Who Really Wrote the Works of the British Renaissance? thread 2

Copied and pasted from AbigailAdams26 introductory post https://www.librarything.com/topic/337240#7670449

Hi All: I want to draw your attention to a new interview we've just posted on the LibraryThing blog, with scholar Anna Faktorovich, in which we discuss her new project, the British Renaissance Re-Attribution and Modernization Series.

ER participants will recognize this name, as the author's main study, as well as some of the newly available texts from the period that she has presented in the series, were offered as giveaways during this past month's ER batch.

Dr. Faktorovich discusses her new computational-linguistic model, and how she has used it in the study of 284 works of the British Renaissance, coming to the conclusion that these works were written by the same six ghostwriters.

Come read her argument, and tell us what you think!

https://blog.librarything.com/2021/12/an-interview-with-scholar-anna-faktorovich....

Summary of the previous thread

LT posts an interview with an author who claims she's proven that most of the works of the British Renaissance were written by six ghostwriters. The author reacts poorly when everyone says she's jumped to conclusions based on suspicious data generated by faulty methodology.

And since I'm sure this summary will draw comments:

>1 amanda4242:
Thanks! Loading times were getting rather cumbersome. Out with the old, in with the new!

I saw you'd opened a continuation thread, but I pulled a sneaky one and posted one final tedious Petroglyph comment anyway. I'm done explaining basic data management to someone who won't hear it, but it felt fitting to post the last-ditch effort in the old thread.

>2 amanda4242:
I would add something to that description along the lines of "a secretive Workshop of six ghostwriters", or even "most of the works of the British Renaissance, including the plays of Shakespeare, the letters of Queen Elizabeth I and the King James Version". But then again, I'm pedantic that way.

---

Edit: Poor Crypto-Willobie. Or perhaps now they get to try and get a whole new set of significant digits.

>3 Petroglyph: I wonder if there's a collective noun for ghostwriters that I could employ. A scribble? A cabal? Oh! How about a haunt of ghostwriters?

>4 amanda4242:

How about a séance of ghostwriters?

I believe the prescribed term is "workshop."

But I'd vote for "cramp."

A Communion of ghostwriters?

I do rather regret this continuation. There was something uniquely appalling about the original thread that was somehow enhanced by the increasingly long time it took to load.

I was hoping it would die the death it deserved.

I guess you have conspired within your Cabal to start this new thread with a volley of insults (that are failing to appear humorous) tossed in my direction. Yes, the term I apply to the Renaissance's six underlying authors is the Ghostwriting Workshop because this is the term that was used during the Renaissance to describe collaborative writing clubs (as I explain in BRRAM's Volumes 1-2).

And this statement is incorrect: "The author reacts poorly when everyone says she's jumped to conclusions based on suspicious data generated by faulty methodology." A more accurate summary would be: "The author presents accurate and precise data and an entirely new attribution method, both of which prove all previous attribution studies of the Renaissance to be fraudulent or faulty in their data and methods. The authors of some of these previous faulty attribution studies (writing under pseudonyms, and thus without acknowledging their bias) react poorly towards being discovered to have erred, and so they attack the author with insults and nonsensical and cyclical anti-arguments intended to disqualify the findings without actually finding any real fault with them."

I asked Faktorovich to show me "Verstegan's authorial signature? And Percy's? And Jonson's? All six of them, in fact. " As well as the authorial signatures as they emerged from your Lunch Break Experiment (tm).

Faktorovich wrote: "I narrate what the six Renaissance signatures signify across the BRRAM series {...} lengthy chapters of explanation. Can you clarify what you are trying to ask here, if you are not asking me where to find my data?"

The data I am asking for is not something you have provided in one convenient spot; rather it is spread out across several volumes.

You test texts for various measures: full stops, question marks, Auxiliary verbs, % passive voice, syllables per word, lexical density, etc, and the unquestionably apophenic nonsense that is your "patterns" of the 6 most frequent letters and words.

You claim to be able to recognize an author's signature as a result of 27 of your tests (with more added apophenia in the "most frequent three-word phrases".

So. There has to be a certain range of full stops that you think is typical for Percy and, and one that you think is typical for Ben Jonson's. There has to be a unique combination of outcomes for these 27 tests that allows you to say "Sylvester wrote this".

I want to see the "authorial signature" for Percy and Jonson and Verstegan and the others, all in one place.

I want to see a table with a header row and with six data rows, one for each of your six ghostwriters. The columns are the 27 tests. In each data row, I want you to put the range of scores that are "typical" for each author for that test. So in the Percy row, for, say Auxiliary Verbs, I want you to put the range: put something like "3.2 - 7.5" in that cell -- the typical range in which Percy's use of auxiliary verbs falls. The typical Percy number for % Passive voice, the signature Percy range for syllables per word.

I want you to do that for all of the members of your secret workshop. And, if you can, also for Austen and the Brontës and Marie Corelli. But mainly I want the workshop signatures.

There are scattered remarks in your book about how the a-pattern is favoured by author X, and some other pattern is favoured by author Y, and so on. But I want to see all of that in one table.

You should be able to test a renaissance text that is not in your corpus, and, comparing the results for that text with the authorial signatures in that table, match that text to its putative author(s) with a certain probability.

I would think that such a table would be the crowning achievement of your method.

>10 faktorovich: The authors of some of these previous faulty attribution studies (writing under pseudonyms, and thus without acknowledging their bias) react poorly towards being discovered to have erred,

I don't really know what to say. It's always a conspiracy, instead of a bunch of random people disagreeing with you because you didn't convince them.

>4 amanda4242: The Stratemeyer Syndicate employed a lot of ghostwriters so you could use

a syndicate of ghostwriters

James

Replying to this Faktorovich post #1151

Great Cthulhu in the deep below, I was right. She uses the red font colour way of counting because counting her data properly (i.e. the yellow highlighter way) would return absolute garbage as result.

"Many of my tests return unusably small ranges and test for features that are used too sparsely to meaningfully separate authors. So I need to change the way I'm counting so I can keep using the tests."

the exclamations test is frequently near-zero or zero for a significant portion of texts. But there are some that have high exclamation values. If you are measuring 18% of the highest value in the range, you might judge not only all 0's, but also most of the other exclamation values to be similar to each other or a 1 (if a text falls in this group), whereas only a few outliers near your highest-value would be dissimilar or 0.

and:

Then, you have an epiphany that I have "some very narrow ranges" in some of the 27 tests, such as Adjectives, where it is a range of 2 points between the minimum and maximum. If I measured Adjectives by calculating 18% from any value at such a close curve, it would increase the number of matches for each test, and it would dramatically increase the percentage of tests on which texts would match. Just as the exclamations are clustered around 0, Adjectives also have a clustering point around which most of the texts are within 18% of each other.

If a test returns results that are concentrated within an unusably narrow band or that are too erratic to be systematically meaningful, then a proper scholar would conclude that testing for that particular feature is, in fact, not meaningful, and would remove it from consideration in favour of a better test.

You, however, continue to use the meaningless test and you devise a ridiculous way of counting, specifically so you can keep the meaningless test around and base your revolutionary new method on it and several others like it.

I'll say that again: Your results are based (in part) on tests that return results too erratic or too narrow to be useful in separating out authors. And you've twisted the way you count these results in order to keep including the meaningless tests instead of discarding them for the useless results they are.

Garbage in, garbage out.

So by following your method, I would be engaging a trick that would make it seem as if a lot more texts are similar to each other. But the problem with this approach is that while the total "matches" would go up, the number of false-positives would also increase. My goal is not to see the highest possible percentile match, but rather the most accurate possible results, and this is achieved by just counting a percentage of texts and not the percentage between their numeric outputs.

It's not a "trick". It's revealing those tests as unfit for the purpose you want them to fulfill.

Your method can only work in a corpus with a single test-type, such as word-frequency, as in this method you might be able to choose a specific percentage that would make sense. Because I am combining the outputs of 27 different tests, I cannot use any standard percentage cut-off for all tests."

What? This is nonsense. You use the ~9% above + ~9% below as based on corpus size rule consistently (the red colour rule); you could just as consistently apply the yellow highlighting rule. And if you chose sensible tests, whose results would not return ranges with insignificant differences between min and max, you wouldn't have to resort to creative data counting.

>1151 faktorovich: (from the previous thread):
Consider this experiment, you are trying to figure out which out of a group of children are most likely to be the sons of which fathers. You don't have access to DNA testing, so you are sorting them by obvious traits like hair color and eye color etc.

I.e. you're ignoring everything we know about the inheritance of traits like hair color and eye color in humans.

You create a table for each of these tests. Let's look at the test for eye color, you have chosen to register specific colors instead of just sorting the kids into blue, brown etc. categories. Perhaps you happen to have a pool of children that mostly have variants of the brown eye color. If you use the 18% measure as a cut-off while placing children on the rainbow spectrum, you might have up to 95% kids falling closer to the brown color than to the other colors. You have thus made this test statistically useless because it is not registering the variations between brown that your careful photographing of eye color and separation of it on the full spectrum was designed to provide. If you separate the children on this spectrum and then choose the 18% of children that are closest in eye color to a tested father; then, you will receive the specific answer to the question which of the children are similar in eye-color to each of the potential fathers. Do you have any questions about this?

So you're presuming you have a complete list of fathers upfront. That seems problematic in reality, and in analogy.

Secondly, you seem to be putting getting an answer above getting a correct answer. If you match the kid with blue-grey eyes with the father with blue-green eyes, then any of the brown eyed kids should match with any of the brown eyed fathers. Even without looking things up, I'd expect fine details of eye color to be based off of environmental factors, making any fine-grained comparisons meaningless. Note that in your literary comparisons, that would sweep up outlying works into a small set of ghostwriters, instead of recognizing outlying works as possibly being the only work by an author.

For anyone who is confused by memes, this is a resource:

https://knowyourmeme.com/

James

Here is another report on the silliness and nastiness that is Faktorovich's mind-bogglingly wrong-headed book Re-attribution.

While I think that the paragraph I want to highlight in this post is a particularly clear example of the hopelessly convoluted mess that is Faktorovich's thinking, I do realise that this may come across as yet more overexplained tedium that's not worth your time. Which is fair enough. I'll cover the annoying bits with spoiler tags.

In my defense, I can only say that I wanted to use this paragraph in my review, but then the fact-checking got out of hand, and now I have a whole bunch of notes about just one single paragraph of distilled stupidity, nastiness and just sheer wrongness, and I didn't want them to go to waste. So here they are.

---

1. Necessary background

If you want to understand the absolute dog's breakfast that Faktorovich makes of things, I need to explain a few things first, so here's a list. I've tried to put the critical information first, and the expanded background behind spoiler tags.

1. Some vocabulary: hapax legomena: words that occur only once in a corpus
   
   
2. the proportion of hapax legomena in a newspaper article can be as high as 75%; the proportion of hapax legomena in one corpus from the 1970s was 40%
   Kennedy (1998, p. 100)says the following:

   "In a short newspaper item of about 200 words, as many as 150 words may be hapax legomena. In large representative corpora, the proportion of hapaxes is normally much smaller. For example, in the American Heritage Intermediate Corpus (Carroll, Davies & Richman, 1971), which is an unlemmatized corpus of 5,088,721 words, 35,079 of the 88,741 different types are hapaxes (39.5%). Nevertheless, the fact that almost 40% of the words in a corpus of over five million words occur only once shows that a corpus of even that size is not a sound basis for lexicographical studies of low frequency words."

   
   
   
3. In one particular 10-million word corpus, the top 2000 most frequent words accounted for 80% of all the words in the corpus
   O'Keeffe et al. (2007, p. 32) talk about "a mixed corpus of ten million words of English (made up of five million words of spoken data, from CANCODE, and five million words of written data taken from the Cambridge International Corpus, CIC)". When talking about this graph (image link), O'Keeffe et al. say this:

   If we examine the frequency of words in a large corpus of English, a picture emerges where the first 2,000 or so word-forms do most of the work, accounting for more than 80% of all of the words in spoken and written texts. As we progress down the frequency list, each successive band of 2,000 words covers a progressively smaller proportion of all the words in the texts in the corpus, with many words occurring only a small number of times or, indeed, only once. (2007, p. 32)

   
   
   
4. Jackson, in his 2003 book, tries to assess how the play Pericles fits within the Shakespearian body of work. At some point he finds it useful to looks at words occurring with a frequency between 2 and 10.
   Jacson sees a separation in vocabulary between Pericles Acts 1-2 and Pericles Acts 3-5. I do not have access to a digital version of this book, so I'm forced to go by what Faktorovich says about it. Why Jackson looks at vocab that has a frequency of between 2 and 10 -- that is to say, excluding the hapax legomena (they occur only once and so are too unreliable to base much on), and excluding those words that have very, very high frequencies. My guess is that Jackson aimed to look at mainly content words, and excluded the really rare ones and the high-frequency function words. But like I said, I don't have access to the Jackson book, so don't quote me on that one. In general, though, taking multiple looks at your data through multiple lenses, and isolating certain variables is a good idea.

---

2. The headache

Alright. with these preliminaries out of the way, I present to you today's headache-inducing Faktorovich quote (here is a screenshot of this quote in a larger context):

Thus, by narrowing the frequency to between 2 and 10 occurrences, Jackson excludes both the most frequent words (which are likely to occur more than 10 times per play: 80%), and the hapax (single-occurrence: 40-75%) words; thus, it is possible that all words (or 80+75 = 155%) in a given sample can be excluded by choosing these parameters. These statistical absurdities can be avoided by comparing only the most frequently occurring letters, words and phrases, as these are representative of the common elements in authorial style. (Re-attribution, p, 475

The thrust of Faktorovich's drivel is that she takes issue with Jackson's (2003) analysis of the play Pericles. Her disagreement takes the form of a particularly juicy befuddlement as she takes a few unconnected corpus statistics she's taken from introductory textbooks and blends them together in an unholy, anti-mathematical mess that culminates in the nasty, paranoid allegation that Jackson has chosen his frequency range of between 2 and 10 words just so he can exclude all the words in a text and run his analyses on hand-picked words. Aside from conflating "frequency higher than 10" with "covering 80% of the words in a corpus", and that thing where she assumes that that 80% + 75% = 155% happens in other people's minds, too, there's just so much unthinking blather here that it requires a lot of context just to disentangle.

There's much more nonsense in the screenshot of that passage (e.g. assuming that what's "confusion-generating" for her is also confusing for others; 2% of 10,000,000 is 2000??; conflating her renaissance corpus with a properly balanced representative corpus). But that's for the diehard drama aficionados.

---

3. Why bring up so much background for a paragaph of nonsense?

Why all this? Well, the fact-checking kinda got away from me.

But there's a larger point to be made here, too.

The paragraph I quoted is part of Faktorovich's uncomprehending discussion of Jackson's arguments, which are a large part of an entire chapter on Pericles (pp. 462-180). Look at the last sentence of that paragraph I quoted above:

These statistical absurdities can be avoided by comparing only the most frequently occurring letters, words and phrases, as these are representative of the common elements in authorial style

In other words: "other researchers nonsense; my method true."

This is one of the most constant, repetitive drumbeats to Faktorovich's train of thought. The rhetorical point of that paragraph, like the rhetorical point of so much else of her drivel, is to present her own method as superior by comparison. All this ignorant and accusatory meanness serves the goal of making her own method look not just better, but the only correct one. (Queen Gertrude might say something about the lady protesting too much.)

And there's just a constant barrage of paranoid shit-flinging. In the two and a half pages leading up to that paragraph I highlighted, Faktorovich is very diligent in accusing Jackson of, what else, BiAs and dAtA mAnIpUlATiON. No fewer than seven explicit accusations in the span of two and a half pages:

1. "It is likely that Jackson selected these links with an undisclosed bias if{, in Faktorovich analysis, Ben} Jonson’s light touch came across as the dominant signature" (p. 473)
   
   
2. "Jackson does not provide the 377 words that led to this erroneous conclusion; this non-disclosure prevents an audit of the cause of this strangely high match between these texts" (p. 473)
   
   
3. "all words in these texts and not only handpicked ones have to be measured to determine if these echoes indicate a greater contribution from one of these ghostwriters." (p. 474)
   
   
4. "Jackson’s findings can be double-checked more fully when he discloses a few revealing elements of his chronological analysis" (p. 474)
   
   
5. "Perhaps in an attempt to complicate audits, Jackson did not include these dates, so these had to be derived by checking first-performance data" (p. 474)
   
   
6. "Jackson then employs a summary table (again without the raw data) to demonstrate that an unknown combination of these early plays is significantly more similar to Pericles’ Acts 3-5 than to 1-2. Since my tests indicate that Jonson had a slightly greater contribution to Acts 3-5 of Pericles than to 1-2, this earliest category should be a closer match to Acts 1-2." (p. 474)
   
   
7. "Jackson’s data indicates a complete reversal in linguistic attribution that defies the rules of statistical-logic" (p. 474)
   
   
8. And the "statistical absurdities" smear discussed above (p. 475).

It continues like that in the subsequent pages, too. It never ends.

Faktorovich, uncomprehendingly lashing out at actual scholars, throws bad-faith accusations and paranoid strawmen at this study, casting Jackson as one of the mean, nasty academicses performing statistical mumbojumbo to manipulate their results and hide the truth. The Truth, which has only been uncovered by her own idiosyncratic research that bears all the hallmarks of a conspiracy theory.

It's bluster. Mean-spirited bluster. Strawmanning with dishonest intent.

---

4. I want this to conclude!

Ugh. The whole book is like that. It's a never-ending assault on quality work and good taste and there's no end to it. Like, look at the length of these paragraphs!



By Grabthar's Hammer, you guys. The stupidity is just so concentrated and so relentless! This entire book is a dense, mis-argued, mean-spirited muddle of anger and confusion. A dogged insistence on an extremely literalist and context-less reading of some half-understood corpus linguistics, and she'll use the resulting mess to discredit someone else's words and make her own look better by comparison.

Writing about this book is a deeply unpleasant experience.

---

References

Faktorovich, Anna. 2021. The Re-Attribution of the British Renaissance Corpus. 1st ed. British Renaissance Re-Attribution and Modernization Series 1–2. Quanah: Anaphora Literary Press.

Jackson, MacDonald P. 2003. Defining Shakespeare: Pericles as Test Case. Oxford: Oxford University Press.

Kennedy, Graeme. 2014. An Introduction to Corpus Linguistics. New York: Routledge.

O’Keeffe, Anne & Michael McCarthy, Ronald Carter. 2007. From Corpus to Classroom: Language Use and Language Teaching. Cambridge: Cambridge University Press.

>17 Petroglyph: I have electronic access to Jackson's book – did Faktorovich provide any page numbers for the claims she makes about Jackson's methodology, at the bottom of page 474? I think that when she quotes Jackson as distinguishing between words "occurring 2-6 times" and "occurring 2-10 times", she refers to Jackson's discussion of Eliot Slater's research, and Jackson's own method which builds on Slater:

"... Inevitably, the section that follows contains a forbidding mass of figures and cites statistical tests of significance. But the basis for these is the simple rule that works written by the same author at about the same time are apt to have more of their low-frequency words in common than works whose dates of composition are separated by many years.

Eliot Slater compiled card-indexes recording all instances of words that occur in at least two Shakespeare plays but not more than ten times altogether. In a series of articles and a University of London doctoral dissertation, published posthumously as a book, he demonstrated that the vocabulary of any one Shakespeare play tends to be linked most closely with the vocabulary of other plays that he wrote within the same period. Likeness or unlikeness of subject matter and genre may increase or decrease the degree of linkage, but the chronological factor is dominant, or, at the very least, considerable. Slater’s method was to compare, by the use of chi-square tests, the actual number of word links between plays with the number of word links to be expected were the distribution purely random. He calculated ‘expected’ figures on the basis of the relative size of each play’s total vocabulary, but expectations may more appropriately be derived from the proportion of the total number of link words (those occurring 2–10 times in the dramatic canon) that are present in each play. Expected figures worked out in this way differ only slightly from those produced by Slater’s more approximate procedure: overall patterns remain unaffected, but a few marginally significant associations between plays disappear when the more precise measure is used. The analyses that follow, though accepting Slater’s figures for actual links, employ my own more precisely computed expected figures." (Jackson 2003:40-41, referring to Slater, Eliot. (1988). The Problem of ‘The Reign of King Edward III’: A Statistical Approach Cambridge: CUP)

And a bit further down:
"If we confine analysis to the rarer vocabulary—words occurring 2–6 times in the canon—the predominance in Pericles, 3–5, of links with late plays persists. The most significant associations are now with Pericles, 1–2 (8 links, 3.58 expected, chi-square 5.46), The Tempest (16: 9.49, chi-square 4.47), and Timon of Athens (16: 9.7, chi-square 4.09). For Pericles, 1–2, on the other hand, the pattern changes. The significant linkage with Antony and Cleopatra and Coriolanus disappears, The Merchant of Venice replaces its near-contemporary 1 Henry IV as a play with marginally significant associations with Pericles, 1–2 (12:6.6, chi-square 4.42), and the truly significant relationships are with Titus Andronicus (15:7.97, chi-square 6.2) and 1 Henry VI (19: 9.64, chi-square 9.09)." (Jackson 2003:43)

Edited to add: Jackson doesn't by any means uncritically accept Slater's method – see also
Jackson, MacDonald P. (2015). Vocabulary links between Shakespeare's plays as a guide to chronology: a reworking of Eliot Slater's tables. Shakespeare, 11(4), 446-458. DOI: 10.1080/17450918.2014.985604.

-Linnéa

>11 Petroglyph: As I explained in a previous post, the "Lyly"-assigned and Percy-ghostwritten "Sapho and Phao" matches 17 Percy-group plays, and no texts from any of the other linguistic-groups among the other 267 texts. This is equivalent to rolling "1" on the dice every one of the 17 times you bet on "1" at the casino, and in the same game rolling "3" every one of the other 267 times you bet on "3". If you had those 100% odds of winning at a casino, would you call them "random"?

The "most frequent three-word phrases" test is not one of the quantitative 27-tests, but rather is the 28th test that is not part of the statistical attribution conclusion; it is used to check if it matches the overall conclusion, and to research further what the attribution signifies.

I began with a corpus of around 100 texts, and gradually expanded it to 284 texts. With each text I have added, the "signature" for the author of the added text mutated slightly to accommodate its dimensions. For example, the artistic "style" of Van Gogh is defined by analyzing his paintings across his career, and reaching a conclusion about the stages and variants he was capable of. If we had only 2 surviving Van Gogh paintings, the art world's understanding of his style would have been very different than it is for the full range that we actually have. Even if changes are small from each added text, it would statistically inaccurate to give you the specific ranges for the current texts in the corpus, as they might change by a bit from an expansion of the corpus. Additionally, as I explained, there is a lot of collaboration across this corpus. Jonson and Percy, Verstegan and Harvey, Sylvester and Harvey, and and some of these with Byrd frequently co-wrote texts. The co-written texts are grouped with the dominant ghostwriter, but co-written texts have a percentage of their signature that fits with the co-writer's linguistic range, and not with the dominant hand's. Thus, probably only completely or mostly single-authored texts should be considered when establishing a linguistic range for a specific author. Even if such a range is derived, if you choose a new text from the Renaissance and compare it to this pure single-author range, if the text in question is co-authored, it would fall between two pure ranges. This is why it would be a mistake to start with the range and check works against it. You really have to go through the full 27-tests process to reach a precise attribution conclusion. If you want to know the current ranges, or to experiment with them, the data is available on my GitHub, so you can go ahead and check for yourself what the ranges are. The data is already in one place. No computational-linguistic study I have looked at has claimed to know or to list the precise signature-range for any given byline.

Why are you requesting me to create a table for you with specific dimensions etc. that you already have in mind. I posted the data on GitHub so that anybody can manipulate or experiment with the data. So just manipulate it and come up with the signature ranges that you want to see. Yes, I comment on the a-pattern and several other more complex patterns across the series. The data tables on GitHub include all of this data already. You can re-order the table for the 27-tests to see which character/ word patterns appear in which of the texts and groups. My summary of these patterns in a single table would be simplistic and would lack all of the fine points in the complete set of data. The most-common pattern for a given signature proves the existence of the signature style, but the second and third most-common patterns also communicate points in the larger narrative about the components of an author's style.

Nothing I have said so far in the overwhelming amount of evidence I have provided has convinced you, but you are saying that a summary table with a bunch of signature-defining statements would convince you? Why do you want to see such a table?

>13 Keeline: Again, the ghostwriters in the Renaissance called themselves a Workshop (and "ghost-writers"), so that's why I called them the Ghostwriting Workshop.

>14 Petroglyph: The data might appear to be in "an unusably narrow band" only if the 18% in its concentrated data-peak is compared with the data in the outlying ends of the range. But when the numbers are compared as individual texts, there is an identifiable pattern that spots the different authorial signatures between the underlying authors.

The tests I choose for a given corpus are the most suitable ones for it that register the stylistic preferences of the underlying author. The consistent accuracy of my results prove that I have chosen the correct tests for the Renaissance.

>15 prosfilaes: If it helps, imagine that there has been a mix-up at the fertility clinic and the samples from a known list of fathers have been mixed up, so you know who the fathers are, but not whom among the children they fathered. Alternatively, you can imagine there are only 2,000 sperm donors (equivalent to the no more than 2,000 or so British authorial bylines from this century), and you have to check the 284 children against each of these donors characteristics tables.

If you are concerned eye-color alone is insufficient to achieve a match, this is exactly what I believe as well. This is why I combine 27 different tests, whereas Stylo/ the accepted method in computational-linguistics is to only measure word-frequency, or only 1 test (equivalent to only measuring eye-color). And when you consider that eye-color might not be exactly matched between a father and a child, you are considering the impact of the mother's eye-color-characteristics; and this is indeed another element my approach helps to reveal, as it is sensitive to co-authorship and spots texts with two parents as opposed to clones with a single parent, etc. Environmental factors are things like vocabulary-range for an author, and these define a linguistic style (as we are discussing linguistics and not eye-color). And as long as I considered the full range of 2,000 potential donors (as I did), there is no exclusion of "outlying works" in favor of any popular byline. If I had not reviewed a broad corpus of potential authors, I would not have even considered Percy, Harvey, Verstegan, Byrd or Sylvester as none of these authors have ever been proposed before as alternative "Shakespeares" or as the "real" authors behind any other anonymous or pseudonymous Renaissance texts. Only by reviewing thousands of potential bylines did I eventually come across these specific donors that fit the unique characteristics of the children (and the timing of the donations).

>22 faktorovich: "This is why I combine 27 different tests, whereas Stylo/ the accepted method in computational-linguistics is to only measure word-frequency, or only 1 test"

Are you simply making things up as you go? Stylo is (as has been explained to you) not a method, it is a tool. 27 tests is (as has been explained to you) a very small number indeed, and back in post 196 in the previous thread you got a list of (at least) several hundred potential tests possible to carry out using Stylo.

-Linnéa

>17 Petroglyph: The quote from my "Re-attribution", p. 475 and the surrounding information in that chapter of BRRAM explained exactly what your preceding introductory comments pointed out, but you did not give me credit for explaining it. I did not make any mistakes of believing 2% of 10 million is 2,000; if I had you would have cited a specific quote where I said anything of this sort. My statements in this section are very accurate and logical; only your nonsensical digressive discussion about it is illogical and mistaken.

Your paraphrasing of my "very diligent... accusing Jackson of bias and data manipulation" appears to be accurate. Yes, I accused him of both of these things, and the quotes explain my objections with evidence that this is indeed the case. You do not provide any evidence to contradict my findings that Jackson and other computational-linguists I review across BRRAM are not biased/ in error. Just because you yell back at me a quote of my rational conclusions and argue that the yelling makes them sound insane, it does not mean you have convinced anybody of your emotion-based explosive reaction, but rather you have betrayed you have no logical argument to make to support your generally antagonistic position.

By the way, you broke copyright law when you posted images of 5 pages from Volumes 1-2. You are allowed to transcribe quotes (up to a paragraph on a topic) from books under "fair use", but you are not allowed to publish 5 pages online for the public to access. And the images of my design/ formatting are not covered in "fair use", which is only designed for researchers who need to communicate an idea. For example, if I had published a book of my paintings, and you had scanned and published these paintings, you would be in clear violation of copyrights law. One clue you are failing to employ a text in "fair use" is when you do not actually read any of the pages you are illegally publishing. This is the first time you have hinted you read any part of BRRAM other than the summary of the basic method, and you are only "reading" a paragraph, while illegally publishing 5 pages. And you are restating content from that paragraph as if it is your own ideas, while accusing me of errors that are not actually there or that you cannot quote. The only errors you are citing is that I find fault with Jackson's study, and you think that I do not have a right to criticize errors in somebody else's linguistic research.

>18 anglemark: Now you are asking for second-hand quotes from my book, instead of just asking me for a review copy to check my citations for yourself?

>23 anglemark: Jackson and almost all of the computational-linguistic studies that have been discussed in this thread only use the word-frequency test when utilizing Stylo. I included other software that can be used to test word-frequency in my statement after the /

And no, the list of "several hundred potential tests" did not actually include many applicable tests that are not already included in my 27-tests, and there were certainly not over 100 different tests in the tossed-together list that was provided, and the list included various errors that I explained in response to that post.

>20 faktorovich:

Again, the ghostwriters in the Renaissance called themselves a Workshop (and "ghost-writers"), so that's why I called them the Ghostwriting Workshop.

Kindly show me an image of a vintage document from the 1500s (with full citation) that uses any variant of "ghostwriter" or "ghost-writer" or "ghost writer" or any variation of this spelling. The information available to me is that the phrase is not more than 100 years old. The concept is older.

"Workshop" is said to go back to 1556 though not in a literary sense.

When I call the Stratemeyer Syndicate (1905-1984) a "book packager," I know full well that the term was not in use when it was founded. However, modern groups who use similar methods are often called "book packagers" and to use the phrase today helps to communicate the methods used when a reader is familiar with that phrase. Of course, when I see "book packager" associated with the Stratemeyer Syndicate, it is a clear indication that the writer has seen my work directly or saw it indirectly since the small field of researchers had not applied the term before I did back in the 1990s.

James

>24 faktorovich:

You have an incorrect perception of what is permitted under "Fair Use" provisions under the U.S. Copyright law. Here is a summary which may be enlightening.

https://www.nolo.com/legal-encyclopedia/fair-use-rule-copyright-material-30100.h...

Five pages from a volume with hundreds that are reproduced in a small format is an illustration of the page layout. From a practical sense, court cases have found that usage of up to 10% of a work can be permitted in certain circumstances. There are lots of details and case details so we won't get in to those weeds.

But if you are very concerned with this, get the opinion of a qualified intellectual property attorney to see if you have a claim that is worth pursuing.

James

>18 anglemark:

Thanks, Linnéa, for going to all that effort! Ett stort tack!

The footnoted citations in that chapter are to p. 42, 40-47, 48, 55. So the bits you quoted indeed concern the relevant parts of that book.

I knew that looking at words with frequencies 2-6 and 2-10 would make sense in context! Faktorovich just calls it "confusion-generating" because the same words would be in both tests.

>19 faktorovich:

This is equivalent to rolling "1" on the dice every one of the 17 times you bet on "1" at the casino,

No. It's not like that at all.

The "most frequent three-word phrases" test is not one of the quantitative 27-tests, but rather is the 28th test

I know that. I said: "a result of 27 of your tests (with more added apophenia in the "most frequent three-word phrases")" But I can see how a careless reader who is eager to discover faults can miss that.

I began with a corpus of around 100 texts, and gradually expanded {...}

Not what I asked for. Excuses not to do it. I'm not asking for the definitive, ultimate, perfect result. I'm asking for the summary of all your testing in the form of a table summarizing the styles as you have used them.

If you want to know the current ranges, or to experiment with them, the data is available on my GitHub, so you can go ahead and check for yourself what the ranges are.

More excuses. Shifting the work onto the reviewer. Even here you're terrible at play-acting as a scholar.

You can re-order the table for the 27-tests to see which character/ word patterns appear in which of the texts and groups. and So just manipulate it and come up with the signature ranges that you want to see

Not my job, but yours. Just an excuse not to do it yourself. And I don't want to make any mistakes. When you do it, I know the job will be up to your standards and you'll have no reason to complain about others mucking things up.

My summary of these patterns in a single table would be simplistic and would lack all of the fine points in the complete set of data

Yes, that's why it's a summary. Another excuse.

you are saying that a summary table with a bunch of signature-defining statements would convince you?

No, that's you reading into my question.

Even if such a range is derived, if you choose a new text from the Renaissance and compare it to this pure single-author range, if the text in question is co-authored, it would fall between two pure ranges.

Yes, I know that. But also: if any of the six wrote (most of) that 'untested' text, it would fit. Duh. Is this another excuse?

Why are you requesting me to create a table for you with specific dimensions etc. that you already have in mind. and Why do you want to see such a table?

Why are you so suspicious? This summary table would be the crowning glory of your work! "Here are the six styles that are dominant in this corpus! Anyone can test a text and see who it was by!"

If you're worried about the collaborative thing, it would be easy to add one row for each possible collaboration.

This way, anyone could test texts from the Renaissance, and instead of doing the whole "replacing with ones and zeroes" thing just go straight to the summary table and check against an author.

It would be a quick verification of the whole thing. I can't, honestly, see a reason not to provide such a table.

>19 faktorovich:

I wanted to highlight this particular bit on its own:

I began with a corpus of around 100 texts, and gradually expanded it to 284 texts. With each text I have added, the "signature" for the author of the added text mutated slightly to accommodate its dimensions.

Put differently: Faktorovich adds a text to her corpus that is a little like, say, Percy or Sylvester or Verstegan, but that does not carry that "by-line". She then either expands the style of the author(s) closest to this text to encompass it. This "gradual" expansion happened ~180 times since the initial ~100-text corpus.

I just thought that methodological statement was worth commenting on.

>21 faktorovich:

Yeah, you've already told us you resorted to creative counting when faced with tests that yield results that are too erratic or cluster in too narrow a band to be useful.

It's one of the many reasons why everyone who is not you is justified in rejecting your dreck.

In your lunch break experiment (tm) (partial screenshot here), I count 12 tests where the difference is less than 6; there's 4 more where the difference is between 8-12.

More than half of your tests you have to force into your "analysis" by counting the red-colour way because the things they supposedly test for are irrelevant.

Thank you, lorax (if you're still here), for introducing me to the term fractally wrong. It's *chef's kiss* perfect!

Question for Faktorovich:

When you are detecting your apophenic "patterns" in the 6 most frequent words, do you normalize those data at all? Or do you work with the raw absolute frequencies?

A simple yes or no will suffice. Thanks!

Question for Faktorovich.

Several times, when I mention that, in terms of word frequencies, you only look at the six most frequent words, you have disagreed and said something along these lines:

No, I look at every single word within every word in every one of the texts I evaluate, and from all of these derive the unique 6 most-common words in each text. (Page 1, msg #1103)

Quick clarification question. What exactly do you mean by that?

Do you mean something like this: In order for the software to determine that the, of, in, is, ... are the six most frequent words, it had to count the frequencies of all the words in the text (or the corpus). Therefore, the analysis of the top six words implies the analysis of all others.

Is that an accurate paraphrase of your position? Am I understanding this correctly?

>22 faktorovich: If you are concerned eye-color alone is insufficient to achieve a match, this is exactly what I believe as well.

That is not what I was concerned about.

If you separate the children on this spectrum and then choose the 18% of children that are closest in eye color to a tested father; then, you will receive the specific answer to the question which of the children are similar in eye-color to each of the potential fathers.

If you have one blue eyed father, and 11 brown eyed fathers, and two green eyed children, and 10 brown eyed children, your system will happily tell you the two green eyed children are similar in eye color to the blue eyed father, and pick out two children to be matched with each brown eyed father, instead of saying there are zero close matches for the blue eyed father and 10 for the brown eyed fathers. You're hiding information and forcing it to give you a result, no matter what.

>27 Keeline: You missed my previous post where I explain the earliest uses of this term:

---My definition is ground in lines such as these in “The Grounds of Divinity, plainly discovering the Mysteries of Christian Religion” (1633: https://www.google.com/books/edition/The_Grounds_of_Divinitie_plainely_discov/DV...: “The holy scriptures are all those Books of the Old and New Testament, by the direction and inspiration of the Holy Ghost, written, or approved by the Prophets and Apostles.” Or in the commentary on “Psalm XCIII” in “Holy Bible” (1635: https://www.google.com/books/edition/Holy_Bible_Faithfvlly_Translated_Into_En/WI...: “For in human books the writer and author is all one; but in divine, the Holy Ghost is the proper author, and a man is the writer.” The non-theological application of a similar meaning is found in a byline, “Written by Thomas Nash his Ghost, with Pap with a Hatchet” (1642 edition: https://quod.lib.umich.edu/e/eebo/A94724.0001.001?view=toc)--- (I go on to explain it further.)

Let me know if you need further clarification.

>28 Keeline: If you took 10% of a book and published it (public posting is a publication) you would be committing an extreme copyright infringement; the 10% rule is likely to apply to a poem or another short work where 10% is not an extremely large quantity of information. And all of the information you are re-posting has to be discussed in the new piece of academic writing you are creating; it cannot be simply re-posted to show that a writer has a very long paragraph. Here is a list of songs that have been under copyright dispute/ settlement: https://en.wikipedia.org/wiki/List_of_songs_subject_to_plagiarism_disputes - I recall watching a documentary video from a lawyer that explained a recent case was over a few repeating notes that were similar between songs.

I am the "qualified intellectual property attorney" as I have been representing my own publishing company for over a decade. I have won lawsuits before pro se while suing other companies. I am not a lawyer, but I (and everybody else) can represent themselves in the US.

>30 Petroglyph: "I'm asking for... a table summarizing the styles". I stated as much and explained in my previous post why such a summary table would not serve readers any better than looking at the full complex set of outputs in the complete data tables, which you can use to see the ranges that different ghostwriters fall between. You are again not reading any of my replies, and just repeating the questions, as if I did not already answer them.

Creating a simplification of my research with a table that compresses my intricate findings into only the highlights of those findings would not be the "crowning achievement", but rather would give you and others room to point out that the summary does not include the intricacies I would have to delete to create this abridgement.

The degree of collaboration (percentage of text written by each and if it is written by 2 or more ghostwriters) changes between most of the texts, so it would not be a matter of simply adding a row to quantify the range in a collaboration; in contrast, the tables I have already posted on GitHub provide this full range with all of the intricacies to allow researchers to explore these patterns more closely if this is a subject that interests them.

There is no simple attribution method, as even if there is a program to process texts, somebody has to research the implications of the quantitative results. The only way to simplify my method for readers is to create a program that applies the 27-tests automatically to any texts entered into the system (probably starting with the Renaissance or other periods I already have a completed corpus for). This is not something I have considered doing, as my method has been sufficient for me to solve the mysteries I am interested in researching.

>31 Petroglyph: That is absolutely not what my statement means. I am describing the statistical distribution curve of output values that has to change slightly (if the corpus is large) or greatly (if the corpus is relatively small) every time a new text is added to a corpus. I did not add texts that are "like" any of the ghostwriters to "expand the style". I do not even have any idea what you are trying to say here. I added texts that were needed to verify alternative potential ghostwriter-bylines, or when texts were essential to establish the earliest and last texts a ghostwriter created, or otherwise were necessary for BRRAM (such as when they were discussed in previous computational-linguistic studies and I wanted to check these attributions with my own method).

>32 Petroglyph: You have not read any of my replies that explain why all of the nonsensical accusations you are making against my method are false.

>33 Petroglyph: "To help avoid dependence on the choice of measurement units, the data should be normalized or standardized. This involves transforming the data to fall within a smaller or common range such as −1, 1 or 0.0, 1.0." --https://www.sciencedirect.com/topics/computer-science/max-normalization

I normalize my data with my unconventional or new method. I compare each type of data units separately and then turn them into a binary (0/1) data table; this resulting data table is in a single unit 1/0, as is standard in the above quote where "0.0, 1.0" is an option.

>34 Petroglyph: Yes, the top-6 words test checks the frequency of all words in the text to determine which are the most frequent words. The output this gives is a unique pattern of these words that frequently can identify the underlying ghostwriter on its own (in less cooperative corpuses than the Renaissance). I am not only testing the top words' frequency, but the combination of these words that a given ghostwriter uses that identifies their personality traits, when these are considered beyond their purely quantitative value for authorial attribution.

>35 prosfilaes: If I only had a single test (such as eye-color), and all of the children were clones (or only had a father and no mother) and none of the children's eye colors matched any of the tested fathers; yes, this would have generated a glitch. Upon encountering such a glitch, I would have used a different testing method to determine parentage. This is why, instead, I used 27 different tests, considered thousands of potential authors/ "fathers", and otherwise maximized the degree of precision of the results. If my corpus had 104 authors for the 104 bylines in the corpus, the texts written by the same authors would have still been proximate to the other texts by that author, and would have been accurately attributed (as I found out when I tested this type of a multi-author corpus for the 18th century). I am not hiding any information; everything is on GitHub.

>37 faktorovich:

I am the "qualified intellectual property attorney"

No, you are not. Your understanding of "fair use" is way off the mark. There are other factors than the amount of material published that are considered, and there is no bright line as to the percentage of material that may be used. To quote the U.S. Copyright Office, "there is no formula to ensure that a predetermined percentage or amount of a work—or specific number of words, lines, pages, copies—may be used without permission."

One can reproduce some of the work in order to criticize or comment on it - exactly what people are doing here. The fact that your work is published is a factor in favor of considering this fair use, and no one here is using the material for commercial purposes.

I (and everybody else) can represent themselves in the US

With the permission of the Court, which is not always granted.

45!

>36 faktorovich:

Really???

To suggest that reference to the "Holy Ghost" in claims for the divine inspiration of scripture, or "Thomas Nash his Ghost" whom the text shows to be a conventional post-mortem haunt, are either of them comparable to or the basis for what we refer to in contemporary English as "ghost writers" and "ghostwriting" is bonkers. "Insects are not animals"-level bonkers.

>38 faktorovich:

All I see are silly excuses for not providing the authorial signatures.

I stated as much and explained in my previous post why such a summary table would not serve readers any better than looking at the full complex set of outputs in the complete data tables

This is not true.

You are demanding that people wanting to check the "authorial-signatures" for themselves have to pour over your tables in detail, go through multiple steps and tables and put in a lot of time and work. This is an unreasonable expectation. Having the summarizing table available would be of great assistance. It would most definitely serve readers better than putting in unreasonable amounts of work.

Moreover: at the end of that unfamiliar process, there is no guarantee that they would have accomplished the goal successfully. You, however, are the only one familiar enough with your "newly-invented" and "unconventional" method to create this summary table with ease and, more importantly: without errors. You are the person who is in the ideal position to do so.

You have also not posted the actual text files that you have used -- I mean, your actual corpus. All the sonnets and plays and letters and all the rest. So readers who use differently-formatted texts from yours may get different results -- different file formats are just errors waiting to happen. You'll want to avoid that! The best way is to provide the table yourself.

as my method has been sufficient for me

And I am telling you that this laborious, painstaking method is not sufficient for other people. You need to keep in mind that other people are not as familiar with your methods, and may not want to invest the time needed to become fluent in them.

the summary does not include the intricacies I would have to delete to create this abridgement

Like I said: It need not be a perfect table that covers all the intricacies, and no rational person would expect that. All it needs to be is a clear overview of the six different authorial signatures and what values they consist of.

It should be straightforward enough to apply texts to them using your 27-test method, but of course allowance will be made for in-between cases.

If you like, you can add the titles of two or three texts that you think are mainly written by each person, so that the people using your test have a few illustrative examples to guide their own process.

You are again not reading any of my replies, and just repeating the questions, as if I did not already answer them

I am not ignoring your replies. I am stating that I find your answers and your objections insufficient. Your objections, to me, sound like you don't want to create such a table. But that is not a good enough reason. It is your job, your responsibility as an author, to make your results accessible to the readers in an easily-digestible format. Merely providing a link to multiple tables and expecting readers to figure it out for themselves by going through the entire process step by step is not good enough.

It is part of the author's responsibility that, when offering a new methodology, they should provide the tools to make things easier for others. The end result of all your labours should not be for readers to go through all the exact same steps as you have for several years. You, Faktorovich, who stand at the end of a long process, should be able to develop tools for others who want to employ your method. Tools that were not available to you at the start of this process, but that you, the only specialist in this method, are able to devise and provide. A summary table with a clear, unambiguous statement of the six authors' signatures is an ideal such tool.

For comparison: you imported an improperly formatted text in R once, and immediately gave up and claimed you had demonstrated that R introduces errors. Solutions were offered to you, but you had already decided that there was no point in using R. Do you want that to happen to you? You don't want people to give up at the first sign of trouble when confronted with an unfamiliar methodology. Which is an unfair reason to discard the whole enterprise.

Also, part of your schtick is that you use freely available tools and avoid complex statistics such as z-scores. Providing handy overviews fits right in with an orientation towards openness and applicability by laypeople.

---

But alright. I see that I have not been literal and explicit enough.

I want you to create this table because:

1. It would be the easiest, most straightforward way of applying your methodology to previously untested texts from the Renaissance for people who are not you, for laypeople.
2. It is unfair of an author to expect the reader to produce the summarizing tables that encapsulate the author's central claims. This is, properly speaking, part of your job and your responsibility in publishing a scholarly tome. Therefore, you, the author, should perform this work; it is unreasonable to expect it from the reader.
3. You are the only person who is capable of creating this table without errors. (Such is the burden of "newly-invented" and "unconventional" methods).
4. If the summarizing table I am asking for is insufficient to capture various degrees of collaboration, then so be it. Its value in discovering single-author works that were ghostwritten by someone from your secret workshop is enough. It need not be perfect or capture all the intricacies.
5. If you are truly serious about creating a methodology that is easy to use and involves nothing you need an advanced degree for, then you kinda have to provide some tools. Tools that were not available to you at the start but that you are now in a position to provide with the benefit of hindsight.

>42 faktorovich:

Yes, the top-6 words test checks the frequency of all words in the text to determine which are the most frequent words.

Ok, gotcha. Thanks for confirming that for me.

The output this gives is a unique pattern of these words that frequently can identify the underlying ghostwriter on its own

This is clearly apophenic nonsense. But we'll get to that later.

>40 faktorovich:

*shrugs* Well, I tried. Harder than most other people would have.

>41 faktorovich:

No, I meant, do you normalize your word frequencies in any way before you replace your data with ones and zeroes?

>24 faktorovich: By the way, you broke copyright law when you posted images of 5 pages from Volumes 1-2

As others have pointed out, fair use (in the US) covers my use of your pages for the purposes of criticism.

If you feel this deserves legal repercussions, have yourself call my people.

Alternatively, you can register a complaint with LibraryThing staff. The contact information and the evidence you need to provide are listed on this page, under the heading DMCA.

(also, small correction: I posted images of 4 pages in their entirety, a portion of a fifth page, and a tiny bit of a sixth.)

To everyone who is not Faktorovich: Warning! Extremely tedious and nit-picking post coming up. Stay out! Go to >53 Petroglyph: instead!

>24 faktorovich: I did not make any mistakes of believing 2% of 10 million is 2,000; if I had you would have cited a specific quote where I said anything of this sort.

Ask, and you shall receive.

And in an experiment that considered a 10-million-word corpus, the 2,000 most frequent words (2% of all) comprised 80% of all words (8 million). (Faktorovich 2021, p. 474)

Here is a screenshot of that quote embedded in its context. Note: this screenshot is for the purposes of criticism.

There are five figures in this sentence. Four of these ("10 million, 2,000, 80% and 8 million") come from Faktorovich's source (O'Keeffe et al. 2007); and "2%", which is her own addition, an addition I believe to be in error.

I read all the figures in this sentence as referring to the same 10-million-word corpus. And the source that Faktorovich is relying on here for her numbers on that 10 million word corpus indeed relates its four figures to that corpus: "the first 2,000 or so word-forms do most of the work, accounting for more than 80% of all the words in spoken and written texts" (O'Keeffe et al. 2007, p. 32).

Furthermore, I read the sentence in that quote as a parallel construction that takes the form of a chiasmus. I have given what I believe to be the intended equivalents in each half the same formatting:

the 2000 most frequent words (2% of all)

vs.

80% of all words (8 million)

In my reading, the parallelism is there to set up a paradox: the small figures in the first half are revealed, in the second half, to have very large-figure effects.

The two figures in the second half both pertain to the 10-million-word corpus; both, in fact, refer to the same proportion: 8 million is indeed 80% of 10 million. Furthermore, the "8 million" in the second half is between brackets, and serves as an explicitation of the "80%" figure. So the most natural reading for the first half is to set up that parallelism, and so I interpret it along similar lines as the second half: the author is (mistakenly) saying that 2,000 and 2% are the same proportion out of ten million, and is using brackets to explicitate the figure of "2000" with respect to the 10-million word corpus.

Finally, the part that says "2% of all" -- I interpret its intended meaning to be "2% of all the words in the 10-million-word corpus", as a parallelism to that "80% of all" in the second half.

---

If this reading is incorrect, I would appreciate an explanation of where that figure of 2% came from and what its import is in the discussion of the 2000 most frequent word forms.

---

So, a typo? A genuine error?

I think this is a genuine error. I think the "2%" should have been omitted entirely -- that is the only way for this sentence to make sense.

After all, the 10-million-word corpus does not contain 10 million unique words; there'll be many, many tens of thousands of instances of the, of, is, in, to, I, he, she, from, ... so amending the "2%" to "0.02%" makes no sense -- 0.02% is clearly not the same as 80%. And you can't change the "2000" figure -- that's right there in the source.

The top 2000 word-forms (or "types") will, of course, be some proportion of all the word-forms in the frequency list. But that proportion is unlikely to be anywhere near 2%. O'Keeffe et al. don't give an exact number for the number of types in this corpus, but looking at the graph on that same page, it's above ten thousand; their discussion on pp. 48-49 suggests something between 16,000 and 20,000 (that is, the 10 million tokens in the corpus fall into ~16,000-20,000 types). So the percentage that Faktorovich should have added ought to have been "10%", at the very least.

I think Faktorovich added the 2% because she's confused between word-forms and the individual words in the corpus: the type the vs all the individual tokens of "the" in the corpus (the top 2000 types (or word-forms, as O'Keeffe et al. put it) account for 80% of the tokens in the corpus). Faktorovich shouldn't have added that "2%", but she did out of confusion -- just one more kink in the incredible muddle that is the section I highlighted in >17 Petroglyph:. But that is only my reading, and I'm happy to be corrected in this respect.

---

Faktorovich, Anna. 2021. The Re-Attribution of the British Renaissance Corpus. 1st ed. British Renaissance Re-Attribution and Modernization Series 1–2. Quanah: Anaphora Literary Press.

O’Keeffe, Anne & Michael McCarthy, Ronald Carter. 2007. From Corpus to Classroom: Language Use and Language Teaching. Cambridge: Cambridge University Press.

TL;DR of >52 Petroglyph:: Faktorovich is fractally wrong.

>44 lilithcat: You imagine that it would be legal for you to take a couple of chapters out of "Harry Potter" and publish them online? While there are plenty of pirating websites that sell texts without permission from the publisher. There are absolutely no cases where publishing 5 pages with the original design would be technically legal. Pirates only get away with such actions if the publishers do not want to invest in paying for a lawyer's hours to pursue a lawsuit over minor cases, reserving such prosecution for major financially damaging cases. The "U.S. Copyright Office" is not a lawyer, or a judge; so this statement from them is not at all legally binding. Yes, there are indeed "formulas" for what constitutes an infringement - this formula is the amount of words etc. per text that was previously found to be a violation in case law. So research case law for specific book piracy cases, and then cite specific instances where publishing 5 pages without scholarly engagement with all of the main points of this text was excused as not a violation of copyrights. In all cases I have won in court, I was given permission to represent myself, so these count as experience to make further self-representation in court more likely to be granted.

>46 paradoxosalpha: The "Thomas Nash his Ghost" is equivalent grammatically to Modern English's "Thomas Nash's Ghost". This is only one example of this type of wording from the Renaissance. As I explained in my original explanation in the previous thread #1, the Workshop regularly attributed texts to bylines for the first time after the "author" had died, as was the case with "Marlowe" and "Philip Sidney". In these cases, and in many similar circumstances, the prefaces/ title-pages etc. tend to refer to ghosts writing, or the spirit of the dead, or how they live on beyond their time of death through their writing. The Workshop was actively selling to wealthy people, like Mary Sidney, the glorification of their loved ones, like Philip Sidney (her brother), through the bylines of texts attributed to him, even if he did not write anything during his short life. And they were also using the bylines of the dead because texts by the deceased tended to sell better, as the public had sympathy over their tragic passing. The direct note "Thomas Nash his Ghost" simply restates this insider joke about "ghost-writing". I explain this point across BRRAM. The idea of a dead "author" starting to write after their death with help from a ghostwriter is exactly what this term is designed to mean. Yes, it can also be applied to ghostwriters writing for living "authors", who hire them for the service, or living people who are credited with scandalous text etc. There is no better term in the English language that describes this spectrum of authorship activities under other bylines.

>54 faktorovich:

Yes, there are indeed "formulas" for what constitutes an infringement - this formula is the amount of words etc. per text that was previously found to be a violation in case law.

This is simply not true. You seem to think that the only factor in determining "fair use" is the amount of "words etc. per text". But there are other factors which you are ignoring, and courts consider all the factors in determining whether fair use applies.

>54 faktorovich:

It is true that there is no "bright line" or "magic formula" for the portion of a work that can be copied. When these cases do make it to court, it has less to do with the length of the material and more to do with the permissible uses granted under the Fair Use provision. There is a lengthy discussion on the Wikipedia page ( https://en.wikipedia.org/wiki/Fair_use ) with many case examples cited. For example:

The four factors of analysis for fair use set forth above derive from the opinion of Joseph Story in Folsom v. Marsh,5 in which the defendant had copied 353 pages from the plaintiff's 12-volume biography of George Washington in order to produce a separate two-volume work of his own.10 The court rejected the defendant's fair use defense with the following explanation:

A reviewer may fairly cite largely from the original work, if his design be really and truly to use the passages for the purposes of fair and reasonable criticism. On the other hand, it is as clear, that if he thus cites the most important parts of the work, with a view, not to criticize, but to supersede the use of the original work, and substitute the review for it, such a use will be deemed in law a piracy ...

In short, we must often ... look to the nature and objects of the selections made, the quantity and value of the materials used, and the degree in which the use may prejudice the sale, or diminish the profits, or supersede the objects, of the original work.

There is not the slightest chance that any reasonable person would consider the less than half a dozen images shown to be a replacement for the published work.

One of the reasons that there is an old proverb that says

A Man Who Is His Own Lawyer Has A Fool for a Client

is that having another person who is expert in the law review a claim will inform when the case is unlikely to prevail under the statute and case law. If one is the only person reviewing a case, it is too easy to let emotions guide decisions.

Here is another resource on Fair Use by the Copyright Alliance ( https://copyrightalliance.org/faqs/what-is-fair-use/ ).

Critique and academic discussion are key permissible uses under Fair Use that fully apply to this thread.

James

>55 faktorovich: The "Thomas Nash his Ghost" is equivalent grammatically to Modern English's "Thomas Nash's Ghost".

Ho, look! There's a ramp to a disagreement no one was trying to have! Turn before it's too late!

Edited to add: In "Thomas Nash his Ghost," the deceased Nash is the putative writer in whose name the text is issued, the pretended byline as it were. That is the exact opposite of the uncredited "ghost" publishing to the credit of someone else's false authorship. If I were a political strategist who wrote an op-ed to be published over a local official's name, he wouldn't be the "ghost." It would be me.

Similar deficiencies obtain in your argument from "ghostly" inspiration of scripture.

>47 Petroglyph: I have already created all of the "summary tables" that my studied needed to assist readers with understanding the findings. There are several different tables for the Renaissance that list different elements of the study that might be of use to future researchers. I explain what this data signifies and enhance it with additional research across BRRAM. There is nothing about what you are demanding I do to delete data out my tables to create short-answer cheat-sheets that would be at all useful to researchers who want to understand my exact findings. The texts I used are mostly digitized and available in the sources I list in the Bibliography. I cleaned them up using a standard method that most users should also use, and thus get very similar results. It would be unethical for me to re-publish hundreds of EEBO books on my own website (i.e.: copyrights).

Why are you assuming I have any invested interest in anybody using my method? It would indeed be a great benefit to them to use my method, as they would derive accurate attribution results. But there would be no benefit to me, especially if they did so without crediting my method, or books. And it is absolutely absurd to demand for a computational-linguistic method to be "easy". The process of establishing the true authorship of a pseudonymous text is as difficult of a challenge as solving a 400 year murder mystery. Anybody who thinks that an automatic answer will be generated without significant input of labor, should not be going into this field (which combines many skills, such as reading the texts in question and dissecting them beyond merely their linguistic content). Creating a program that makes this process seem easy by finding matches between isolated texts would be unethical, as establishing a similar/dissimilar answer is only a step in the process, whereas the more important step is checking for other potential similarities with other bylines etc. If you do not have the concentration to read over the entire methodology section in Volumes 1-2, and to closely review the provided tables on GitHub; you need to stop criticizing BRRAM, as you are repeatedly explaining that you have not even understood the basic steps of the method, nor analyzed the posted data. The data I have provided can be re-ordered in the Excel files to check what the ranges are for each of the ghostwriters; if you do not know how to do this, again, you need to stop complaining about a subject you lack the capacity to process. The data tables I have provided are extremely easy to use, as they include intricate information with simple headers that explain what information is contained in which section.

If you or anybody else told me that you want to apply my method, but you are struggling with its complexity, and want my help to make it easier. I would absolutely help you to make it easier. But that's not why you are posting this tirade, as your only goal is to discredit my method by proving it is too difficult for you (or anybody) to use. If you are actually trying to use my method, maybe begin by explaining what about my existing tables is confusing you, and what about my steps is taking too long. There should be a way to solve this problem for you, but it is not making a table with simple ranges.

I did follow instructions to adjust my employment of R/Stylo until I realized that the program is designed to be inaccessible and unusable to a general user, even if they read all of the available manuals (that do not explain the basic steps you guys instead had to give to me in this thread). In contrast, you have not told me about any glitches you have encountered with my method, as there are no such glitches.

The data tables are an overview of my findings. A summary of ranges for the ghostwriters is a simplification that only fits your request, and would only lead casual researchers to miss-quote my findings.

You are attempting to force your method onto me. It is not at all a part of my method to identify ghostwriters by any summary table of what I expect from their style. I attribute texts by putting them through the 27-tests, and I do not recommend to anybody to make attributions from a ranges table instead. It is just as wrong for somebody to engage in computational-linguistic authorial attribution without a PhD in English or computer science, as it would be for somebody to engage in a murder-investigation without being a licensed detective. One can stumble into the right answer, but one can also generate a false-positive, which can have detrimental consequences. Laypeople can use my method and they will generate accurate results, but interpreting their meaning will be challenging; it would be absolutely unethical for me to create a tool that would erode any degree of precision from this already simple approach to attribution.

>50 Petroglyph: It would be a nonsensically duplicating step to first normalize the data (PS: not all of it is in frequencies) into 1/0 or another -1 to 1 system and then again normalize it by replacing it with 1s and 0s.

>51 Petroglyph: If I felt that your conduct deserved legal repercussions, I would have already proceeded with filing a lawsuit. Most such lawsuits require that the aggrieved party first tell the infringer that one believes one is being infringed on. When I inform you that you are committing libel or copyrights infringement across this discussion, I am not doing it for the joy of complaining, but because these are required steps in solving such disputes. For example, if I told you that it is an infringement to post 5-6 pages of my book, and you added to it by posting 20 pages; you would be escalating the problem, after being informed it is a problem. Your knowledge of your actions being wrong is a significant ingredient in such cases. It would not be of benefit to me to go through your "people", as posting my complaints in this public forum is a legally recognized delivery of information to you (the infringing party), and it also serves as a record because it is public and seen by many other people who can testify of seeing me share the information. There is absolutely nothing to be gained by complaining to LibraryThing, as, for example, the rep from LibraryThing posted insulting remarks about my research at the top of this "2" thread: "she's jumped to conclusions based on suspicious data generated by faulty methodology". I have previously complained to LibraryThing about harassing posts in this thread, and they responded that they do not feel you guys are doing anything wrong. This is the standard response from all content-"moderating" companies. I am not at all surprised by it. I am just explaining to you how it all works, in case you are not aware.

>61 faktorovich:

the rep from LibraryThing posted insulting remarks about my research at the top of this "2" thread

amanda4242 is not a LibraryThing employee, merely the person who created a new thread in order to spare all our computers from trying to load the original thread. No one from LibraryThing has commented in this thread continuation. You can tell because when LibraryThing employees post there's a little "L" icon in front of their username.

>52 Petroglyph: One of the reasons scholars do not post images of quotes, but rather transcribe the entire section that one is criticizing is because it allows them to re-read the section as they transcribe it to re-consider its meaning. So, to review my own intention in this isolated quote, I am not going to read over your rambling comments about 2%, and will instead rephrase what the rest of the paragraph in question stated. I begin by pointing out that it is nonsensical for a computational-linguist such as Jackson to claim that he is checking word frequencies for words “occurring 2-6 times in the canon” and those “occurring 2-10 times”. This statement duplicates of repeats the testing of words that occur 2-6 times within the second phrase, while adding a few of the more frequent 7-10 words. Then, I explain that in a corpus will millions of words, 40-75% can be "hapax legomena" or 1-time occurring words; larger texts lean towards 40% and shorter texts to 75%. The size of the text thus significantly impacts the word-frequency measure, and most of these computational-linguists chop up small fragments of text, such as 500-words, and thus have samples with around 75% of 1-time words. Then, I come to the sentence in question. I am citing Kennedy's book when I summarize his findings, which would make more sense if a reader read those before reading my summary. The "10-million words corpus" measure describes the total number of words in a corpus being tested (not in individual texts, and not the number of words when repeating words are only counted once). The "2,000 most frequent words" phrase clearly explains that it is referring to the "most frequent words" (or they are counted only once in this 2,000 measure, so that if "a" occurs 500 times, it would only add 1 word to this "2,000" count). I can see how a reader can be confused what "2% of all" and "80% of all words (8 million)" signifies, but there is enough information here for the reader to guess the 2% is not arrived at by a division of 2,000 by 10 million. 80% are indeed of 10 million. And 2% is a unique measure that Kennedy calculates in his book when he counts the number of words that are part of these 2,000 frequent words (when counting each of these separately). If there is something wrong with this 2% figure, it would be a good idea to complain to Kennedy. My paragraph is explaining the errors with previous computational-linguistic studies, so if you believe you have found an error in Kennedy's claim, you are only adding support for my argument. My point is in the following sentences, where I explain that excluding the 1-time occurring words and those that occur more than 10-times, Jackson is excluding the significant most-frequent words; and additionally, Kennedy's and other linguists claims that such words would make up 80%+75% or 155% of a text cannot be logical, so something must have gone wrong in their calculation of what percentage most and least frequent words make up. The confusion over the 2%/80% calculation is precisely what I am criticizing in this paragraph. I then conclude that the percentage of 1-occurence, or 10+ occurring words, or those in the middle becomes irrelevant, when one simply compares the most-frequent words, which establish a linguistic pattern similarly in smaller and larger text sizes.

>61 faktorovich: As >62 coprime: says, I am not a LT employee. I'm merely the member who continued the previous thread.

My summary in >2 amanda4242: reads, "The author reacts poorly when everyone says she's jumped to conclusions based on suspicious data generated by faulty methodology." I am summarizing the opinions expressed by posters on the previous thread; that many of those posters expressed disbelief in your work is not an insult but a statement of fact.

>56 lilithcat: There are indeed many factors in a "fair use" dispute. Neither you in your post, nor I in my post named all possible factors that can impact a "fair use" judgement. I have pointed to the essential points you should all be aware of because they have been specifically violated in this discussion. And courts cannot logically consider "all the factors" in anything; they can only consider the relevant factors. There is actually an objection for "irrelevance" that can be made in court.

>57 Keeline: The actual case you cite "Folsom v. Marsh", if it (or the section you cite) was used as the sole case to base the decision on, would have decided a case about the use of the 6 pages in my favor. "The court rejected the defendant's fair use defense". The explanation then clarifies that the duplication and publication of another's copyrighted materials at-length without engagement with all of it in scholarship is a violation. Similarly, posting 5/6 pages from my book, when only discussing that it is a long paragraph is a violation. The intent of posting such reproductions can be not to profit from the resale, but rather to decrease the number of people who would need to purchase my book, by making a section of it criticized as the worst part of the book publicly available for free. Publishing a pirated book to shrink profits for the author is just as much of a violation of copyrights, as selling the book or section of a book for direct profit.

>62 coprime: Amanda conducted an interview with me on behalf of LibraryThing, saying it was for their newsletter. If she is not affiliated with LibraryThing, this is indeed a curious line of inquiry.

>67 faktorovich: No she didn't, that was AbigailAdams26, look at the top of the previous thread!

>64 amanda4242: Yes, this is why you believe it is acceptable for you to make such a post. But your summary and other posts are all extremely biased against me in a manner that is more of a direct attack than many of the other posters. For example, summarizing that "everyone" says my "suspicious data" is "generated by faulty methodology" includes the speaker, Amanda, in the statement, so you are expressing your own opinion as well as "everyone" else's. If you are going to insult a researcher in a public forum, there is no room to hide behind such publicly made comments, so you might as well acknowledge you are doing so with malicious bias. My data is fully accessible so it is not "suspicious". And my methodology works to generate precisely accurate attribution results, so it is not "faulty". Instead of providing any evidence to support such libelous accusations, "everyone" here has instead been just repeating the libelous statements under the assumption that if you repeat a lie enough times, it becomes the truth. But this is just never going to happen, when the truth is still the truth, and my data and method speak extensively for themselves.

>67 faktorovich: I most certainly did not conduct the interview! AbigailAdams26 did and I clearly state in >1 amanda4242: that I was copying and pasting Abigail's introduction from the previous thread. Careful reading will save you from making such embarrassing mistakes.

>68 FAMeulstee: Thanks for pointing this out that is helpful. Since I did not check to see the name of the person who had interviewed me, other readers are likely to have assumed Amanda was a LibraryThing editor, and that's why she took it upon herself to police the previous thread, push everybody off it and push everybody into this new thread. It helps to have this explanation at the top of the thread for anybody who might reach a similar misunderstanding.

>67 faktorovich:

Amanda conducted an interview with me on behalf of LibraryThing, saying it was for their newsletter. If she is not affiliated with LibraryThing, this is indeed a curious line of inquiry.

There are two people involved here. One is AbigailAdams26 who is a LibraryThing employee (again, you can tell because there is an "L" icon beside her username). AbigailAdams26 is the person who started the original thread and the person who interviewed you.

The other person is amanda4242. amanda4242 is the person who continued this thread in a second thread because the original took a while to load due to how long it had gotten. It is common LibraryThing user etiquette for someone to start a new thread when the original gets long; anyone can do this. Doing this is not reserved for LibraryThing employees and is frequently done by people who are not LibraryThing employees.

amanda4242 a different person from AbigailAdams26.

>68 FAMeulstee:

All those "A"-initialed people look the same.

>71 faktorovich: other readers are likely to have assumed

Not everybody has the chip on your shoulder or your paranoid ideation.

>71 faktorovich: No she didn't, it is usual. Anyone can continue a thread when it becomes way too long.
No one else but you would assume she is a LibraryThing employee. As you could see at the top of the previous thread, there is a "L-icon" next to the name of LibraryThing employees.

>73 paradoxosalpha: LOL, my first name starts also with an "A", I might look identical ;-)

>66 faktorovich: Kindly read it again and note the basis for the rejection:

defendant had copied 353 pages

and the court's decision:

A reviewer may fairly cite largely from the original work, if his design be really and truly to use the passages for the purposes of fair and reasonable criticism. On the other hand, it is as clear, that if he thus cites the most important parts of the work, with a view, not to criticize, but to supersede the use of the original work, and substitute the review for it, such a use will be deemed in law a piracy ...

Are you really trying to claim that those 5 or so pages represents the essential part of your 17+ volume series?

This is not the same case since the usage here is to analyze and critique an aspect of your presentation and content. This is how academic reviews work in most cases. Some diagram or text is presented and the critique responds to it.

In a fair world, lawsuits would be the complete responsibility of the losing party, including any attorney fees to defend against the frivolous claims. Sometimes this is true. Often it is not.

In reality, most copyright and other IP infringement claims are settled out of court, making the circumstances of any given case harder to use as a precedent for future cases.

James

I just want to make a few things very, very clear:

I am not now, nor have I ever been, a LibraryThing employee.

I have never claimed to be a LT employee.

I have never conducted an interview on behalf of LT.

I have never claimed I have conducted an interview on behalf of LT.

I made it clear in >1 amanda4242: that I was copying and pasting AbigailAdams26 introductory posts by making the very first sentence of >1 amanda4242: "Copied and pasted from AbigailAdams26 introductory post https://www.librarything.com/topic/337240#7670449."

All LT members have the ability to continue a thread once it surpasses 150 posts. All it takes to continue a thread is the willingness to click the "Continue this topic in another topic" link that appears once a thread has gone over 150 posts.

As I stated in post #1148, the previous thread was taking far too long to load so I continued it as a courtesy to other members.

I did not continue the thread to try to halt the discussion, but rather to make it easier for it to continue. I even explained in post #1153 how to link to posts in the previous thread so members could easily reference the previous thread.

Many people expressed unflattering opinions of faktorovich's study in the previous thread. To summarize part of the thread by saying "everyone says she's jumped to conclusions based on suspicious data generated by faulty methodology" is not an insult, but a statement of fact.

That I also have an unflattering opinion of faktorovich's work does not make the summary any less true.

I am not AbigailAdams26.

I'll just leave this here; it seems apposite.

Faktorovich, page 1 msg #1050: "I have not made a single false statement in this thread, and I don't believe I have ever made a false statement in my life."

>36 faktorovich: Yes, I think more clarification is called for. Do you have other examples?

This seems like another example of "reading between the lines" (or "subtext") to go from a literal meaning to the figurative meaning that you wish to apply.

I don't see "ghost writer" with anything or nothing between the words in your example.

The usage in some religions of "the Holy Ghost" has nothing at all to do with writing or ghostwriting. It would be expected to find "ghost" used in a Bible or similar religious volume.

Perhaps this blog-like article is not authoritative. But the author claims that "ghostwriter" was coined in 1921 ( https://www.thewritersforhire.com/whats-the-origin-of-the-term-ghostwriter-and-h... ).

It seems like it would be earlier and certainly the concept is much older, just not the word. Stratemeyer always referred to his people as "writers" not "ghostwriters". I have looked at hundreds (perhaps thousands) of his letters from the first 25 years of his Syndicate. When a "ghost" is mentioned, it is a supernatural spirit or at least the representation of one since "ghosts are not real" is a tenet of these books (unlike the recent CW Nancy Drew TV show). Here's one example that Stratemeyer personally wrote.

https://stratemeyer.org/edward-stratemeyer-author-and-book-packager/dime-novels-...

{The original story in a newspaper did not have an illustration. A period illustration was added which fits the text.}

The extra term is needed to clarify a distinction between the owner of a copyrighted work and one who may have contributed to the content of the work such as writing some or all of it. For copyright, the "author" is the "owner," even when there is a clear transfer of title for the work (product and labor).

James

Is it just me, or does anyone else see it as a bad sign that the owner of a small publishing house doesn't know relatively basic things about copyright?

Anyway, >29 Petroglyph: – my pleasure, and I have to say that as frustrating it is to read even a single post by Faktorovich, it is counterbalanced by the posts by your good self and everyone else who does know what they are talking about. In fact, I'm becoming more and more inspired to get back to doing a bit of research, myself, and that can't be a bad thing :-)

-Linnéa

>61 faktorovich:
K.

>63 faktorovich:
I am not going to read over your rambling comments about 2%, and will instead rephrase what the rest of the paragraph in question stated

Yeah: your usual modus operandi.

In brief: you pulled the 2% figure from yet another source, and felt it was appropriate to add this proportional figure into a completely different context and apply this proportional figure to some unrelated numbers. Alright.

I was wrong. You're not just confused about word types and word tokens; you're even more confused. There's always a deeper fractal.

>59 faktorovich:

Well, it is my suspicion that the only way to achieve your results is to repeat all your steps one by one. Sounds like you agree with me.

I'll stop prodding you to provide a clear, unambiguous, non-overlapping set of criteria that demarkate your ghostwriters from one another. Pity. It would be such a revealing table.

>78 Keeline: The earliest attestation of "ghost writer" in the OED is from 1908, from a local newspaper in Lincoln (Nebraska): "There are a dozen or so known, catalogued and labeled ‘ghost’ writers in town. Few of them have ever seen their names in print."

"Ghost-written" is first attested in Arthur Ransome's Bohemia in London, published in 1907. "There are agents who make a living by supplying ghost-written books to publishers who keep up for appearance sake the pretence of not being in the know."

The very earliest use of "ghost" meaning "ghostwriter that the OED mentions is from as early as 1881 (so still more than 200 years after the end of the English Renaissance) : "They have never been introduced to a ‘ghost’—that is to say, a person employed by incompetent artists secretly to do up their work and make it artistic." Vanity Fair 24 Sept 1881.

-Linnéa

>60 faktorovich: That's.... not what normalizing is.

-Linnéa

>73 paradoxosalpha: All those "A"-initialed people look the same.

"Well if they didn't want me to confuse them, they shouldn't have put numbers in their name."

>79 anglemark:

Cheers!

The fun thing about doing more research is that you can do it pretty much any time. Sadly, that is also one of the drawbacks: you can't just leave it at work and stop thinking about it. But if this whole debacle inspires you to do research, it'll have been worth it!

faktorovich (#71):

I know you have said that you never make mistakes, but this is one case where saying "oops, got the names mixed up, sorry" would really be to your advantage. Nobody else thought amanda4242 and AbigailAdams26 are the same person just because their names both start with the same letter, especially since amanda4242 doesn't have the characteristic "L" icon indicating an LT employee. (It's okay that you didn't notice it, especially if prior to this thread you weren't active on Talk much, but don't assume everyone else made the same error.)

>78 Keeline:

Oh, never mind. Next time I'll read all the posts . . .

But the author claims that "ghostwriter" was coined in 1921

The OED gives an earlier use, citing a July 23, 1908, article in the Lincoln, Nebraska, Daily Star: "There are a dozen or so known, catalogued and labeled ‘ghost’ writers in town. Few of them have ever seen their names in print."

>83 anglemark: That seems closer to what I would imagine for public usage of the phrase. As I wrote, the 1921 seemed hard to believe, particularly with the extensive use of the process in the 19thC by story paper and dime novel publishers. It might go back earlier than 1881.

Often there is an understanding of a writer adopting a personal nom de plume ("Mark Twain" = Samuel L. Clemens, etc.) but "house names" are far less understood, even by many today. Library cataloging systems, including LT, don't have good ways to handle the published name vs. the person or people who worked on it for even a single work. But that is a topic for another thread.

Curious, I looked at my notes and transcriptions of Syndicate correspondence with 4,372 outgoing letters and 2,656 incoming letters. I have page images for tens of thousands of pages but a good portion of them are not relevant for my work. I don't note and transcribe everything, mainly the interesting ones.

Edward Stratemeyer died on May 10, 1930. As I wrote, he called his people "writers" and "artists" (for those who made illustrations). The first use of "ghost writing" appeared in a 1943 letter to a publisher editor when she passed along a request of an autographed Nancy Drew book. Here are the first two paragraphs of the reply:

Following a precedent decided upon years ago among Mr. Stratemeyer, Mr. Alexander Grosset and Mr. Reed, none of the series books ever have been autographed. It is unfortunate that this word was not passed along to you when you became head of the juvenile department, and I hasten to explain the reasons to you, why it is inadvisable for our writers to give out autographs.

In the first place, the names on the series are pseudonyms, which are the property of the Syndicate. Most of the authors of the stories are engaged in some other type of work which makes it inexpedient for their identity to be revealed, and in a few case in the past where a writer has died, the series has been carried on by someone else. Furthermore, as you know, our Syndicate retains all rights to these series, and the "ghost" writers have no further claim on them, once the manuscripts are accepted by us. You can readily see why it has seemed best not only to keep these matters a business secret among Grosset and Dunlap, our writers and ourselves, but also why it is not feasible to have signatures put in our books.

I would note that the Syndicate did sign some books in later decades, particularly in the 1970s.

There are other references to ghost in this context in the 1940s and 1950s. By this point the conventional usage of "ghostwriter" was being used by the principals in the Stratemeyer Syndicate.

In this field there are many terms that collectors have coined to describe features or processes connected with series books. For example, the practice of issuing three or perhaps more books at one time to get a series launched are called "a breeder set" or "breeders" by collectors. But you don't find this term in the contracts and correspondence of the period for the Syndicate. The earliest use I have found for series books is in a 1976 memoir by Leslie McFarlane and this is probably where the collectors read it and started to use it.

The collectors refer to blue spine Hardy Boys and yellow spine Nancy Drew books as "pictorial covers" (1962+ for most titles) but the trade referred to these as "art boards" in the correspondence. Other terms were used by book manufacturers in the trade publications. Similarly, a "next-title announcement" used by collectors is a "throw-ahead notice" by people working on the books.

There are many more examples where the terminology after the fact is different than what the people doing the work would use. That's why I was really surprised by the claim that "ghost writing" in ay variant was used in the 1500s. I shall await an example I can actually see.

James

>60 faktorovich:

It would be a nonsensically duplicating step to first normalize the data (PS: not all of it is in frequencies) into 1/0 or another -1 to 1 system and then again normalize it by replacing it with 1s and 0s.

Ok. If that is your stated opinion.

So please explain to me why you normalize the frequencies of the punctuation marks you include in your tables before you do your ones/zeroes thing to these figures.

Edit: and please explain why you also normalize the parts of speech counts. (But to a different baseline than the punctuation marks.)

>83 anglemark:
>89 Keeline:

It's very common for phrases to be in use for a while before they are put down in writing. The OED staff are very proficient in tracking down very early (if not the earliest) recorded attestations, so missing it by like four hundred years is not something I would expect from them.

>91 Petroglyph: In the case of the Lincoln, Nebraska newspaper, when I read it (from NewspaperArchive.com) it sounds like a reprint from another publication like one of the New York City newspapers (often without credit). This was a very common occurrence. Here is the article.



James

>92 Keeline:
Thanks for tracking down that article!

Archival research always turns up these whiplash-inducing articles -- I thought that "the ermine -- née tiger skin -- rug" was funny. And then I got to the laundry list bit.

>75 Keeline: There are no non-essential pages in my 17+ volume series, and it would be absurd to consider these pages as a percentage of the full series and not at most only of Volumes 1-2 (as this is a separate book with its own ISBN number). After I informed Petroglyph that I viewed his posting of designed pages from my book as a plagiarism, he posted at least one other different page deliberately to harass me (after having learned this activity troubles me). This would prove malicious intent, as opposed to an accidental or an unknowing piracy of copyrighted materials. He did not discuss the content of the pages in question, so he did not comply with the "fair use" allowance. "Academic reviews" are when somebody engages with the content they are borrowing to make an argument of their own with help from this content. But failing to review any details or to summarize the material in question beyond just saying it's a long paragraph; Petroglyph did not meet the minimum requirements for "fair" scholarly "use". I have never filed a "frivolous" lawsuit. I only file when there is overwhelming evidence that I have been wronged. Since Petroglyph cannot even comply with my request to stop posting screenshots and to only post transcribed quotes of no more than a paragraph and with explanations of the content; it is unlikely he would ever "settle" any case.

>94 faktorovich:

I viewed his posting of designed pages from my book as a plagiarism

How could this possibly be plagiarism? The last thing Petroglyph did was claim that he wrote them! (Or do you have a different definition of "plagiarism" than the rest of the world?)

>78 Keeline: You are all intentionally blind to every one of my actual arguments, as you keep preferring to make up what you imagine I am arguing instead.

In this case you have missed the joke the Workshop was employing. Yes, I know the term "Holy Ghost", and the Workshop subverted this term (across the century when they ghostwrote) to apply it to this Holy Ghost writing or the Ghost of the Deceased writing posthumously with help from a "Ghost-Writer". This joke is not only lost on you, but it was missed by scholars of this period for a few hundred years.

Before 1921, the term was revived as early as 1889 to describe "Miss Carrol" as a "ghostwriter": https://www.google.com/books/edition/Proceedings_of_the_Botanical_Section_in/WGe...

In 1895, Hawthorne is described lurking like a "ghost writer" of short stories: https://www.google.com/books/edition/Harper_s_Round_Table/xFQ_AQAAMAAJ?hl=en&amp...

A man describes his profession as a "ghostwriter" in court proceedings in 1896: https://www.google.com/books/edition/Z%C3%A9linde_comedie/srgvNbywffEC?hl=en&amp...

It appeared in a non-fiction story called "The Ghost Writer and His Story" by Graves Clark (1919), where it was applied to a hack writer of ghost stories: https://www.google.com/books/edition/The_Editor/M58uAQAAIAAJ?hl=en&gbpv=1&am...

The term "'ghost' writer" is also used in Publishers Weekly (1916): https://www.google.com/books/edition/The_Publishers_Weekly/hygzAQAAMAAJ?hl=en&am...

As you point out, the only reference to a "writer" I found on the page you liked to is: "Founded Syndicate in 1905 to hire ghostwriters to write books from his outlines." There is no hint of his application of the term "ghostwriter" in the books he had ghostwritten for him.

Yes, an "author" is generally the copyright holder, or the byline that is used in obtaining copyrights. However, if a work is pirated, the stated "author" might not be the legal owner of copyrights. And there are several other complex layers of meaning to the act of writing a text without receiving byline-credit for it, and all of these are best summarized with the term "ghostwriter". You have not suggested an alternative with equally multifaceted applications.

>81 Petroglyph: In brief, I used the 2% figure to demonstrate how a couple of previous computational-linguists used misleading and confusing statistical figures. You only read the confusing figure portion of the paragraph, realized that it was confusing, and thus leaped to the conclusion that I was the one who had created confusing data.

>82 Petroglyph: Yes, I designed a method with very specific steps that are all absolutely necessary to arrive at an accurate attribution. There is no shortcut without designing a program to execute all of these steps for the user, which would become progressively harder to build the more steps you want it to do for you (such as cleaning a file of glitches in the transcription, or checking the birth and death years of all of the bylines from the relevant period in the relevant genre(s)).

>96 faktorovich:

A man describes his profession as a "ghostwriter" in court proceedings in 1896: https://www.google.com/books/edition/Z%C3%A9linde_comedie/srgvNbywffEC?hl=en&amp....

Wrong. That case is from 1940: https://casetext.com/case/cassidy-v-gannett-co-inc

>94 faktorovich: After I informed Petroglyph that I viewed his posting of designed pages from my book as a plagiarism, he posted at least one other different page

Are you referring to the page in >52 Petroglyph:? I already posted that page in >17 Petroglyph:. That screenshot was one of the things you were complaining about.

In >52 Petroglyph:, I explicitly said "Note: this screenshot is for the purposes of criticism." Like, right next to it.

But I guess that "must complain about something" and "I IS VICTIM" being decoupled from actual facts is a persistent trend with you, so I'm not surprised. Lies, Faktorovich. Lies.

>97 faktorovich:

K.

>98 faktorovich:

Your method only works if you go through all the steps and take on board all the confirmation bias, apophenia, creative counting via red-colour method, irrelevant tests, data-obliteration via 1/0, and all the other Greatest Hits, and if you have a sufficiently gradual creep in what the six ghost-writers' styles represent so they can be widened in small steps and accommodate an ever-increasing body of work. I suspect that, if you were to put actual ranges (of % passive voice, clout, adjectives, lexical density and all the others) to your six authors, that you'd end up with a set of data rows that would not unambiguously distinguish between the six.

Prove me wrong. Show me that table with six distinct authorial signatures as "evidenced" by your data.

>89 Keeline: "Nom de plum", or pen name, is the same as a pseudonym, but none of these have been changed into verbs such as "ghostwrote", so they are not suitable for broad application in a book about a Workshop of Ghostwriters. It would not even be grammatical to say a Pseudonym(s?) Workshop. I explained already that the Workshop used this term in the "Nashe"-byline and cited several other examples of its application. The Workshop started many ideas that appear "new" to scholars today; for example, Verstegan's "Restitution" is credited by some with starting the "Saxon"/"Anglo-Saxon" origins (that morphed eventually into the Aryan concept) concept, whereas other scholars credit much later scholars. Before Verstegan, "Saxon" mostly referred to the people of the Duchy of Saxony. There are many other points, I raise across BRRAM, and you guys seem to be glued to "ghostwriter", without even at least reading my previous explanations why this was indeed the Workshop's humorous invention.

>59 faktorovich: "And it is absolutely absurd to demand for a computational-linguistic method to be 'easy'."

"I did follow instructions to adjust my employment of R/Stylo until I realized that the program is designed to be inaccessible and unusable to a general user"

You couldn't make this up!

>90 Petroglyph: I do not "normalize" any of the data before turning it into a binary and thereby normalizing it. I enter texts into automated counting systems that return a frequency of punctuation marks or parts of speech, etc. The parts-of-speech numbers are given as a percentage of all word-types in the text. There is nothing "normalized" about a number like "10.44 %" of verbs, as this is not a number between -1 and 1. You have no idea what you are yelling about, so figure out what you are trying to say, or stop yelling.

>95 lilithcat: You are right, it is piracy. This typo is a testament of how emotionally impacted I am by somebody republishing or pirating my work without permission, and then attacking me for pointing out that this action deeply troubles me, as I find it to be technically illegal.

>96 faktorovich:

For "Miss Carrol" I get "no preview available", so I can't check that.

Hawthorne as a "ghost writer" is an artefact of the text being in two columns. The actual sentence, at the bottom of the left-hand column, is "lurked like a ghost among the old church-yards".

>99 lilithcat: Wow. You guys are seriously committed to finding mistakes in every word I write. I posted 5 examples that Google Books judged to have been published prior to 1921, to correct one of your claims about the first application of this term, and you looked up all of them to find the one that was incorrectly labeled by Google with the wrong year of publication? You should contact Google Books to file a complaint. The other examples are still valid cases of earlier applications of this term that whoever claimed the earliest usage was in 1921 was not aware of. We have been able to find in this thread usages as early as 1881 and 1896, and yet these are not mentioned in most definitions of the term "ghostwriter", so it is entirely likely that the Renaissance Workshop used variants of the term "ghost-writer" without these registering in any dictionaries or discussions about this term.

>100 Petroglyph: Including a note that you are pirating a text in "fair use" does not counter the act of piracy, but rather stresses that you are aware your action is a violation and that you are attempting to get away with it by using an empty descriptor.

>102 Petroglyph: You have previously stated that you have stopped repeating that I should create a nonsensical summary table that simplifies my data, and now you have proven your previous statement to have been a lie by again repeating this nonsensical request. There is no "data-obliteration" by applying the binary division, as it achieves the exact "normalization" you were just insisting on a few posts ago.

>94 faktorovich:

I viewed his posting of designed pages from my book as a plagiarism

I would note that plagiarism is not illegal; it is merely a violation of some codes of ethics at universities and some professions. It is defined as passing another person's work off as your own.

Perhaps you are conflating this with copyright infringement. However, if you are going to represent your company in legal matters, you should learn the difference.

James

>107 Petroglyph: You're right, that's 2 out of six. How about you explain the other 4. I challenge you to find something wrong with all of them. I just briefly found some curious examples of "ghostwriter" usages in response to your request. I have never reviewed the application of this term in the 19th or 20th century before, as these earlier usages are irrelevant, when the term I am borrowing is from the Renaissance.

>96 faktorovich: In 1895, Hawthorne is described lurking like a "ghost writer" of short stories: https://www.google.com/books/edition/Harper_s_Round_Table/xFQ_AQAAMAAJ?hl=en&amp....
It appeared in a non-fiction story called "The Ghost Writer and His Story" by Graves Clark (1919), where it was applied to a hack writer of ghost stories: https://www.google.com/books/edition/The_Editor/M58uAQAAIAAJ?hl=en&gbpv=1&am....
In 1895, Hawthorne is described lurking like a "ghost writer" of short stories: https://www.google.com/books/edition/Harper_s_Round_Table/xFQ_AQAAMAAJ?hl=en&amp....
The term "'ghost' writer" is also used in Publishers Weekly (1916): https://www.google.com/books/edition/The_Publishers_Weekly/hygzAQAAMAAJ?hl=en&am....

In all three of these, the terms are expressly used to denote writers on the topic of supernatural ghosts or spectral entities. None of them have anything whatsoever to do with feigned authorial attributions.

>111 Keeline: Being so harassed by a barrage of insults that I make a typo and confuse "plagiarism" with "piracy" is proof of emotional distress. I would not make this mistake when calmly designing a court filing. The terms are related: "plagiarism can warrant legal action if it infringes upon the original author’s copyright, patent, or trademark." While Petroglyph did not put his name on the pages he pirated, he did not put my name on it either, as he did not cite the page numbers, or chapters from which he took these pages, and he did not even include a footnote that specifically identified these pages as coming from my book. Only at the bottom of his post did he cite the volume in general. So, I was not incorrect to call it "plagiarism"; for example, synonyms of "piracy" include "plagiarism": https://www.thesaurus.com/browse/piracy There is really no good term for stealing a book equivalent to "plagiarism", as copying, infringement, piracy etc. do not really convey the precise intended meaning. "Plagiarism" can be unintentional: https://www.ox.ac.uk/students/academic/guidance/skills/plagiarism So by failing to include my name while posting my work, this can be an act of unintentional plagiarism, including if most readers would not assume that you were attempting to claim you wrote my book.

>113 paradoxosalpha: The description of the author specifically as a "hack" writer before she is described as a writer of "ghost stories" humorously explains the title of the story is about hack-writing or ghost-writing and not merely about writing about ghosts. And the Publishers Weekly article similarly stresses that the "ghost writer's" work occasionally actually has "real artistic merit", a phrasing that is commonly applied to distinguish hack/ ghost writers from highbrow or true "writers".

>112 faktorovich:

>108 faktorovich: "I posted 5 examples that Google Books judged to have been published prior to 1921, to correct one of your claims about the first application of this term, and you looked up all of them to find the one that was incorrectly labeled by Google with the wrong year of publication?"

(implying it's wrong of us to do so. Taken out of context, this is something a legitimately upset person would say)

Also >112 faktorovich:: I challenge you to find something wrong with all of them"

(stating we should do so; taken out of context, this is something a defiant person would say)

Also >112 faktorovich: "I just briefly found some curious examples of "ghostwriter" usages in response to your request. I have never reviewed the application of this term in the 19th or 20th century before, as these earlier usages are irrelevant, when the term I am borrowing is from the Renaissance"

(she wasn't serious, you guys, it was just a quick search don't take it seriously). Back-pedaling time

>108 faktorovich: "The other examples are still valid cases of earlier applications of this term that whoever claimed the earliest usage was in 1921 was not aware of. We have been able to find in this thread usages as early as 1881 and 1896,"

(she's still right, and these claims were serious. Taken out of context: this is something a confident expert would say.)

>115 faktorovich: The description of the author specifically as a "hack" writer before she is described as a writer of "ghost stories" humorously explains the title of the story is about hack-writing or ghost-writing and not merely about writing about ghosts.

"Hack" is a pejorative label for a writer who produces in quantity and on demand. Although there is certainly an overlap between ghostwriters and hacks, there is nothing like an equivalence between the terms. Writers of supernatural stories for the popular market have been frequently derided as "hacks" when honestly using their own bylines.

>105 faktorovich:

A note about normalization.

For texts that do not have similar lengths, counting absolute frequencies and judging texts as "similar/not similar" based on those absolute frequencies leads to wrong results. Obviously. I shouldn't have to point this out. And Faktorovich indeed makes use of relativized frequencies (though not always!).

So you should at least relativize or even "normalize" your data -- transform absolute frequencies by relating them to a shared norm, say, recalculating them to the average frequency per 10,000 or 100,000 words, or per 100 or 1000 sentences. That way, when you compare many different texts of different lengths, all your frequencies for that particular test will have the same baseline. Maternal mortality and other medical causes of death (such as covid-19) are often expressed in incidences per 100,000. This is to accommodate both very frequent diseases and very rare conditions.

Expressing absolute frequencies in percentages is a form of relativizing them: you are recalculating them relative to a shared basis of "100". In scientific results, percentages are often expressed as a range between 0.0 and 1.0¹ (thanks, Faktorovich, for googling that for us in >41 faktorovich:). In corpus linguistics, textual features are often expressed per 100 sentences, per 100,000 words, or even per million. Here is a graph from O'Keeffe et al. (2007, p.14), a book I know Faktorovich has at least briefly glanced at:



The website Analyzemywriting.com, one of Faktorovich's online services, presents the punctuation marks both in absolute frequencies, and terms of per 100 sentences:



In her tables, Faktorovich uses these normalized results. Pace her comments in >60 faktorovich: "a nonsensically duplicating step" and >105 faktorovich:, "I do not "normalize" any of the data before turning it into a binary and thereby normalizing it"

That website also offers the option of normalizing punctuation per 1000 sentences, or even per 100/1000 words:



But that is an extra two clicks.

---

¹ Normalization between 0.0 and 1.0 would take the form of 0.01 for 1%, and 83% would be 0.83. It's not the 0/1 either/or choice that Faktorovich imagines it is.

>118 Petroglyph:

If I stop posting here after this, it's probably because I'll have been speedily deported to the US to stand trial for my plagiarism piracy of analyzemywriting.com's website.

>108 faktorovich: faktorovich: "I posted 5 examples that Google Books judged to have been published prior to 1921, to correct one of your claims about the first application of this term, and you looked up all of them to find the one that was incorrectly labeled by Google with the wrong year of publication?"

If you looked at any posts after #78, you would see the clarification that I and others made to find earlier examples. In #78 I expressed doubt about the 1921 claim. Stuff on the Internet is often wrong so each claim needs to be assessed to see if it is valid.

Perhaps this blog-like article is not authoritative. But the author claims that "ghostwriter" was coined in 1921 ( https://www.thewritersforhire.com/whats-the-origin-of-the-term-ghostwriter-and-h.... ).

It seems like it would be earlier and certainly the concept is much older, just not the word.

It was not my claim. It was the claim of the author of that page. They were wrong.

So the only partially successful attempt to show earlier examples of the word or concept is not especially necessary since it only joins the content of replies #83, #88, #91, and #92. Welcome to the party.

When someone claims that they never make an error, when the conversations show many examples of provable errors (not merely differences of opinion), how much confidence are we to have in the body of work being discussed here?

Now since this is starting to get emotional, rather than focusing on the theory, methods, and results of these extraordinary attribution claims, I suggest taking a break from the forum. It will still be here later on.

James

>96 faktorovich: the term was revived as early as 1889 to describe "Miss Carrol" as a "ghostwriter":

If you read the page, available at https://babel.hathitrust.org/cgi/pt?id=hvd.hx7rw7&view=1up&seq=208&q... , you'll see it mentions a book published in 1940. Obviously this is misdated by Google, which is not infrequent. It's irrelevant; whether ghostwriter was first used in 1889 or 1921, it's still a word created long after the British Renaissance, which makes your stretched interpretations of "ghost" in those works completely untenable.

Actually, that got me thinking.

Faktorovich uses analyzemywriting.com as a source for her data. Not occasionally, but systematically. She puts that in books and tries to sell them. Can she do that?

From their "terms" page, emphasis added:

You must not conduct any systematic or automated data collection activities on or in relation to this website without www.analyzemywriting.com's express written consent.
This includes:

• scraping
  
• data mining
  
• data extraction
  
• data harvesting


• 'framing' (iframes)
  
• Article 'Spinning'

{...}
You must not use this website for any purposes related to marketing without the express written consent of www.analyzemywriting.com.

Faktorovich, do you have their express, prior, written consent to use their stuff in the books you sell on your website?

>121 prosfilaes:

Yes, it looks as though Google digitized several different and unrelated items, yet catalogued them all under Proceedings of the botanical section in memory of Dr. Asa Gray (Brackets around "of the botanical section" removed by me because they created a weird touchstone.)

>120 Keeline:

Peace, Keeline. I appreciate your calm, rational, and knowledgeable contributions.

I just would like to thank everybody posting in this thread!
I'm learning a lot.

>125 indeedox: I'm learning a lot.

I think every single person reading this thread is. Well, with one exception.

-Johan

>107 Petroglyph:

You couldn't make it up. This is such a basic mistake, a mistake I wouldn't expect an 11 year old to make, it can only be an attempt at a deliberate misleading statement - despite faktorovich knowing that people here are more than willing to double check her statements.

>127 andyl: No, I think it's just a result of searching GBooks really quickly for something that could "prove" her theory, and posting the search result without checking it. The autogenerated GBooks preview says

pastures and meadows , hills public , which had become familiar with Hawthorne as a and valleys and wild - pine groves , and lurked like a ghost writer of short stories , now saw that it had been entertaining a genius unawares .

You or I would have checked the page in the book to see what it really said, but that requires an extra click.

I don't think using the term "ghostwriter" is a problem in itself, because if you are talking about the group and their (hypothesized) function, you'll need a term for it. (The same thing goes for "byline", which to me is a much weirder choice.) What is completely bonkers more problematic about "ghostwriter" is the arsy-versy etymology.

-Linnéa

>128 anglemark:

Ahh OK, but she probably ought to have tracked the primary source - especially for something like ghost writer, and ghost, where the are multiple possible meanings.

Personally I would be amazed if there were not people performing the role of a 'ghostwriter' in renaissance England even though they didn't call it that. Which of course isn't expressing any sort of belief in Faktorovich's 'research'.

>118 Petroglyph: "For texts that do not have similar lengths, counting absolute frequencies and judging texts as "similar/not similar" based on those absolute frequencies leads to wrong results." As I argued in a previous post, short texts (under around 1,000 words) lead to less precise attribution results (with any testing method).

There is nothing unique about "absolute frequencies" that would make this measure uniquely useless when applied to short texts; it is as problematic or helpful as any of the other standard measures, as all units in a short text shrink at a similar rate (there are fewer punctuation marks and words overall).

I am now sure you do not understand what the terms "absolute frequency" and "relative frequency" mean, as what you are saying is nonsensical. All of the tests I apply that use frequency use relative frequency (such as the percentage of nouns, or commas per 100 sentences); it would be nonsensical if I recorded the absolute number of commas in a text when the text-length vary so widely. You argue I "should" use per-100 sentences measures, as if you have not read that I indeed use this measure (as I explain across BRRAM's Volumes 1-2). The website I use does not give the "option of normalizing punctuation per 1000 sentences"; it gives the option to create relative frequencies; refer back to the definition of "normalizing" I provided; the website in question refers to this as "per X" not "normalized" on a -1 to 1 scale. And whatever you are trying to say here is statistically nonsensical: "Normalization between 0.0 and 1.0 would take the form of 0.01 for 1%, and 83% would be 0.8." Just so you will understand the difference between "absolute" and "relative", here is a definition from Investopedia: "Absolute Frequency vs. Relative Frequency: Absolute frequency can be the starting point for a more nuanced statistical analysis. Relative frequency, for example, is derived from absolute frequency. When the absolute frequency of values is tracked over the entire trial, the absolute frequency for a particular value can then be divided by the total number of values for that variable throughout the trial to get the relative frequency. The relative frequency is what we most often reference, whether it is the winning percentage of our favorite sports team or the percentage of fund managers that beat the market. Unlike absolute frequency, relative frequency is usually expressed as a percentage or fraction rather than a whole number." https://www.investopedia.com/terms/a/absolute_frequency.asp

And all attribution methods judge "texts as 'similar/not similar'"; that's what attribution is determining which texts are similar to each other, and which are not similar or by different authors.

>130 faktorovich: The website I use does not give the "option ...

I will be interested in your reaction to >122 Petroglyph:

>122 Petroglyph: I immediately sent the following message to AnalyzeMyWriting:

---Dear Administrator:

I have developed a new computational-linguistics author-attribution method that includes using your website (and other websites) to measure certain elements of texts, such as passive voice. Here is the series where I mention your tools: https://anaphoraliterary.com/attribution/ And here is a free copy of the series for your review... I cite your website in Volumes 1-2. I believe using and advertising your tools to other potential users is a positive thing for you. Let me know if you object to me mentioning your website, or if you might want to collaborate on developing a platform that combines a number of your tests into a tab on your page to assist other attribution researchers. Thanks in advance for your time. Sincerely,

Anna Faktorovich, PhD, Director, Anaphora Literary Press---

I did not send it because I agree with your interpretation of their rules, but rather just because they and you invited an email, and I felt it was the polite thing to do. As for the rules, there is nothing in their policy that bans free advertising from users who write about their tools being helpful to attributors. You again do not understand the terms they are using and leaping to the assumption they are relevant. They are forbidding activities that "causes, or may cause, damage to the website" or to other people. Such as "data mining" or collecting data about private users to sell it to corporations who want to sell them stuff, or even to criminals who want to target them for fraud. It would be absurd if a website like Analyze did not allow the usage of their tools for the main purpose it is created for, or for calculating linguistic measures in texts. Another term they use in those terms: "Data scraping is a technique where a computer program extracts data from human-readable output coming from another program." This is the covert extraction of data that sneaks into somebody else's system illegally and steal things that are not publicly accessible. The tools they provide are public and open to all, so there is no sneaking, or hacking involved in utilizing the tools in the manner they are designed to be used. When they say the website cannot be used in a manner related to "marketing", they mean in a way related to the other activities on this list like "data mining" or it cannot be used to phish or to harvest and process personal data to sell data about thousands of people.

>132 faktorovich:

Another term they use in those terms: "Data scraping is a technique where a computer program extracts data from human-readable output coming from another program." This is the covert extraction of data that sneaks into somebody else's system illegally and steal things that are not publicly accessible.

The definition of data-scraping in quotes is correct.

However it is not 'covert extraction' it isn't about sneaking into somebody else's system illegally or stealing anything. In fact the whole point is your are collecting data in a form you can use elsewhere from a source that you can see as a member of public or your own logon. It is not automatically illegal in any jurisdiction (although some jurisdictions to have laws about scraping email addresses). It can be a contravention of a website's terms of use though.

When I was doing searches for terms like "ghost writer", "ghostwriter", "ghost writing", and "ghostwriting" in four newspaper databases to which I have subscriptions yesterday, I found some interesting false hits.

In some cases the nature of the search engine allows there to be words between the two I defined even though I used quotes. This becomes more like a

ghost near2 writer

search for the search engines that support that syntax that says something like "within two words". It is a little less helpful but the inaccurate nature of OCR and newspaper text layout means that this is sometimes necessary to catch most hits. They assume that a human will look at the result and see if it is relevant.

In some British newspapers of the mid-19thC had an often-reprinted ghost story where they would talk about the ghost and within a couple words mention "the writer" of the article/story. It clutters up the listings.

In the 1890s and first decade of the 20thC it is fairly common to see "ghost writing" (usually two words, sometimes hyphenated) to refer to collecting intelligence from the dead by entering a trance and writing what comes to the person with pen or pencil in hand. Spiritualism was popular in this era and there were some famous people who embraced it and others who worked hard to debunk it.

Decades later, after Albert Payson Terhune (December 21, 1872 – February 18, 1942) died, his second wife and widow produced a book that she claimed was a spiritual transmission of his thoughts. Today it is treated as some of his early writing with her notes to prepare it for publication.

In 1913 the president of the American League of baseball expressed concern about players who were going to be identified as writers of articles, particularly the key players of the teams participating in the World Series. This was, of course, ghostwriting in most or all cases. Some sports writer did the actual writing and the player was paid for the use of the name. Whether there were any interviews to collect the players' reflections is not noted. But the president of the league, Ban Johnson (almost like Ben Jonson), disliked the practice and tried to ban (sorry, an unintentional pun) the practice through the league rules and contracts.

Around this time, Christy Mathewson (August 12, 1880 – October 7, 1925) had published articles and some books between 1910 and 1919. A few names have been suggested for ghostwriting the various work, including two sports writers, Bozeman Bulger and John Nevill Wheeler, and one series book author for the brief series of baseball stories, Ralph Henry Barbour. The latter may be because Barbour wrote stories about several sports, including baseball, and had a football series for the same publisher at that time with similar titles. Yet, lists of Barbour's books in reference volumes which are generally reported by author do not include them. It is an area that may be a fun exploration of authorship attribution techniques once the texts can be gathered as well as some baseline examples of the writers. Barbour baselines are easy but the other writers may require more effort since they mainly wrote for newspapers. On one of these Christy Mathewson books had a special edition with a signature that did not match authenticated signatures by the player on contracts, letters, and baseballs.

But this all shows that it is not enough to look at a search results page. Each item has to be scrutinized. In some cases the date associated on Google Books or a newspaper database is wildly wrong and the date on the publication has to be found. For example, a periodical bound volume scanned by Google books may have a year that corresponds with the first year for the publication and not the first (or last) issue in the bound volume. The metadata in the newspaper databases can be wildly wrong as I discovered when looking for Disneyland articles. In general the usage should be no earlier than 1952 and mainly from 1954+ for the theme park. There were some other uses. But if you see something from the 1910s, or 19thC, then it is a case where the metadata is wrong. The date on the first page of the newspaper is generally reliable and should prevail unless there is one of the occasions when they made a typographical error (in the original sense of the term) in setting the banner for the paper for that issue. But usually this is not years off as the metadata problem reveals.

The close columns of newspaper and some magazine print can cause OCR systems to be confused to jump words across a column gutter and lead to confusing results.

It is an imperfect world.

James

>130 faktorovich:

When you quote me, you conveniently leave out this sentence: "And Faktorovich indeed makes use of relativized frequencies (though not always!)". And, later in >118 Petroglyph: "In her tables, Faktorovich uses these normalized results.".

So anything you wrote in >130 faktorovich: that rests on the assumption that I accuse you of using mainly absolute frequencies is false.

It's a very convenient omission. One that allows you to write a haughty, dismissive, imperious post that pretends you know what you're talking about and write as though others have not understood. It's one of your favourite kind of conceits. But if you have to resort to quote-mining to pull it off, you're probably wrong.

that would make this measure uniquely useless when applied to short texts

More reframing my post into something it did not say. "Varying lengths", I said. Such as the Mansfield, Emma, Northanger discussion.

I am now sure you do not understand what the terms "absolute frequency" and "relative frequency" mean, as what you are saying is nonsensical. All of the tests I apply that use frequency use relative frequency

I, again, refer you to >118 Petroglyph:, and to the top of this post where I point out the bits that you avoid quoting. The bits you avoid quoting make this assertion of yours wrong. The implication that I claim you use mainly absolute frequencies is also wrong. See the bits you avoided quoting.

You argue I "should" use per-100 sentences measures, as if you have not read that I indeed use this measure

In >118 Petroglyph:, I said "The website Analyzemywriting.com, one of Faktorovich's online services, presents the punctuation marks both in absolute frequencies, and terms of per 100 sentences: {image} In her tables, Faktorovich uses these normalized results."

If you are responding in detail to post >118 Petroglyph:, how is it possible that you have missed this?

Just so you will understand the difference between "absolute" and "relative", here is a definition from Investopedia

If you search page one of this thread for "absolute frequenc" and "relative frequenc" (to catch both "frequency" and "frequencies"), you'll find I have been arguing you should uses relative frequencies for your word counts. Relativized against your entire corpus. Not most frequent words/phrases in one particular text (absolute frequency). But most frequent in a text relative to a corpus. Did you only google those terms just now? If so, I can't say I'm surprised...

The fact you googled a finance and investment website for some utterly basic concepts in corpus linguistics is yet another indication of your complete out-of-your-depthness.

The website I use does not give the "option of normalizing punctuation per 1000 sentences"; it gives the option to create relative frequencies; refer back to the definition of "normalizing" I provided; the website in question refers to this as "per X" not "normalized" on a -1 to 1 scale

Corpus linguistics does not usually follow statistical standards and notational conventions from STEM fields, no. If you were familiar with the field you would know that. But normalizing "per 100,000" words, say, or per million words (as O'Keeffe et al do in that screenshot I posted in >118 Petroglyph:) is entirely unremarkable in corpus linguistics. Or normalizing the number of, say, commas per x number of sentences or per x number of words.

There are also other ways of normalizing things. There is no such thing as "one normalization to fit all." Expressing the number of commas in a text as a percentage or on a scale from 0.0 to 1.0 would be, well, useless? The proportion of all single characters? Comma's as 62% (or 0.62) -- of what? Total punctuation mark use? Expressing it in terms of 100 or 1000 sentences, though, that is a measure that tells you something about the average internal complexity of sentences. So that is a sensible normalization in this case.

But you've demonstrated numerous times during this thread that, once you form a "brisk impression" of how something works, you take that impression literally and use it as a black/white all/nothing right/wrong mental category. It becomes an easily-applied test with which to declare someone else's statements "nonsensical" and your own statements "correct", though only in the most superficial, literalist and unhelpful kind of way.

That's your way of doing things, though. Most other people don't think like that. Or act like that.

You know who really doesn't act or think like that? A scholar.

>134 Keeline: Out of the initial entries I found one stands out as the likely source that popularized the term in later appearances, "The Ghost Writer and His Story" by Graves Clark (1919), and others in this thread have mentioned other non-glitchy appearances as early as the 1880s. Those are some interesting stories you have dug up while exploring the appearances of this term; so I am glad I inspired you to look further in this direction.

>135 Petroglyph: I left it out because it was repetitive, but, "relativized frequencies (though not always!" is still false, as my frequencies are indeed "always" "relative".

I have not missed anything in your replies. Your comments on these topics, and your attempts at rebuttals have been deeply nonsensical, so there is no rational defense you can come up for them, and so you are relying on adding still more nonsense.

Looking up terms in a dictionary before discussing them in public is an essential part of being a professional linguist/ English-language-researcher. Anybody who does not check their terms before defending a position that requires a complete understanding of terms, is not a scholar who cares about the meaning they are conveying.

"Comma's as 62% (or 0.62) -- of what?" Commas can be counted as a percentage of all punctuation marks, but this is not the type of count that would be useful (as there are some rare punctuation marks or variants in punctuation that are likely to introduce glitches). My commas-per-100-sentences test has a range between 413 and 1075; this is not a percentage; it is a "relativized frequency" that is normalized during the binary conversion process into 1s/0s.

Whenever I detect nonsense in texts that are or pretend to be scholarship, I always call it by this term, as there is no better term to describe words that are used incorrectly and thus do not carry any sense. I have been reviewing books with this degree of honesty since 2009, so it is my instinctual position. It would be highly unusual for me to ever stop myself from pointing out a mistake of this sort, and especially when there are so much nonsense in a text that the whole thing is too nonsensical to allow anybody else to attempt to read it without a warning.

>137 faktorovich:: "Looking up terms in a dictionary before discussing them in public is an essential part of being a professional linguist/ English-language-researcher. Anybody who does not check their terms before defending a position that requires a complete understanding of terms, is not a scholar who cares about the meaning they are conveying."

The point being made in the post that you responded to was not criticism of the fact that you looked up a term, but rather 1) that you needed to look up the term, which should already be familiar to someone doing corpus-based linguistic analysis and 2) that the source you chose to back up your findings is from a completely different discipline.

A dictionary is a good starting place when one encounters terminology with which one is not familiar.

However, there are many words that have specific technical meanings in particular contexts. Different disciplines may have their own particular ways of using some of these terms that differ from common usage or the usage in other disciplines. Often these words convey concepts that require extra-linguistic knowledge--scientific or mathematical models and theories, for example. A dictionary cannot and is not intended to provide this sort of knowledge. In such cases, the confused reader would be better advised to consult a textbook or encyclopedia on the subject matter in question.

***This service announcement brought to you by someone whose work regularly requires her to read and understand texts on subjects she knows little about.***

>138 spiphany: Even with a filmographic memory, it would be impossible for any human to remember the precise definition of every word in all dictionaries (up to a million words in usage). Thus, any critic that never looks up terms, but instead relies on their intuition or vague recollection of terms is likely to misuse them, as Petroglyph has done in his misapplication of terms like "normalization". I and everybody else in this discussion are likely to be "familiar" with terms like "normalization", but linguistic debates are not solved on vague familiarity, but rather with precise knowledge of the rules of grammar, definition of terms, etc. Anybody can discuss a subject based on their "familiarity", but experts must discuss subjects based on precise knowledge, which is always found by checking sources, including checking dictionaries.

Statistics is a field that covers a wide array of disciplines. I first learned statistics during my economics degree classwork, so Investopedia is a familiar source that consistently provides useful definitions with mathematical elements that help to visualize the options of a mathematical problem. The statistics and the statistical terms do not change in meaning when they are applied to investment calculations or to computational-linguistics. To "normalize" data still means the same thing, especially since the process of normalization is all about taking whatever type of data units the raw data is in, and altering it to a numeric system where different types of measurements can be compared to each other. There are no classes I have heard of purely designed for computational-linguistics statistics, so most who want to be trained in this subject have to step outside of linguistics and computing and into fields like economics that offer textbooks, dictionaries and classes on statistics as the primary subject.

Non-quantitative philosophical theories, such as "extra-linguistic knowledge", are irrelevant when a scientist is working to arrive at precise quantitative results by applying the most relevant statistical tools.

I am particularly interested in the subject of investment statistics at the moment because next week I am doing a special review project for PLJ where I am reviewing textbooks (that publishers sent to me for review) about investment, and a few of them are purely quantitative, or describe the advanced mathematical/ statistical tools investors can use to make investment decisions.

>139 faktorovich:

Even with a filmographic memory

I have to admit that this expression is new to me. I have certainly heard of the more colloquial "photographic memory" that shows up in popular media and "eidetic memory". In attempting to find "filmographic memory" online I am seeing some of the other terms coming up but my browser spell check is not familiar with this word. I don't say it isn't used by someone but it is not a common expression, nor does it seem to be a technical term with a specific meaning. For the other two I found the usual Wikipedia entry (detailed) and this page.

https://www.betterhelp.com/advice/memory/difference-between-eidetic-memory-and-p...

Looking up words is something every scholar might (or should) do when there is uncertainty of the definition. Whether they cite the definition found is another matter which depends on the purpose and audience of the communication. However, I agree that when looking up a word, a generic dictionary often falls short, especially when one is looking for a scientific or mathematical meaning. Likewise, if one is looking for a usage from a past decade or century, modern dictionaries may emphasize current usage which can be misleading. We can all think of words that have been transformed in the past 50-75 years to have entirely different meanings.

Take the word "syndicate" which was used in the late-19thC and early 20thC as a group of people preparing content (articles or stories or pictures) for newspapers and magazines. Thus the Stratemeyer Syndicate who was involved in producing books for publishers mainly in juvenile series fit at the time it was established in 1905. But, with the rise of Organized Crime in the Prohibition era, the word "syndicate" became a "crime syndicate" and that became the dominant definition.

One of the employees of the Stratemeyer Syndicate described an interview that the head, Harriet Stratemeyer Adams, had with some vaguely-identified men in suits who wanted to know the nature of the business conducted by the Stratemeyer Syndicate. She explained, with as little detail that she cared to, that they wrote and produced these children's series books. At the end of the interview she asked them if they were from the F.B.I. They seemed a little surprised by this question, which was apparently close to the truth. When asked why she thought so, she replied, "well, I do write mysteries." The prevailing thought is that these investigators thought that the combination of a German name and the term "syndicate" might be something darker than what it really was.

My reaction to citing an investment dictionary/encyclopedia for a term used in statistics for linguistic computing was skeptical as it has been expressed by others. I know that words like "average" are often ill-defined in popular use but this and related terms (mean, median, standard deviation, etc.) have very specific meanings in statistics. The two-part question raised when using a source like this is — is it the best source for an authoritative answer; are there differences in the way the term is used in investments than it is in statistical analysis?

James

>140 Keeline: Indeed, I use the term "filmographic memory" because I mean something different from "photographic memory", as I recall things as if I am watching a film, though at times it can be more like looking at a photograph, or a series of photographs. It is indeed essential to look up such terms when they are even vaguely unfamiliar, so that we understand an author's intended meaning. Citing definitions has been essential in this thread because whenever I object without quoting a source such as a dictionary as the source for my objection; other writers in this thread simply say that my interpreted meaning is wrong; and the meaning can only proven to be correct by citing a dictionary meaning to avoid a debate about a basic term. Since I am currently translating a Dictionary-and-History via "Restitution", I am vividly aware of the shades of changes of meaning over time. But a term like "normalization" is not an ancient concept that is out of date, and so modern dictionaries, and especially those that specialize in statistics/ investment are suitable for this task. The problem is not that I used a dictionary to define these terms, but that Petroglyph misused these terms or used them in a nonsensical manner that is contrary to all definitions of them across all known dictionaries (specialized and non-specialized). Arguing that the problem is my source-checking or diligence with finding dictionary meanings is similar to arguing that the problem is in whoever outed the abortion opinion, and not the fact that turning abortion into an illegal act is problematic.

I see a relatively small number of pages thad use "filmographic memory". It is clearly not a common term used by others and it was not defined by you when you used it. Is this the definition you mean as covered on these pages?

https://www.google.com/search?q=%22filmographic+memory%22

James

>139 faktorovich: Non-quantitative philosophical theories, such as "extra-linguistic knowledge", are irrelevant when a scientist is working to arrive at precise quantitative results by applying the most relevant statistical tools.

For one, you have told us that "Holy Ghost" refers to the fact that they were ghostwriters, which is extra-linguistic knowledge. For another, we can go back to near the start of last thread for the discussion about the most relevant statistical tools where yours were referred to as "amateur hour". There's a lot of choices of statistical tools, but random websites with hidden code is one of the least trustworthy.

>139 faktorovich: Re: dictionary definitions of "normalization".

You can't have looked very hard or very carefully. Possibly just enough to make yourself feel in the right and vindicated and whatnot.

From Collins Cobuild English Dictionary:

Normalize
in British English

{...}
2. to bring into conformity with a standard

{...}

in American English

1. to make normal; specif.,
{...}
b. to bring into conformity with a standard, pattern, model, etc.

Derived forms: normalization: noun.

From Merriam-Webster:

Normalize
2. mathematics: to make (something) normal (as by a transformation of variables)

The page at the link behind normal has this as the relevant definition:

3 a: approximating the statistical average or norm

Or even #6:

6: of a subgroup: having the property that every coset produced by operating on the left by a given element is equal to the coset produced by operating on the right by the same element

Therefore, as per Merriam-Webster, dividing the number of commas in both Mansfield and in Northanger by (# sentences / 100) entails normalizing them to per 100 sentences.

Quod Erat Dictionariandum?

For this Lunch Break Experiment (tm) and Petroglyph's After-Dinner Stylomety corner, I wanted to take a look at lexical density:

As analyzemywriting.com defines it: "lexical density is simply the sum of the percentages of nouns, adjectives, verbs, and adverbs as stated in the definition of lexical density." In other words: count the proportion of all the content words as a percentage of the entire text, and don't count the function words (such as auxiliaries, determiners, pronouns, prepositions). It is a measure of how many content-words there are in a particular text.

Faktorovich's "newly-invented" "27-tests method" spends 8 of its tests on lexical density, broadly speaking: there's one for the lexical density of the entire text, and then a one test each for nouns, adjectives, verbs, adverbs, prepositions, pronouns, and auxiliaries.

Let's see if she can get reliable results for these tests.

---

1. What about genre?

Faktorovich happily compares genres (sermons, letters, poetry, plays, prose, philosophical works, etc, etc.). Plays like Hamlet (~32,000 words, ~3900 sentences) are blithely compared to individual poems like Thomas Edwards L'Envoy to Narcissus (511 words; 15 sentences.)

It is well-known that lexical density is a feature that changes with genre. Plays, which are virtually entirely dialogue, are likeley to contain a much higher number of pronouns than, say, philosophical works. Even analyzemywriting.com knows this! They even performed this brief and informal comparison between some random wikipedia entries and some short stories. Their main finding is that "Lexical density, as measured by this website, tends to be higher in informative writing than in fiction on the order of about 7%."

This is, in fact, one of the things you can use lexical density for: differentiating between genres, and between different kinds of texts, and between different styles. If the same person writes a play, and also a theological tractate or something, the fact that those texts will have different lexical densities does not justify the conclusion "must be two different people who wrote this". It's just the concept of "genre". Individual differences in lexical density may play a role, too, obviously. But that's something you have to consider in conjunction with things like register, age, genre, level of education, familiarity with genre, creativity, etc. "Different density means different people" is not a conclusion you can jump to, or, as Faktorovich does, merely assume it's true.

Or, as McArthur et al. (2018) note: "Different registers will often differ in terms of lexical density; generally academic texts have a greater lexical density to informal spoken texts" (q.v. Lexical Density).

Let's look at what Faktorovich makes of this concept. From Re-attribution p. 96 (see reference in >17 Petroglyph:):

The 6th test for Lexical Density had one of the most pronounced clusters. This is a measure of an author’s tendency to be verbose or to compress their thoughts into challenging language (indicating greater lexical density), or to write lightly and with brief strokes (registering as low lexical density). (p. 96)

So no. Lexical density is a property of texts: chatty, spoken language tends to include more sentences like "I gave it to her" (where the lexical words have been underlined, for a lexical density of 1/5 = 20%); and a more technical, written text might put that as "The person was given the book" (for a lexical density of 3/6 = 50%).

Lexical density is not a feature of a person. It is a feature of a text. This is a very important point, and Faktorovich has mistaken the latter for the former.

She continues:

While one might assume that Verstegan’s sermons and speeches were particularly dense linguistically to bombastically express propagandistic positions, they actually predominantly cluster on the lower end of this scale. 9 of Verstegan’s texts took the 9 lowest positions on this range. The lowest among these was Verstegan’s translation of King James Bible that has a lexical density of only 43.69, probably because it was intended to be simple enough to be grasped even by the semi-literate masses. “Elizabeth I’s” speeches follow, then the Marprelate Tracts, and then other rhetorical texts. This pattern is only interrupted by the similarly lexically light Byrd-ghostwritten Addition III of Sir Thomas More. Then, 5 more of Verstegan’s texts follow, including “James I’s” Demonology. The upper-end of the lexical density spectrum is dominated by Harvey and Sylvester. Harvey’s self-attributed Pierces Supererogation has the highest density among the 284 texts at 64.02; Harvey appears to have chosen this piece to represent his own style in 1593 to defend his scholarly credentials around two years after his fellowship at Cambridge ended, and while he might have still been hoping to rejoin academia." (p. 96)

So. Faktorovich assumes that sermons and speeches (i.e. intended to be delivered orally in a largely illiterate society) would have high lexical densities, which is one thing she got arse-backwards; and she associates different lexical densities with different authors (with most of "Verstegan's" texts having low densities). This, too, is arse-backwards.

In other words: Faktorovich has misunderstood what lexical density is, what it measures, the kind of thing it is used to measure, and what kind of genres can be expected to have low lexical densities.

But that's par for the course.

This, in and of itself, makes whatever she makes of the results of her tests unreliable.

---

What about Early Modern spelling?

Under the section "Assumptions and Limitations", analyzemywriting.com says that "We first note that our calculation of lexical density assumes that a text is written in English. Furthermore, it is assumed that a text is properly punctuated and apostrophes are used correctly."

If this free, online service is calibrated for present-day spelling conventions, that means that it stands to reason that Early Modern spelling conventions, or the lack thereof, may have an impact on the results that this black box spits out. Enough of an impact to validate the results, perhaps?

Let's send up a few trial balloons and see if renaissance spellings can influence lexical density counts.

To begin with, I tried this query:

loved lovèd lov'd

and got this as the result:



So: loved is analyzed as a verb (correctly!); lovèd is analyzed as a noun (incorrectly!); and lov'd is analyzed as an auxiliary (incorrectly! -- Analogous to I'd, you'd, she'd, ...). Yikes! A verb being re-assigned to a noun (merely because of un-modern spelling!) impacts both the percentages for nouns and for verbs, but it does not impact the calculations for lexical density. But a mis-tagging of a lexical verb as an auxiliary does.

So yeah. That little test right there is reason to doubt the results that analyzemywriting.com produces for Early Modern spellings.

What about thou thee thy thine? These four words, pronouns all, should yield a lexical density of 0%.



Only two of these are parsed correctly as pronouns. thy and thine are tagged wrongly as "adjectives". Having been analyzed as lexical words instead of function words, they drive up the lexical density.

Let's continue.

I took this sermon (written in 1604; printed in 1623), cleaned out the page numbers and some of the formatting artefacts. I took one paragraph, the one on pp. 190-194 starting with "in conclusion", and re-spelled it into modern English. (See here for a .txt file). I then fed both versions into analyzemywriting.com, just to see if there were differences.

Oh yes, there were. There were problematic problems, even.

First, the overall figures for lexical density of these paragraphs were different. The Renaissance original got a score of 51.3; the
Re-spelled version got 46.24. That's 5% difference.

Where do those differences come from?

Well, turns out that there were things I thought could be problems when I was respelling the text, and that actually are: to teach vs, that wee must not "vs" is analyzed as a noun, and not as the pronoun "us"; "wee" is analyzed as an adjective (as in "a wee lass") and not as the pronoun "we"; though it bee neuer so small "bee" is analyzed as the noun and not the copula "be".

(I must add that there were things I thought could be problems but that turned out not to be: heauen, vnderstand, keepe, needefull, feare: these things are indeed analyzed as lexical items. The plural pronoun ye also is correctly identified as a pronoun. Though I did not check whether they're analyzed as the correct subtype of lexical word).

Here are two screenshots. On the left is a set of 10 sentences from the re-spelled version; on the right is the original. The words that analyzemywriting.com judges to be lexical (i.e. noun, verb, adjective or adverb) are in green; the percentage of lexical density for each sentence is to the right. I've underlined in orange a few differences. (Right-click to embiggen)



The lexical density of nearly every sentence is different across the two spellings.

The culprits are the un-modern spellings of doe "do", haue "have" and bee "be" (and also shal "shall"). Now, these are all a) high-frequency words, and each occurrence adds up over time to (likely) significant differences in the counts; and b) they should be counted as function words, but instead they are coloured green and, therefore, drive up the lexical density. Sometimes by as much as 15-20%! (as is the case in sentences 22 and 20)

Here are two more screenshots, this time showing what proportion of a text is taken up by each part of speech that analyzemywriting.com recognizes. The first image shows the percentages for the re-spelled version, the second image for the Early Modern paragaph.



Every single one of these is different. Nouns differ by as much as 8%; and the count for auxiliaries doubles. And they're the same text! These texts contain the exact same number of nouns, and auxiliaries, and pronouns, etc.

If Faktorovich were to import both versions of this text into her corpus, the only chance of these two identical texts ever being marked similar in her kooky and wrongheaded red-font-colour way of testing is if they happened to be adjacent to each other when sorted low-to-high in every single of the 8 columns. Adjacent in the sense of ~9% of the corpus size. Despite them being the exact same text -- just in different spellings.

That's a big ask. I don't think I can trust that to be the case.

Conclusion: This free online service is explicitly designed for present-day spelling conventions, as the makers admit themselves. Applying the software to Early Modern spellings yields unreliable results.

---

But it gets worse

So. Does Faktorovich use modern spelling, original spelling, or both (which would be the worst of both worlds)?

She uses both: "90 of the tested versions of the texts had modernized spelling, while the other 194 had their original or only slightly edited Early Modern English spelling" (Re-attribution, p. 29)

Insert eye-rolling emoji here. Early Modern spelling, modern spelling and an inbetweeny version. Immediately after this sentence she mentions that versions of Shakespeare with modernized spellings and punctuations were registering as different from Early Modern punctuations, so she switched to original spellings for these texts. But not, apparently, the 90 other modernized texts!

In characteristic Faktorovich fashion, she ignores the issue and goes on to use tests that she herself has found to be unreliable: "While it is best to test texts with the same level of modernization; many of the groups {of texts she attributes to the same ghost-writer} included both old and new-spelling texts, so modernization in many cases leaves enough of a signature for a text to be identifiable" (Re-attribution p. 30).

In other words: IT wORKs sOME of THE tIme So Let'S aSSume IT wORKs alL Of The TIme. I also note that she omits a very important part here: did the similarities between the original and the modern spelling come from the lexical densities tests? Or were other tests responsible for this similarity? That would, in fact, be important to know in deciding whether or not wE CaN aSSume IT wORKs alL Of The TIme.

---

Conclusion: What does all this mean for her tests?

Well, 8 of her tests concern lexical density and the percentages of various parts of speech. As I have shown here, and as the makers of the site pretty much explicitly note: analyzemywriting.com is not the appropriate tool to use for versions of English that deviate too much from standardized spelling conventions.

I should note here that 5 of her tests are for punctuation marks -- which are often altered by printers, later editors and by the digitization team. This is a problem she seems to be aware of (and it's been brought up numerous times in this thread by others). So, properly speaking, without a systematic treatment of punctuation (either all-original, or all-modern), these five tests should be included here, as well, as "tests that are unreliable because the tests measure a spelling-dependent feature".

But that is a post for another day.

---

The final facepalm (for now)

When two versions of the same “Funeral Elegy by W. S.” were compared in old (4,450) and new (4,459) spelling varieties, they matched each other on 16 out of 27 tests. Since two nearly-identical texts should match each other on all 27 tests, it follows that texts matching each other on 10 or more tests despite being from separate old/new categories, can be assumed to actually have far greater similarity if they had been compared without this spelling divergences. In fact, a 10-test match is sufficient in this corpus to establish a match because digitization-generated spelling irregularities and various other glitches have introduced linguistic inconsistencies into most of these 284 texts." (Re-attribution p. 30)

Near-identical texts only match as just under 60%, as measured by her method. From this, Faktorovich does not conclude that she must be doing something wrong. Instead: YAY i Can DEpLoY EveN loWEr sTAnDarDs.

---

References
McArthur, Tom, Jacqueline Lam-McArthur, and Lise Fontaine, eds. 2018. The Oxford Companion to English Language. 2nd Edition. Oxford University Press.

>142 Keeline: I have never seen the term "filmographic memory" used by anybody other than myself before. I started using it at least a decade ago to describing remembering events as recorded films, in addition to recalling pages from books photographically. For example, I was a town hall meeting where oil executives surprised us by announcing they were going to drill new holes in Quanah. I asked several questions during and after the meeting. I did not have a pen and paper or a laptop in my hands, so it did not look like I was recording the conversation, so they were uniquely honest. But then when I got home, I remembered every word that was said by all sides across this 2-hour or so interaction, transcribed every detail and published an article about it. Here is the resulting article: https://www.prforpeople.com/news/%E2%80%9Cit-ain%E2%80%99t-no-california%E2%80%9...

>143 prosfilaes: You are misinterpreting what I am saying by imagining what I said instead of quoting what I actually said again; go back and read the full post 414. Every appearance of the term "Holy Ghost" from the Renaissance absolutely does not indicate a confession they were ghostwriters. The specific instances I cited subvert this term to refer to the authorship of the theological text by the "Holy Ghost", and do not merely describe the "Holy Ghost" without putting the credit for authorship into this "Ghost's" hand. The credit line that a text was written by "Nashe's Ghost" is not at all convoluted, but is a direct satire about "writers" who do not really do their own writing (perhaps because they are dead, or perhaps because they are illiterate) and instead have a "Ghost" (i.e. ghostwriter) write their book(s) for them. If you think there is something metaphysical about this, you have missed the joke. The joke is that a Ghost cannot write, and that a professional writer has to step in to write the text that is credited to a posthumous byline of an author such as "Philip Sidney" or "Marlowe", who did not have any book credits to their name during their lifetimes, and then seemingly start writing for the first time post-death as "Ghosts". No, the Workshop was not saying the ghosts were actually writing. It is amazing that you guys keep repeating this as if you do not understand it. Can you clarify what you are confused by? It is not "extra-linguistic", but basic linguistics to grasp that the Workshop was repeatedly confessing to their own authorship by joking about the possibility of Ghosts (instead of them) doing the work. There are several statements regarding ghost-authorship made in the front-matter poems to "Fletcher" and "Beaumont's" 1647 Folio that was one of the last Workshop projects released while the last one of them (Percy) was still alive. There is a chapter in Volumes 1-2 about these references in the Folio that explores their confessional implications. The statistical tools I use are the exact right tools to establish an attribution conclusion.

>145 Petroglyph:

To begin with, I tried this query:

loved lovèd lov'd

I have a few questions and comments about the baseline tests described.

Is lovèd with the grave accent what is normally found in vintage printed and/or modern texts under consideration?

I've seen the first and last but don't recall the middle one.

Were these three words the only ones in your corpus?

I ask this because there are a couple ways for software to identify the parts of speech (POS). One of them is a word list but as we know, a given word can have different meaning based on the sentence structure and other context hints (e.g. "made of lead" as a noun vs. "lead a group" as a verb).

From the descriptions, https://AnalyzeMyWriting.com, is a "black box" experiment where you input a text, run some tests, and get some report. But we only know what they say the software does inside.

Since it is a free service, they probably have not spent hundreds of hours to develop new and innovative code to analyze a text and come up with new measurements that have never been done before. Instead, it is more likely that they have collected open-source tools (not identified?) and made a friendly web user interface to implement them. The terms of service rules are designed to prevent network and processor abuse at the Denial of Service (DoS) level.

Supposing they did use some open-source textual analysis tools, the source code for them is available and they are documented. But not knowing which ones means that we have to play a guessing game like the "black box" experiment of a physics class.

I agree with the concern expressed:

If this free, online service is calibrated for present-day spelling conventions, that means that it stands to reason that Early Modern spelling conventions, or the lack thereof, may have an impact on the results that this black box spits out. Enough of an impact to validate the results, perhaps?

The word lists, POS identifications, and sentence structure of the tools used are almost certainly optimized for modern English writing. Other languages and other time periods of English are likely to produce weaker results. It will recognize some words and structures but fail to identify others. This noise in the results could be the difference in identification, especially if the later analysis method tends to further mask details.

As an illustration, a colleague of mine applied some reading scores to certain series books that had been digitized. The reading scores were provided by Microsoft Word, I believe, and were named. They gave grade levels based on the documented methodology of each test. So even though Microsoft Word is notoriously the opposite of an open-source application, the tests most likely follow the reading store methodology. As it turned out, these reading stores had some interest but were not a means of distinguishing one ghostwriter from another. There were other factors involved, including the outlining and editorial influences of the Stratemeyer Syndicate and the publisher editors. To complicate things, these people and standards changed over time. The books published today, well after the Syndicate ceased to be a separate entity, are directed to much younger readers than were the originals.

My assessment of this situation is that you can use available tests but not all of them will be relevant. Let's say you have a chemistry set and you are going "off book" in your experiments and combining chemicals as they appeal to you. You could use up your supply of red and blue litmus paper on the results and not learn anything if the reaction (if any) did not result in a change of the pH (base-acid scale) of what is left in the test tube.

Over the years I have seen many individual and suites of textual analysis tools that can be deployed on a server. They can be used at the command line or in conjunction with software to produce dynamic web pages such as Perl, PHP, or Python. Here is a write up on one such tool

https://aicbt.com/authorship-attribution/

Remember that the main motivation for such software is to identify authors of modern texts. This could be for an ethical concern like plagiarism or for some criminal investigation such as a terrorist manifesto. It can also apply to anonymous or pseudonymous text questions such as the insider book Primary Colors or the cases where a famous author tries to secretly release a different kind of book than for which they are famous under a pseudonym. There is more of a financial motivation (along with other sources of motivation) to build tools for this kind of analysis than there is to answer authorship questions of the distant past.

The emphasis on punctuation, which I argue is more relevant in modern texts where the author's punctuation is typed on paper or into a computer and likely to be left standing with less intervention than 400-year-old texts which were typeset from manuscripts (perhaps with little or no punctuation at all) and edited many times, is indicative of the tools above and possibly inspired by analysis like the one described here which has been rebroadcast in many blogs.

https://medium.com/@neuroecology/punctuation-in-novels-8f316d542ec4

In this, the words are removed, leaving only a handful of punctuation marks (; : ' " , ! ? .) to be counted or displayed as walls of punctuation to give a sense of the patterns. It is interesting and relevant only so far as the number of hands manipulating the text does not become so large that we can't tell who contributed what from it. This page (a chapter of a larger work) may be of interest:

https://sites.ualberta.ca/~sreimer/ms-course/course/punc.htm

which contains this paragraph which relates to this discussion (emphasis added):

Generally, manuscripts tend to be more lightly and less consistently pointed than printed books (and with the exception of the punctus, virgule, and the blank space, almost all of our modern marks of punctuation have come into use only since the thirteenth century). Modern punctuation, designed to clarify syntactic structures rather than to indicate breathings, is largely a Renaissance invention, developing during the first generations of the printing press, and codified in the eighteenth century (about the same time that capitalization and spelling became fixed in more or less their current form). Among the earliest works showing "modern" punctuation is Francis Bacon's Essays. An interesting early discussion of the nature of modern punctuation can be found in Ben Jonson's English Grammar (composed ca. 1617, printed posthumously in 1640). Eighteenth- and nineteenth-century punctuation practice varies considerably, but tends to be "heavy"; current "light" punctuation is largely the invention of H. G. and F. G. Fowler, The King's English.

There are many textual analysis tools out there and some have more interest than others. Here's one page with links to projects:

https://digitalresearchtools.pbworks.com/w/page/17801708/Text%20Analysis%20Tools

This one is also of interest:

https://corpus-analysis.com/

Note that most of these make individual measurement, counting one type of thing in an absolute form (how many X in the text) or relative form (how many X per Y words/sentences).

James

>146 faktorovich:

I have never seen the term "filmographic memory" used by anybody other than myself before.

If you believe this, then that is the time to define it rather than make us guess or think you have confused it with the more conventional terms I mentioned.

James

>144 Petroglyph: Your argument is that I should not use financial dictionaries to establish the meaning of a mathematical/ statistic term, and instead you are using general dictionaries like Collins and Webster? This is clearly a case of you arguing for the sake of arguing without any weight or significance to the counter-argument that you are making that is lacking in any rational meaning.

"Normalization" is "to bring into conformity with a standard"? This is a non-mathematic philosophical definition that is irrelevant when one is discussing computational-linguistic methods for calculating authorship attribution in the most non-biased and the most accurate manner possible. And if you apply this definition, "approximating the statistical average or norm", you are going to change all of your data from the actual data into averages, which would be the maximum possible level of none-sense, as your attributions would all point to a single average author. As for the last attempt at the definition, it is absolutely irrational to divide all 27 tests I used by 100 sentences. This division makes sense for tests like commas (because there needs to be some standard "X out of Y" calculation to account for a text being a wide range of possible words in total), but it does not make sense for characters-per-word (because dividing this measure by 100 sentences only does a second breakdown when you have already calculated the average characters-per-word for the whole text). This is why what you are nonsensically insisting is wrong with my method, are the exact right things that lead my method to return accurate attribution results.

>145 Petroglyph: Interesting – thanks!

Just checking, though: you say

to teach vs, that wee must not "vs" is analyzed as a noun, and not as the pronoun "us"; "wee" is analyzed as an adjective (as in "a wee lass") and not as the pronoun "we"; though it bee neuer so small "bee" is analyzed as the noun and not the copula "be".

– do you deduce their analysis from the POS percentages? Because I find no list of tokens classified as nouns, tokens classified as adjectives, tokens classified as pronouns, etc, just the pretty coloured bars and percentages. And that's more than a little problematic if it's going to be used as the basis of an analysis of the proportions of each part of speech!

I carried out a slightly similar exercise: I ran four versions of the "Once more unto the breach" speech from Henry V through the lexical density* analyser at analyzemywriting.com.
1. A First Folio text found at https://internetshakespeare.uvic.ca/doc/H5_F1/complete/index.html
2. Gutenberg.org's version at https://www.gutenberg.org/cache/epub/100/pg100.txt
3. A version of uncertain provenance at https://poets.org/poem/henry-v-act-iii-scene-i-once-more-unto-breach-dear-friend...
4. Another First Folio version, with modernised (American English!) spelling at https://internetshakespeare.uvic.ca/doc/H5_FM/complete/index.html

All four texts had the same number of words (273), but they differed quite a lot in other respects, with 34.43% nouns, 4.76% adjectives, and 4.76% auxiliary verbs in Text 1, and 29.3% nouns, 6.23% adjectives, and 2.93% aux in Text 4. For instance. Texts 2 and 3 were the most similar, both of them having 28.98% nouns, 6.23% adjectives, 6.23% adverbs, 15.38% prepositions, and 5.49% pronouns. The difference between those two versions lies in the verb / auxiliary proportion, specifically hard-favour'd in Text 2 which is hard-favoured in Text 3. (Both texts have several other identical 'd verbs, call'd, swill'd, etc).

But this points to another problem, a big one regardless of the age of the text – analyzemytext doesn't tokenize words! That is, "you'd" is counted as one single function word with one single part of speech, even though it is actually two parts of speech ("you" + "'d", pronoun + auxiliary). To illustrate this, I ran a lexical density analysis on the string
"you would you'd call'd". This was parsed as four words with the following POS distribution:
Nouns -25%**
Adjectives 0%
Verbs 25%
Adverbs 0%
Prepositions 0%
Pronouns 25%
Auxiliary Verbs 50%

Checking the words one by one, I verified that "you" was classed as a pronoun, "would" as a main verb, and you'd and call'd as auxiliary verbs.

So, yeah. No. I was well aware of the fact that analyzemytext.com was not built for research, but this is a pretty big flaw which makes the word count and lexical density analysis unreliable even for a modern-day text.

I was going to say something about the CLAWS POS tagger, too, but I think I've been verbose enough here.

-Linnéa

* or, as I have occasionally called it to the amusement of my students, lexical dentistry
** no idea what that's all about

(edited to fix own brainfart – I do know what an auxiliary verb is.)

>148 Keeline:
>145 Petroglyph:
You might be interested in the following paper:
Rayson, P., Archer, D. E., Baron, A., Culpeper, J., & Smith, N. (2007). Tagging the Bard: Evaluating the accuracy of a modern POS tagger on Early Modern English corpora. In Proceedings of the Corpus Linguistics conference: CL2007. Available here: https://eprints.lancs.ac.uk/id/eprint/13011/

It's a small pilot study, but it highlights some of the issues involved here. The TL;DR version is that with a tagger (CLAWS) that is consistently 96%-97% reliable when POS tagging Modern English texts, the reliability for the Shakespeare First Folio was about 82%, and for another EModEng corpus with texts related to law and the sciences it was 88.5%. Their spelling variant detection tool VARD improved the reliability a bit, though. The VARD tool is available at https://www.clarin.ac.uk/vard, and the CLAWs tagger (which is rather more useful than analyzemywriting to a researcher) is at https://www.clarin.ac.uk/claws.

-Linnéa

>145 Petroglyph: "It is well-known that lexical density is a feature that changes with genre. Plays, which are virtually entirely dialogue, are likely (I had to correct this typo from "likeley") to contain a much higher number of pronouns than, say, philosophical works." I already tested the "lexical density" measure output on 284 texts just in the Renaissance corpus, so there is no need to go beyond it to answer the question of how genre impacts it. On the lower-end of this spectrum are mostly Verstegan's texts, including the KJ Bible, but there are also speeches, theological satire pamphlets (Marprelate), and Vere's letters. Byrd's section of poetry-drama, "Addition III" out of "Shakespeare's" "Sir Thomas More" is also on this low end of the spectrum. And Verstegan's rare play "Wilkins'" "Pericles Prince of Tyre" is also on this lower end of the lowest 19 rates out of 284. On the other top extreme of this range are mostly Sylvester and Harvey, with the top spots going to Harvey's poetry, such as "Rowlands'" "Betraying of Christ". There are also plays on this top end such as Sylvester's "Lodge's" "Wounds of Civil War". And a bit less high but still on the higher end is Verstegan's non-fiction rhetorical text, "Meres'" "Wits Common Wealth, Part 2", which ranks 240 out of 284 on this "lexical density" test. Intuitively one might have guessed that rhetorical non-fiction would have the higher "lexical density" scores, but the numbers indicate the opposite to be the case. There are different genres both on the high end and on the low end. There is more of a pattern with Verstegan and Jonson being on the lower end and Harvey and Sylvester being on the higher end of this spectrum than between specific genres being at the top or bottom.

"Lexical density, as measured by this website, tends to be higher in informative writing than in fiction on the order of about 7%." This points to a major problem of assuming one knows what the results are going to be before starting an attribution experiment. While the specific texts they chose to test between non-fiction websites, and fiction might have pointed to a higher lexical density in non-fiction; there are many writers that write extremely linguistically complex fiction, and many who write very light or linguistically simply non-fiction. An attribution method is designed to spot linguistic differences between texts without giving any weight to the genre. This is why my combination of tests works, as it (as I have explained) shows linguistic patterns that override genre-lines.

A single test indeed does not lead to a precise attribution. So I do not claim anywhere in the study that merely, "Different density means different people". By putting this statement in quotes you suggest that I wrote it, when in fact this statement does not appear anywhere in BRRAM. Because any single test can produce an attribution glitch for an unpredictable reason, I use a combination of 27 different tests. In contrast, the standard accept Stylo method uses a single test only for word-frequency, and this indeed creates glitches, and assumes erroneously that: "Different frequency of word-usage means different people".

The test measures: Lexical words/ All words. Lexical words: "nouns, adjectives, verbs, and adverbs." Non-lexical words: articles, prepositions, conjunctions, etc. Here is what you state:

---"I *gave* it to her" (where the lexical words have been underlined, for a lexical density of 1/5 = 20%); and a more technical, written text might put that as "The *person* was *given* the *book*" (for a lexical density of 3/6 = 50%).---

I do not believe Analyze is calculating these measures correctly, and if I had designed this test, I would have done it differently. One problem is that "pronouns" can be used instead of "nouns", so they should be counted as "lexical words". The definitions for "lexical"/"non" do not specify that pronouns are "non-lexical", but this is indeed how the system calculates them. And "was" is technically a verb, so even if it is a "linking verb", it should not be counted as a non-verb if the definition states all verbs are "lexical". While I would not have designed these parameters for this test, whatever the rules are, when they internally consistent (and relevant to author-attribution) and they are applied consistently to a set of texts generate results that distinguish between linguistically different signatures.

My statement in the book about this measure holds as true. "This is a measure of an author’s tendency to be verbose or to compress their thoughts into challenging language (indicating greater lexical density), or to write lightly and with brief strokes (registering as low lexical density)." The second example has more "verbose" or more complex words. Nouns and adjectives are more verbose than articles or other "non-lexical" words; a sentence with 10 words with 50% of them as adjectives and 30% as nouns is more compressed in meaning, than one that has 80% of prepositions. An article is technically a "brief stroke" as it is shorter than longer nouns, verbs etc. (especially if the shorter pronouns and "linking verbs" are not counted as lexical).

I address the question of Modern vs. Early Modern spelling in an extensive section in the "A New Computational-Linguistics Authorial-Attribution Method Described and Applied to the British Renaissance" chapter of Volumes 1-2. And yet now that it would be relevant, Petroglyph is not quoting what I explained about spelling differences, and is instead paraphrasing what I said why not crediting me for having said it, suggesting I ignored it and he is saying these points himself. In my explanation of how I prepared texts for testing, I explain that I altered all unreadable for programs characters such as the è in "lovèd" into an e automatically, by applying the auto replace feature across each manuscript. So this typo in the original transcribed text, would not have negatively impacted my results. I also explain that there are indeed major changes between the Old and Modernized spelling texts, such as those that include 'lov'd" vs. those that do not have such contractions. This is why previous computational-linguists have not been able to distinguish "Shakespeare"-bylined texts into the two distinct Percy and Jonson signatures, as they have been texting Modernized versions of these texts that are very different from the original versions as editors have made an enormous impact on their linguistics. To fix this problem, I re-tested all of the "Shakespeare" texts in their original-spelling versions, and this made the test sensitive enough for the authorial signatures to be distinguished. The presence of a similar percentage of unreadable words such as "lov'd" in all texts for each of the signatures, just means these have a similar white-noise impact on the results, as they do not change the attribution outcome, but present some haziness. This haziness is one of the reasons the Renaissance texts have more glitches or mismatches or smaller percentage matches between texts than texts in the 18th-20th centuries that do not have some glitches. Despite this problem, it is far less likely glitches will impact the decision when there are also tests for punctuation and other non-word-frequency related items; in contrast, most other computational-linguists only test word-frequency, so such glitches occur at the same frequency for them, but they have a far greater impact on skewing the attribution decision towards an error.

I explain the special case of "Funeral Elegy" across Volumes 1-2, and not only in this passage. The other observations explain that it has been re-attributed to nearly a different other bylines by different linguists across the history of its scholarship. The main reason it only matched on 16 tests to the other version of itself is because one was modernized spelling and the other was original-spelling. As I explain across the series, Modern-spelling versions of "Shakespeare" etc. do not properly match the Old-spelling versions because the editors introduce their own style. By testing both of these versions for many different texts, I proved the editor-bias or editor-skewing to be a fact, and especially in the case of this "Funeral Elegy". Linguists who have been comparing the modern or old spelling versions might have been coming up with different attributions simply because they were only considering one spelling style. And I have seen almost no computational-linguistic studies that disclose if they are testing original-spelling or modernized-spelling versions of texts, or review how impactful this difference can be (as I have done across BRRAM).

Huh. There seems to be something pretty odd about how analyzemywriting handles verbs. Could someone else do a lexical density analysis at analyzemywriting.com of the sentences "You did it. You will do it. I said that you would do it." and tell me what your result is, and if it makes sense?

>148 Keeline: It is irrational to reject a tool that accurately performs a basic test in a manner that is stated in its manual, simply because it is free and you are assuming this means the developers have not invested time to polish the provided tool. In practice, Analyze's tools have worked to distinguish between linguistic styles in the corpuses I have applied them to. Based on Petroglyph's test, the system only makes mistakes when it is applied to corrupted letters such as an e with a stress-mark above it. Otherwise, it consistently groups verbs etc. in the right categories. It is unnecessary for it to recognize Early Modern English spelling variants. And there are no programs I have come across that are capable of sifting linguistic statistics for the many variants that come up in these Renaissance texts.

In my experience, EEBO and other digitizing services preserve most of the original punctuation to make punctuation tests very accurate even for Renaissance texts. They introduce some glitch punctuation such as strange large dots in the middle of lines (mostly when a letter cannot be read by the system, as when there is a line above a letter to indicate it is a contraction, with an n/m that is missing but should be after this letter); these dots can be and were removed automatically by me during the text-cleaning process before testing. Pre-Renaissance punctuation oddities are not relevant to my tests of Renaissance texts that use generally similar punctuation rules to those we currently use.

>149 Keeline: I assumed you all know what "filmographic" and "memory" mean separately, or that you could have looked up these terms, and then derived what the two of them when used together would have meant. It would indeed be amazing if Petroglyph defined each one of his terms every time he used them to make sure we all knew exactly what he meant by "normalized" etc. He cannot even give a single coherent definition that makes this term make sense when asked about it, and instead gives a rainbow of possible meanings to add nonsense on top of nonsense, instead of acknowledging he is using nonsense terms to make it sound like he has some cryptic expert knowledge of terminology that proves my method wrong; and nobody can understand how my method is wrong without understanding his special made-up meanings for these terms. If you misunderstand what I mean by "filmographic memory", you are simply confused about the type of memory I have; whereas, if you believe Petroglyph has reached some deep understanding that proves my method wrong in the miss-used terms he is applying to it; then, you are being led to an erroneous or false conclusion about a scholar's research (and not about Petroglyph and his personality or memory).

>155 faktorovich:

It is irrational to reject a tool that accurately performs a basic test in a manner that is stated in its manual, simply because it is free and you are assuming this means the developers have not invested time to polish the provided tool.

You seem to have missed my point.

My objection to this site as an analysis tool is that without knowledge of how it identifies parts of speech or the basis of other counts, it is an unknown "black box" experiment. We know only what they say about the tests.

But since they are unlikely to have developed the site from scratch, devising every string manipulation to perform counts, they are far more likely to take existing tools and put a web page wrapper on it and package that as a tool for people to use, payment or not.

I think your use of "irrational," "nonsensical" and similar terms is a bit overused, especially when you seem to miss key points such as what I was trying to convey in my post. Perhaps it was my responsibility for not wording things clearly enough or making the post too long. I will accept that. Will you do the same?

But my comments were mainly addressed at this particular experiment and the tool being used. If three variants of the same word are used and the software does not have two of them in its word lists, its ability to identify the correct parts of speech will be hindered because it cannot do so from the sentence structure (it's not that smart).

Using the right tool for the job is essential. If something is set up to work with modern texts, it may have some output for older ones but that does not mean that it is relevant and reliable. That has not been established to the satisfaction of the readers of this thread. You may be happy and confident with it but we've gone through more than 1,000 messages back and forth and convincing is not occurring.

You have asked for questions about your method. We have made them and often the responses, if they are made, are not clear and convincing. I'm willing to listen but this process has not gone well.

James

>156 faktorovich:

I assumed you all know what "filmographic" and "memory" mean separately, or that you could have looked up these terms, and then derived what the two of them when used together would have meant.

No, that's not a safe bet. Neologisms should be defined, especially since you believe you have invented it.

If I were to go on the first five or so pages of responses on the Google search, I would be led to believe that you are experiencing a condition associated with Asperger's, a form of high-functioning autism. I don't think that is what you are trying to convey with this usage and that is why I asked. Instead you are responding with a "you and everyone knew what I meant by this term I invented" response.

My request is that you should be specific. Be clear. Be concise. Be accurate.

Probably I am not successful at all of those and that leads to misreadings and incorrect conclusions of what I have written.

James

>151 anglemark: Glitches such as counting "you'd" as a single word are applied consistently when the compared texts are all in Early Modern English, and thus include a similar percentage of these types of contractions. This only becomes a problem when Modern and Early texts are compared to each other, and then they register with very different outputs via Analyze, as I have explained previously.

Why isn't Analyze a more precise tool? Can you cite a paid-for program that does not generate some glitches, or that can recognize all variants of Early Modern vocabulary and distinguish between "you'd" and "you"+typo-letter? Analyze is one among a few of the best tools I have come across (as I use a few different tools that I list in Volumes 1-2) and it happens to be free. I did not receive any funding for any of my research, so it would be strange if the best programs of this sort received any significant funding; instead the worst services are likely to be the most efficient at manipulating their way into funding. I demonstrated previously the various extreme glitches that Stylo generates that makes even running a basic test without the system gargling up or adding typos to the text impossible. The best method is the one that achieves the central goal of a study in the most convenient and accurate way possible, and the tools I have used achieve accurate attribution conclusions.

>152 anglemark: I see, all of you have been selling this CLAWS tagging system as superior. But you state that it is designed for "Modern English" texts, so it does not solve any of the problems with Early Modern texts using variant spellings etc. You also do not state if the "Shakespeare First Folio" tested was in Early or Modern English; if in the latter, it does not solve the general Early spelling glitches that you guys have been discussing. I have not seen any of you point to specific tagging errors in Analyze, as instead you have all pointed to how there have to be spelling errors or other glitches in a text for Analyze to misattribute a word to the wrong word-type category. Just because somebody wrote a bogus study that makes their tool sound superior to other tools does not mean their tool actually identifies more Standard-English-spelled common nouns than a rival free service.

I tested CLAW just to see if it is usable. The output is a file that places tags next to words using this abbreviation system: https://ucrel.lancs.ac.uk/claws5tags.html Because there are around 50 different tags, a researcher would have to manually add up each tag and then combine tag types to receive even a simple measure like percentage of nouns. Only by writing a program would I be able to perform the step of this calculation automatically, so it is designed for somebody who is writing a program and not for a statistician that just wants to apply an automated tool to determine linguistic measures that can then be compared and analyzed. VARD also appears to tag different words with different colors, but does not even appear to give data that calculates the number of different word-types in the text. I did not download this program to find out, since I assume it is similarly limited as is CLAW. I would absolutely use VARD or any other program that could calculate for free the number of nouns etc. in a text, especially if it increased precision. But based on what I have seen these are not better tools.

>154 anglemark:



That is strange. Total lexical density: 7.14%. Zero in the first two sentences.

It appears that did is only counted as an auxiliary; and so is do. That makes sense for things like do-support for questions and negatives: "Did you ask her? I did not ask her; I do want to ask her". It does not for "You did it" and "I said that you would do it."

It looks like have is counted as an auxiliary as well, all of the time. In "have your cake and eat it", it's counted as an auxiliary (though it should be main verb); and in "have you done it", it's correctly counted as Aux. In "Have you any wool?" it's counted as aux (though it should be main verb).

"Hast thou any wool?" -- hast counts as main verb.

That made me wonder why analyzemywriting counted the spelling "haue" as lexical all of the time. What does it think that word maps to? Hate or something?

Based on the results for "have you met me?" and "haue you met me?", the website thinks haue is a noun.

Looks like when analyzemywriting.com analyzes the words in a text pasted into the website, it runs a comparison against a stored dictionary where every word is tagged with a single part of speech. There's some fuzzy mapping going on between input text and dictionary -- feare, goodnesse, vnderstanding are mapped to the right type of word. Is the software just replacing letters until it gets a match in the dictionary? And it chooses the option with fewest replacements?

Now, since passive voice in English is a specific conjunction of BE + participle, I wonder how AMW handles passives. How accurate are passives on this website? (That was already going to be a future lunch break experiment (tm), but I'm curious now).

I've sent an email to the website asking for details on how their tokenization and POS-tagging works. Will report back if they reply.

>157 Keeline: You guys have tested Analyze's output, and concluded that the measures it provides fit the basic rules its manual describes. While it is strange how they count different types of verbs, they do apply their rules consistently. It would be impossible for any researcher to check every word in an open tagging system for accuracy when there are millions of words in the corpus. And doing such checking and correcting a miss-tagging for every instance of "you'd" would require not only an enormous amount of effort, but also avoiding adding new glitches when automatically correcting all of these and erroneously changing "you'don't" or the like. Unless you suspect that Analyze manipulates noun-counts by hacking into the system and altering results (which you can check by manually testing or testing with other tools the same texts I give Analyze measures for in the GitHub file), the "black-box" nature of the output they provide is not a problem. The data I need for attribution is the number of nouns, and not the specific nouns' tagging; the latter is in a way calculated in my most-frequent words test.

You are all obviously affiliated with the rival tools you are selling such as CLAWS, Stylo, and the general computational-linguistic attribution word-frequency-only method. Thus, it would be just as absurd to believe it is possible to convince any of you in the precision of my attribution method, as it would be to believe it is possible to convince a Tobacco administrator to post that he has come to the realization it is better for humanity if they just stop producing Tobacco.

>161 Petroglyph: No you object that "feare" might be correctly identified as "fear"? As long as all such cases are tagged consistently, in a corpus with millions of words, the overall attribution decisions will not be negatively impacted. You can guess if "passive voice" might not be accurate, but until you actually prove this system makes specific errors in cases where any rival system could identify the item correctly, you are not raising any actual objections against using Analyze.

Quick note:

I am attaching so much importance to have being counted as an auxiliary only, because that does make a difference.

In present-day English, "have been, have seen, have gone, have arrived, have come" are normal, but "Have you any wool?" is not. It was in Early Modern English.

Have was used much more often as a main verb in Early Modern English than today. So if the website marks every occurrence of have as an auxiliary, that is a big mistake with lots of impact on the numbers for verbs and auxiliaries (and, therefore, the lexical density). All instances of haue are mapped onto a *noun*, for chrissakes! That, too will have an effect on counts of nouns, auxiliaries, and, therefore, lexical density.

(Early Modern English preferred the auxiliary be for verbs of motion: "they are arrived; she is come". It took a few centuries for have to completely take over).

>160 faktorovich:

I tested CLAW just to see if it is usable. The output is a file that places tags next to words using this abbreviation system: https://ucrel.lancs.ac.uk/claws5tags.html Because there are around 50 different tags, a researcher would have to manually add up each tag and then combine tag types to receive even a simple measure like percentage of nouns. Only by writing a program would I be able to perform the step of this calculation automatically, so it is designed for somebody who is writing a program and not for a statistician that just wants to apply an automated tool to determine linguistic measures that can then be compared and analyzed.

Being a programmer or having access to one is a basic requirement of this kind of work. Just as being well grounded in statistics is needed.

Depending on your computer system (Windows is harder, Mac/Linux easier on this), it is not too difficult to write scripts that count occurrences of a given tag. I'd have to see some sample output to try my hand at it. Right now I see just the list of tags. On the Mac or Linux command line I would look to standard tools like awk or sed and wc to get some of the counts.

Counting and other repetitive tasks is one of the things that computers are well adapted to handle. They just need to be told how you want it done in terms they can understand. This is programming.

The whole business of identifying parts of speech is very hard to get right. But in this kind of study it can be important if your authorship claim depends on verb and other POS usage relative to the whole work.

James

>163 faktorovich:

I'm not objecting to that. The fuzzy mapping does its job correctly sometimes, and that is good. But it turns out that the fuzzy mapping makes mistakes, too: the high-frequency verb and auxiliary haue gets mapped to a *noun*.

until you actually prove this system makes specific errors in cases where any rival system could identify the item correctly, you are not raising any actual objections against using Analyze

You're play-acting at being a scholar, and it shows again.

We're demonstrating that AMW has issues with identifying auxiliaries correctly. Since the English passive is a specific set of combinations of Aux and Participle, that means that the counts for passive voice may be inaccurate and unreliable, too. That should concern you. (But you're not a real scholar, so it doesn't.)

Also, it is not anyone's job to provide you with better tools. You're the one who selected a free tool that spits out answers you do not understand but that are easy to (ab)use. If it can be demonstrated that this analyzer of *present-day English* is inadequate for analyzing Early Modern English, well, the only one impacted is you.

If AMW is shown to be inadequate, then that simply means that you cannot base a massive re-attribution on its results that flies in the face of everything we know about the period, the people, the texts, and how linguistics works.

>162 faktorovich:

It would be impossible for any researcher to check every word in an open tagging system for accuracy when there are millions of words in the corpus

That's why you work with taggers that have proven accuracy rates of +95%. Duh!

Your manual workflow and your limited imagination are showing again. Why not apply this obsessive need to be able to check everything by hand yourself to the black box that is AMW?

Stylo offers various options for tokenization out-of-the-box: you can choose whether things like don't is counted as a single word/token, or as a combination of do + not. The effort involved (in the GUI) amounts to checking a box; on the command line, you change the argument corpus.lang = "English.all" to something else.

The data I need for attribution is the number of nouns, and not the specific nouns' tagging; the latter is in a way calculated in my most-frequent words test

AMW counts haue as a noun. Surely that's not right? Surely such a high-frequency word throws off the counts in a way that matters?

A quick test for passives:



"It is done. It gets done. It has been done."

The middle sentence is not recognized as a passive. The other two are. Looks like the website looks for the combination of a form of BE with a participle.

The image behind that link also shows that the website says: "Identify where you might be using passive voice". It also says "about 2 out of 3, or 66.67% of your sentences look like they might be passive". And the sentences that are flagged as passives are explicitly flagged as potential passives: they all read "Check for passive".

So. The website is telling you not to just accept the total number of passives at face value -- it tells you that they *might* be passives, and it says to check if they are.

Wanna bet that Faktorovich just copy/pastes the total number?

I quickly tested that one paragraph from that sermon again -- in the original spelling. The site correctly identified three passives, but also this thing:



Look at the words marked in blue: the website is not looking for a form of BE and a participle (it does not use that as one of its parts of speech). Instead, it looks like a form of BE followed by something that *looks like* a participle (ends in -ed).

It is easy to trick such easy apps into giving wrong answers (try "They were bound for Austria. I am prepared to die" -- both are incorrectly marked as potential passives). But the fact that a tagger doesn't get everything 100% correct is not a real objection.

This, however, may be, I think:



While the blue words indeed constitute a passive, there are many more verb phrases/clauses in that passage, and the software only counts it as one. And therefore, all of this passage gets counted as passive.

Same story here:



If one verb phrase in a sentence is passive, the entire sentence is counted as passive. One sentence with multiple passives only counts as one passive. Not sure if that is the most useful level of granularity to look at passives. At least it is used consistently?

Preliminary conclusion: passives are kinda iffy on AMW. It's not as big a failure as Faktorovich's handling of lexical density (or so, so, so much else in that vast mess of misunderstandings and critical failures that is her "method"). AMW not being the greatest at identifying passives isn't that much of a critical error, either. The biggest problem I have is that Faktorovich uses as "% of passive sentences in the entire text" a measure that is clearly marked as the percentage of potential passives.

In other words: it's not so much the small shortcomings in the software that bother me. It's the careless overgeneralization of the person using the software.

>152 anglemark: Looks like an interesting study. Perhaps I'll have some time next week to look at it. Thanks!

>162 faktorovich:

Perhaps there are millions of words but how many different words ? That should only be in the thousands.

I'm not sure why you have addressed this response to me. I have not used these tools and I am certainly not shilling for them. But I do want to know exactly what is being measured and how this is achieved. I have no good reason to trust "black box" analysis. That is like saying:

"This is what it says. TRUST ME."

Well, we are not there yet.

Something is strange about the CLAWS page link provided since the many links I have tried don't go anywhere. The filesystem on the server must be rearranged. Backing up, I went to the Wikipedia page for CLAWS

en.wikipedia.org/wiki/CLAWS_(linguistics)

From there I found the project's website:

https://ucrel.lancs.ac.uk/claws/

and their FREE web tool which has similar terms of service to AnalyzeMyWriting.com. There are licenses for universities and people who want to integrate their program into another application being written.

http://ucrel-api.lancaster.ac.uk/claws/free.html

I took a text I had handy, Tom Swift and His Photo Telephone (1914) and selected CLAWS5 vertical output. Here are a couple sentences of output:

0003420 001 ----------------------------------------------------
0003420 010 " 00 PUQ 
0003420 020 It 00 PNP 
0003420 030 certainly 00 AV0 
0003420 040 is 00 VBZ 
0003420 050 , 00 PUN 
0003420 060 " 00 PUQ 
0003420 070 agreed 00 VVD 
0003420 080 Tom 00 NP0 
0003420 090 . 00 SENT 
0003420 100 ----- 00 PUN 
0003421 001 ----------------------------------------------------
0003421 010 " 00 PUQ 
0003421 020 And 00 CJC 
0003421 030 the 00 AT0 
0003421 040 next 00 ORD 
0003421 050 time 00 NN1 
0003421 060 you 00 PNP 
0003421 070 go 00 VVB 
0003421 080 for 00 PRP 
0003421 090 a 00 AT0 
0003421 100 conference 00 NN1 
0003421 110 with 00 PRP 
0003421 120 such 00 DT0 
0003421 130 men 00 NN2 
0003421 140 as 00 CJS 
0003421 150 Peters 00 VVZ 
0003421 160 , 00 PUN 
0003421 170 look 00 VVB 
0003421 180 out 00 AVP 
0003421 190 for 00 PRP 
0003421 200 airships 00 NN2 
0003421 210 . 00 PUN 
0003421 220 " 00 SENT 
0003421 230 ----- 00 PUN 
0003422 001 ----------------------------------------------------

I see that LT Talk does not use the <pre> tag in a conventional way that preserves spaces. Here's a graphic that shows the output in proper columnar format.



With text output like this, it would be possible to determine counts for each of the 62 (I let the computer count them for me) tag types in CLAWS5. There are 137 tag types in CLAWS7.

I saved my output as a text file and ran this sequence of piped commands on the Mac command line (same would work in Linux/Unix).

cat TS17-CLAWS.txt | awk '{ print $NF }' | sort | uniq -c

I did an extra step to make it into columns so it is not too hard to read here:



This shows the count of each type used. I could start with a list of the available tags and get counts even if 0 were used.

If there's interest I can explain the steps in the command above but this isn't Stack Exchange and I don't know if there is interest.

You are all obviously affiliated with the rival tools you are selling such as CLAWS, Stylo, and the general computational-linguistic attribution word-frequency-only method.

Please stop with the sweeping and inaccurate accusations. If the tools are FREE for CLAWS and Stylo, there is no selling. They are established tools that people in the field know and trust. CLAWS was developed in the 1980s and has had refinements since then.

James

>161 Petroglyph: Yep, that's what I got as well, so it's not just me then. It's probably inevitable that some instances of do as a main verb are mistagged, but it does like they have programmed the AMW tagger to always treat the primary auxiliaries do, have, and be as auxiliaries, even when they function as main verbs. In "You did it" there isn't really any room for misinterpreting "did" as anything other than a main verb! (I tried some sentences with be as a copula and it consistently tags it as a function word.)

As for why it interprets haue and thy as nouns, it's pretty weird, but I suspect you are right – the tagger just matches words to a dictionary, instead of applying rules about English syntax. A normal part of training an automated POS tagger would include which parts of speech can appear in a particular slot.

Thanks for emailing them about it, and about the tokenization! I'll be very interested to hear what they reply.

-Linnéa

>151 anglemark:

do you deduce their analysis from the POS percentages?

I do. I tried these out in small phrases and on their own.

just the pretty coloured bars and percentages. And that's more than a little problematic if it's going to be used as the basis of an analysis of the proportions of each part of speech!

You're quite right there! There's no other colours in the sentences at the bottom other than "lexical". That's insufficient.

The main issue with using AMW for this kind of research is, as you rightly point out, that the software was not developed with that kind of thing in mind. And so everything else follows from there (the results not being precise enough; the black-box approach; the bare-bones POS tagging; the lack of more specific colours in the sentences at the bottom; the rigidity of the spelling requirements; not being able to add non-standardly-spelled words along with their expected POS; most of the tests not being suitable for authorship attribution ...). The uses to which this very simple tool is put are unreasonable.

I ran four versions of the "Once more unto the breach" speech from Henry V through the lexical density

Nice!

>153 faktorovich:

I already tested the "lexical density" measure output on 284 texts just in the Renaissance corpus, so there is no need to go beyond it to answer the question of how genre impacts it.

Given that your results are complete garbage, there is no need to trust them. Your study is not the be-all and end-all of anything to do with genre -- you basically completely ignore all of it.

"Lexical density, as measured by this website, tends to be higher in informative writing than in fiction on the order of about 7%." This points to a major problem of assuming one knows what the results are going to be before starting an attribution experiment

That quote you repeated was the result of a test. Not an assumed conclusion, which is your modus operandi.

While the specific texts they chose to test between non-fiction websites

Wikipedia articles and short stories. Are you saying that short stories are non-fiction? Or did you just read carelessly? Or do you simply not care, in your rush to sound authoritative and play the role of a scholar?

One problem is that "pronouns" can be used instead of "nouns", so they should be counted as "lexical words".

No. Wrong. Such a basic misunderstanding that I'm not going to explain it to you.

And "was" is technically a verb, so even if it is a "linking verb", it should not be counted as a non-verb if the definition states all verbs are "lexical"

Auxiliaries are a fairly constrained subset of verbs. How is it you do not know this?

My statement in the book about this measure holds as true

Nope. Still wrong. But that sentence, taken in isolation, sure does sound like something a real scholar would say.

lovèd ... unreadable for programs characters such as the è in "lovèd" {...} this typo in the original transcribed text.

Ahem. It's not a typo.

>164 Petroglyph: Having translated 15 different Renaissance texts by now, I can report that the use of "have" as the "main verb" was not uniquely more common than it is today. And even if it was, if this was a pattern across all Renaissance texts, it would be marked in the different category for all of them, and thus this consistency would nullify the significance of this divergence for all texts in the corpus with the original-spelling. On a brief search in "Restitution" I found "have" in "have been", "have had", "will have", "would have" etc. I stopped at this point, as there are no instances so far of "have" without a main verb next to it. Yes, the bigger problem is that "have" was frequently spelled as "haue", but it was thus spelled across most of the old-spelling texts I researched, so if it was always mapped as a "noun", it would also be on-average cancelled out as white-noise because the glitch occurs in all old-spelling texts. "Early Modern English preferred the auxiliary be for verbs of motion: 'they are arrived; she is come'. It took a few centuries for have to completely take over". I have no idea where you found this idea, but in all the texts I have translated I do not recall these types of odd phrasings being common. You have to cite specific examples of a text with many such usages, or the source where you found this conclusion, so readers can check how or if what you are saying is relevant.

>165 Keeline: If I start counting the occurrences of these tags with a computer program, it would be more efficient for me to write a program for the entire set of 27-tests, vs. investing time in building a program and having it only be applicable to repeat one (or a few) of the step that already works if I just use Analyze's free software (and then having to repeat the other steps manually). It is amazing how much time you are investing in criticizing my failure to have written a program to make my 27-tests method extremely easy for you to use, and telling me how easy it would be for you to create the perfect program that would have applied such tests flawlessly. Meanwhile, I have to invest years into translating texts like "Restitution" to find intricate evidence to further support my quantitative conclusions, having explained in Volumes 1-2 that they have already provided extremely precise and verified results. There is no need for me to write a program because the methodology works by applying tools that have already been pre-programmed. The wheel has already been invented, it would be a waste of time to re-invent it, instead of using that wheel to build a new type of vehicle.

>167 Petroglyph: Stylo cannot even manage a basic pre-programmed experiment without destroying the data fed into it, and it can only accept files that are posted on websites, and not uploaded into its system. Arguing that Stylo is more precise because Stylo or one of these other programs rated itself as being "95%" accurate is like saying that we should all eat Company X's fast-food because their advertisement said it was "the best".

>168 Petroglyph: Yet again, the main goal for an attribution test is that it consistently applies the same set of rules/ tags etc. to all of the texts fed into it. Even if a sentence is not in fact "passive" and only "looks like it might be passive"; the Analyze program's rules are going to always identify as "looks like passive" the same types of sentences. Thus, while a grammarian who checks the sentences might find glitches in this system, the same grammarian would have to agree that the same types of glitches occur in the same types of situations; thus the grammarian would state that statistically speaking this measure would be just as a good at identifying patterns of usage, even if its definition of "passive" does not precisely match the grammarian's. The 27-tests attribution method is not a grammar-checker; it is a system to identify patterns of linguistic usage that are similar between texts written by the same author; however one defines "passive" (and it can even be defined as the exact opposite of "passive" by a program) the attribution method will be accurate if it repeatedly tests for this specific measure in a similar manner across all texts and gives answers that are consistent with whatever definition it is using. The accuracy of the "passivity" test can be verified by looking at the data; and you can check this accuracy by going into my 284 texts table re-ordering the "Passive" column by least-to-highest numbers, and you'll see that Percy's texts make up most of the bottom end of this range, and Verstegan's texts cluster on the upper-range. If this test had failed to establish divergence between signatures, the results would have been random at both ends, or with equal mixtures of all or most of the six potential ghostwriters. To check how this works on the large scale of a 30,000 word text, you would have to check all cases of miss-identified passive voice, and check such cases in different texts from different bylines, and then see if there is a similar percentage of miss-identified instances of passive voice that basically cancel themselves out. I can see that this is the case in the data; if the data failed to show divergence, I would have chosen a different test. It does accurately show divergence, so there is no need to fix this test. Yes, in an ideal world a better program can be devised, but apparently specialist programmers in this field are busy finding fault with other programmers' work.

>147 faktorovich: The joke is that a Ghost cannot write,

That is extra-linguistic. It is not universally believed; as an aspect of God, the Holy Ghost certainly can write, and many people have recorded what the people from the other side have said. Whether or not you or I believe that happens doesn't really have an impact on what people using the words mean.

I don't think you have much respect for these author's religions. The fact that someone would credit the Holy Ghost then just like many people would credit God now is not surprising to me.

The credit line that a text was written by "Nashe's Ghost" is not at all convoluted, but is a direct satire about "writers" who do not really do their own writing (perhaps because they are dead, or perhaps because they are illiterate) and instead have a "Ghost" (i.e. ghostwriter) write their book(s) for them.

It was credited to "Tom Nash his ghost", and is generally believed to be written by John Taylor. No need to talk about satires on fake authors; it seems quite clear to me that no one at the time believed the author was Tom Nash, and that the author was merely adopting a name to write under, like someone might write as Ronald Reagan's ghost or Abraham Lincoln's ghost to make a point about how those figures might view today.

It is not "extra-linguistic", but basic linguistics to grasp that the Workshop was repeatedly confessing to their own authorship by joking about the possibility of Ghosts (instead of them) doing the work.

The only linguistic information from "Title by Name" is that it is asserting that Title is by Name, even if that Name is "the Holy Ghost". Everything you're saying is extra-linguistic.

>170 Keeline: The main problem you missed is that you have created 62 different tags for different word-groups. Checking all of these groups separately against a complex corpus such as the Renaissance is more likely to generate glitches than when testing only major groups such as nouns and verbs. While the test for "passive voice" or for "nouns" in Analyze might generate a few glitches from misspelled words, the same errors would also appear in CLAWS, and they would be magnified because you will be using 62 different tests instead of the fewer count that I include in the word-type category in my 27-tests. You have to simplify these results by adding a calculation step that groups these types. Or if you use all of these 62 tests, you are going to have to spend even more time in the later stages of my attribution method, where you calculate and add up proximity results. If you keep advertising tools like CLAWS and Stylo as superior to the specific tools I am using like Analyze; then it is a fact that you are advertising for them; you have not disclosed an affiliation with them, but this is the only logical reason you would keep insisting their tools are the only ones allowed to be used for computational-linguistic attribution.

>173 Petroglyph: My results are exactly correct, as I have confirmed in finding evidence to support them across borrowings and various other pieces of evidence in 17 volumes of BRRAM so far. You have barely read the few paragraphs from Volumes 1-2 that you have quoted from. Whenever you read other parts of BRRAM, you attempt to take credit for what I explain in BRRAM as if it is your own ideas, and you did not find out about them from my explanations in BRRAM.

"Wikipedia articles and short stories": this statement is clearly creating a duality or stating the two things are different from each other, and thus one is non-fiction and the second is fiction. You are finding fault with absolutely true statements to avoid responding to my actual criticisms of the flaws in your suggestions. Across the following comments you are just saying false nonsense that is not responding to anything I actually stated.

>178 prosfilaes: It is "extra-linguistic" to seriously argue that the "Holy Ghost" has been proven to be able to "write", and not to argue the opposite; the "Holy Ghost" is a fiction, and it is rational to interpret authorship by the "Holy Ghost" as an absurd or satirical statement, and not as a factual one.

Earlier today, I translated this section from "Restitution":

---The ancient pagan Germans, especially the noblemen, as both Krantz and other writers testify, sometimes took the names of beasts, as one would be called a Lion, another a Bear, another a Wolf, and so in a like manner these two before named princes (referring to "Hengist", whose name is derived from "Horse" and "Horsa", whose name is also derived from "Horse") had their denominations...---

Here is my annotation for this section (Verstegan cites Krantz by name later on):

---Albert Krantz’s "Chronicle" and "Saxony" both include names such as Leo dux Belgicae (Lion, the Duke of Belgium), Lupus Vasconiae dux (Wolf the Basque Duke) and Lupus Karoli (Wolf Charles). Verstegan might have especially noticed a line such as this one in "Chronicle": “…Ursus, alius Leo, alius etiam Catulus dicetertur.”, Latin for, “…Bear, alias for Lion, alias for Cub to be named.” This is an example of Verstegan hinting at his multi-bylined ghostwriting under clearly fictitious pseudonyms (such as “A Monday”) as well as under historical figures' bylines, such as “James I”; he is stressing that the use of such aliases in history books suggests the author is writing clearly fictitious names because the actual names of the historical figures discussed might have been lost to history.

There are thousands of similar references and discussions about aliases and pseudonyms etc. across the Renaissance texts I am studying. Lines such as "Nashe's Ghost" or references to pseudonyms are not the exception, but occur at some point in most of the texts I have translated.

>181 faktorovich: It is "extra-linguistic" to seriously argue that the "Holy Ghost" has been proven to be able to "write", and not to argue the opposite

Do you know what extra-linguistic means? Things like the existence of the Holy Ghost are factual questions, outside the field of linguistics.

the "Holy Ghost" is a fiction, and it is rational to interpret authorship by the "Holy Ghost" as an absurd or satirical statement, and not as a factual one.

So you don't believe it, so they didn't believe it? If it was accepted to be placed on the title page, it was because the readers did not interpret it as an absurd or satirical statement. And if the average reader would not have interpreted it as satire, why would you assume the author intended it as satire?

There are thousands of similar references and discussions about aliases and pseudonyms etc. across the Renaissance texts I am studying.

Sure, if you interpret the statement about "The ancient Germans" as a claim about something entirely different. I think this proves that Verstegan was a furry, and I'm sure I can find thousands of similar references proving that the British were all about furrydom at the time.

>179 faktorovich:

Your continuing accusations of collusion by me or anyone else to these publications is not only false but growing to be tiresome. If you have concerns then there are professionals with whom you can consult for this kind of issue.

I have not even heard of Stylo for R or CLAWS before reading about them in this group.

It is CLAWS v5 which has 62 parts of speech tags. If you look at and read them, you will find that they have a level of granularity that your favorite web tool cannot approach.

But if you want to aggregate all of the nouns together for your counts. For example:

NN0 noun (neutral for number) (e.g. AIRCRAFT, DATA)
NN1 singular noun (e.g. PENCIL, GOOSE)
NN2 plural noun (e.g. PENCILS, GEESE)
NP0 proper noun (e.g. LONDON, MICHAEL, MARS)

This can be achieved with an egrep statement:

egrep "NN0|NN1|NN2|NP0" input_file.txt | ...

The 40+ year track record of CLAWS as a trusted tool fo identifying parts of speech with a very high degree of accuracy is why professionals in the field turn to it.

Since the programs are free to use, there is no benefit or motivation to sell anything. They have their reputation and 40 years is longer than most computer tools. Sometimes I think you reject anything that is unfamiliar or requires a little work on your part.

You claim that I've spent lots of time with the other tools is also incorrect. The actual work with CLAWS to paste in a text, generate a report, and use my experience with the tools available in the Unix/Linux command line to tabulate the results in a meaningful way adds up to about 15 minutes. It took longer to write about it than to do it. I started with BSD Unix in 1985 and have worked with Linux continuously since 2000. What I don't know offhand, I know how to look up. But the most challenging part was getting the results to display in columns so they didn't become a 62-line list of numbers and tags. Even that was a matter of making a Google search and consulting 5 pages until I found a solution that would work for me.

I don't know what you mean by proximity results. However, I would note that the numbers on the left of the CLAWS output have the sentence and word number so one could certainly work with that.

My main point was to illustrate that your metaphorical "throwing up of your hands" at anything that is not familiar can be resolved with just a little linking of one utility program to another.

Is it your goal to insult me until I go away? Is this how a professional scholar behaves?

Don't let yourself be characterized by the old adage:

When the only tool in your toolbox is a hammer, everything looks like a nail.

Instead I suggest:

There are more things in heaven and Earth, Horatio,
Than are dreamt of in your philosophy.

Give a man a fish and you feed him for a day; teach a man to fish and you feed him for a lifetime

James

>176 faktorovich: Stylo cannot even manage a basic pre-programmed experiment without destroying the data fed into it,

The fact that you don't know how to use a tool doesn't mean it doesn't work.

>177 faktorovich: If this test had failed to establish divergence between signatures, the results would have been random at both ends, or with equal mixtures of all or most of the six potential ghostwriters.

You're using your conclusion to prove a test that the conclusion is based upon. That's wrong. For this entire thread, other people and I have been telling you that you need to calibrate on a known corpus; you can't prove that something separates works by different authors by running it and deciding whether you like how it separates works.

Yes, in an ideal world a better program can be devised, but apparently specialist programmers in this field are busy finding fault with other programmers' work.

Says the person who is dismissing every scholar's work on the authors of the British Renaissance and who hasn't bothered to study existing programs.

>184 prosfilaes:

I thought you would pick up on that and "it can only accept files that are posted on websites, and not uploaded into its system>"

Which of course is pure rubbish. Anyone can establish this in under a minute if they wish. R/Stylo works from a corpus subdirectory* on your computer - there is no posting on websites, no uploading. You just copy your prepared file into the appropriate subdirectory.

Of course petroglyph got there first and explained this on post 251 of the original thread.

* OK you can load the corpus from whatever directory you want, but it is a subdirectory called corpus by default.

>159 faktorovich:

Glitches such as counting "you'd" as a single word are applied consistently when the compared texts are all in Early Modern English, and thus include a similar percentage of these types of contractions.

But it is not just a matter of how many words a text contains nor what the proportion of contractions is in a text, though that is part of it. I looked at the first two texts you list in your bibliography, accessing them at the sources you specified that you had used: The Map and Description of New-England by William Alexander, 1st Earl of Stirling, and "A Pleasant Commodie called Looke About you" by an anonymous author, published in the collection A Select Collection of Old English Plays, Vol. VII (4th edition), 1876.

Before getting to the issue of contracted forms, a different issue: The Bodleian online edition of The Map and Description contains a lot of invisible characters. How many of your texts were affected by that, and how did you fix them before you ran the text through AMW? AMW also removes characters such as the macron over ē in words such as "Patēt" and "cēsured", splitting the words that contain macrons – so "patent" and "censured" become "pat t" and "c sured". How do you handle that type of issue, and how many of your texts are affected by it? (Incidentally, I see that it does the opposite with hyphens, removing them and combining the words.)

As for the contractions, The Map and Description apparently contains no contracted forms at all. "A Pleasant Commodie", on the other hand contains a lot of them, e.g.

"ROB. You see I am weapon'd, do not, I beseech thee.
I'll stab them, come there twenty, ere they breech me."

or

"POR. I wonder how thou cam'st so strangely chang'd!"

But there are also many potential contractions that are written out in full, so on subsequent lines we have

"SKINK. I'll bring it ye to the sheriff's, excuse my absence.
FAU. I will, my noble lord; adieu, sweet prince."

Since this is a play written in rhymed blank verse, scansion was presumably a consideration, but the reason for the variation is less relevant than the fact that there is considerable variation within the same text. We have

"Had you not been, I would have left the place;
My service merits not so much disgrace."

and also

"My lord, but that my case is desperate,
I'd see your eyes out, ere I would be cheated."

Apart from the number of words in the sentences, you have a more serious problem with your indata, which will affect your outdata a lot. "I would" is correctly counted as one pronoun + one auxiliary in AMW. "I'd" is incorrectly counted as one auxiliary. "exil'd" is incorrectly counted as an auxiliary instead of a main verb. "red-cheek'd" is incorrectly counted as an auxiliary instead of an adjective. If the frequency of certain parts of speech are some of your "tests" that allegedly indicate authorship, how do you control for the fact that the tool you use doesn't provide you with the correct proportion of main verbs, auxiliary verbs, adjectives, and pronouns? (This is why tokenization is such a crucial thing.) Just for the hell of it I ran the text of "A Pleasant Commodie" through AMW twice – I did not clean it up, so in both runs, the character labels (the name of the character who speaks on stage) were left in, as were page numbers, stage directions, and other chaff.

Run a) was the text as it is from the source and run b) had all instances of "I'll" expanded to "I will", and all instances of "you'll" expanded to "you will".
Run a) had 26482 words with 10.85% pronouns and 6.75% auxiliaries, run b) had 26708 words with 11.6% pronouns and 5.84% auxiliaries.
What, if anything, does that kind of difference do to your results? (Keeping in mind that this was just the expansion of one single contracted auxiliary with two individual pronouns, in one individual text.)

-Linnéa

>174 faktorovich: , responding to >164 Petroglyph:

Having translated 15 different Renaissance texts by now, I can report that the use of "have" as the "main verb" was not uniquely more common than it is today.

To make that kind of claim you'd need something more than vague assertions based on anecdotal evidence. Why the scare quotes around "main verb"?

"Early Modern English preferred the auxiliary be for verbs of motion: 'they are arrived; she is come'. It took a few centuries for have to completely take over". I have no idea where you found this idea, but in all the texts I have translated I do not recall these types of odd phrasings being common.

See https://www.librarything.com/topic/337240#7814046. That be and have have been used in (constrained) variation as auxiliaries to form the present perfect and past perfect tenses in English is a very well-researched phenomenon. Again, you can't go on what you think you remember from texts you have read – not to mention the fact that you already commented on the construction (in your post 878 in the prev thread) and said that it wouldn't sound odd to anyone who knew English grammar.

To be honest it is a little depressing that you have spent years of your life working with EModEng texts, without being aware of this pattern of variation.

-Linnéa

>182 prosfilaes: A fuller definition for "extra-linguistic" is "not included within the realm of language or linguistics"; and pretty much all subjects that require language to debate about them are within the realm of "language". Thus, there are many terms that apply to the existence or non-existence of the "Holy Ghost", but it being a subject that is "extra-linguistic" or outside of language is not a relative term. If you are applying the term to mean a general incorporeal theological question of the existence or belief in God versus the non-believers being deserving of dismissal... This conversation has strayed into the opposite from the intended satirical meanings that writing by the "Holy Ghost" was meant by the Workshop to express.

The biggest problem with previous author-attribution studies, and especially those about the Renaissance, is that bylines have all been taken seriously, or as if even an obviously satirically absurd byline like "A Monday" represents an actual author, even if most of these bylines are not supported by any documents that support the existence of a person by that name. If you lack a sense for when an author is using satire (such as exaggeration of a fictitious pseudonym into a blatantly fake one to make a sarcastic criticism) or irony (saying the opposite of the intended meaning); you cannot interpret the actual intended meaning. Without awareness of the intended meaning, a purely quantitative analysis of a corpus is going to lead to the broad misattributions that are now accepted. For example, in the "Shakespeare" byline so that some texts with "Shakespeare" in their original byline are now not accepted as having been written by "Shakespeare", while those that had another byline, or were anonymous are now broadly accepted as by "Shakespeare"; the acceptance of this byline in the first place stems in ignoring that this byline with modernized spelling is "Shake-spear"; and "John Shakespeare's" signature handwriting matches "William Shakespeare's". Thus, an absence of a sense of humor is one of the major obstacles attributors have had when handling this corpus. The "average reader" has absolutely failed to see the humor in these jokes, but this failure to appreciate the absurdity is the fault of scholars and professors who fail to notice it.

While it is possible to prove all sorts of things based on fragments of evidence. The fact that absurd pseudonyms were overused in the Renaissance is provable by statistically adding up the number of bylines with silly words in them, like "Kyd"/"Kid", "Donne"/"Done"; "Chapman" is a name that is explained as being absurd in one of Percy's plays, and Percy was the main ghostwriter under this byline. There is an entire section in "Restitution" where Verstegan explains how names can be created out of alternative spellings for common words. Your conclusion regarding "furries" cannot be proven factually because for starters there were no furry-like costumes that were worn in the Renaissance, and your argument requires proving what they did in their private lives. In contrast, the bylines and their meaning are explained by these Renaissance writers themselves in numerous cases, and they were printed on these books so their existence is already substantiated.

>188 faktorovich: your conclusion regarding "furries" cannot be proven factually because for starters there were no furry-like costumes that were worn in the Renaissance

Do you have difficulty detecting sarcasm?

>183 Keeline: Any tool that is 40 years old, and might not have been edited since it was created is most likely inferior to a tool that performs the same function that was created in the past decade, as computing has leaped forward. And Analyze actually gives the types of output data that my method requires (the statistical data on each individual test component), whereas the tools you have pitched require programming to perform additional steps before deriving the data points I need for my method. If you keep arguing for this ancient tool, you at least should stop objecting that you object to me objecting to you being biased towards it/them, and inserting advertisements for it/ them.

"Proximity results" are obtained in one of the steps in my method, as the proximity (degree of similarity or divergence) of each text to each text is calculated for each of the tests.

Your last quote is about the value of teaching things; don't you find that if this is your motto, you should have at least read some of my BRRAM series, or enough to learn what "proximity results" means? As I have repeatedly explained the things you guys are suggesting are extremely anti-helpful, and have no relevance to perfecting my already precisely functioning 27-tests computational-linguistic author-attribution method. Most of what you are saying are indeed insults. I am just trying to respond to your points to avoid readers misinterpreting my research from the false statements you are saying about it without understanding even a basic component of it like "proximity results". Why do you think I would want or need you to "go away"? If you guys keep saying false nonsense this should prove your position to be false for me, as you keep circling around the same falsehoods.

>184 prosfilaes: It is a fact that I did use Stylo (as I described earlier in thread #1), and that's how I know it does not work because it is infested with bugs, and it is unusable for any practical linguistic experiment.

It is absolutely not true that I "need to calibrate on a known corpus", as "calibrate" means: "adjust (experimental results) to take external factors into account or to allow comparison with other data". Such adjustments only distort the data to make it fit with a researcher's desired results; or force the re-attribution of texts to the same bylines that they were previously assigned to, instead of establishing the true authorship groupings of the texts. Instead of "calibrating" or adjusting my results to fit with others' data; it is indeed accurate to check if an attribution test works by if the results it provides distinguish between texts in a meaningful pattern; and if the attributions match documentary research into if such attributions were possible, or if they are the most logical attribution conclusions. There are many other checks that I have performed to establish my results are correct, but at no point have I adjusted the experiment's results to fit any previous researcher's data. Doing so is both illogical and immoral.

I have already done a brief review of the existing linguistic programs, if I look any further it would be just to create a program that applies the 27-tests method simultaneously to a corpus.

>185 andyl: Based on my previous testing, R/Stylo has a glitch that blocks usability of their subdirectories, as it looks like files are added to the correct subdirectory, but Stylo does not see these files, or produces numerous errors and only recognizes files when they linked to from websites. Yes we did discuss this programming error in Stylo in thread #1.

>192 faktorovich:

Well it is very telling that the only one who cannot manage to get R/Stylo to do stuff is you. I am not a computational linguist nor am I an R user but I would bet any money that I would be able to download it, install the stylo package, and get it to analyse plain text documents which exist on my machine within 15 minutes.

I think you have amply demonstrated that your computer skills are lacking over the course of the entire discussion.

>186 anglemark: For "Look About You" I tested the EEBO old-spelling edition, and not the reprint from the 19th century; I used the reprint to help with transcribing the text for the translation of this volume in BRRAM. I also did not use the Oxford edition for "Alexander", but instead used the EEBO old-spelling version of this text, as I did for most of the texts, as EEBO is the most accessible format. With just a few automated corrections for odd punctuation and lines over letters etc., it creates a smooth text without any significant glitches that prevent precise results. I used other sources only when EEBO was not available, or if I had initially tested a new-spelling edition, before adding an old-spelling version, or the like. You can see which edition I used by checking the "old/new" spelling column in the 284 texts data file.

I did not see any ""'censured' become... 'c sured'" glitches with the formats I used because I substituted all of the lines over e etc. glitches prior to running tests.

You might be comparing a modernized edition to an old-spelling edition. If you are not, and you notice more contractions in one of the texts; this is an obvious linguistic difference between two distinct authorial signatures. Even if the counting programs do not understand what the contracted words mean and misinterprets them; they apply the same type of misunderstanding to all texts and this creates a diversion in the output that reflects this over-use of contraction by a given authorial hand. Thus the attribution results are accurate, even if the counting program would have received a failing grade as a grammarian.

Similarly, assuming you are correct in this statement: "'red-cheek'd' is incorrectly counted as an auxiliary instead of an adjective." If it is always "incorrectly" counted and all words like it are always "incorrectly" counted as an "auxiliary"; then, when you look at the pure mathematical data, they are all correctly counted as a similar type of word (be it "adjective" or "auxiliary"). In a perfect world, I would have the time or the money to build the perfect program that precisely counts even an old-spelling test, but to reach perfection, this would be an enormous undertaking. And it is entirely unnecessary, as the simple free tests already reach precise results, and they can also be used by the general public with the no-programming-needed method I already described.

I did delete page numbers and most other irrelevant info, but not the names of characters, and some other repetitions that appeared in the original texts.

A percent change in the data with contractions subtracted is a change that probably would not change the attribution of a text, but it might if it also impacted other tests more significantly. I left contractions as-is to allow all of them be counted similarly (all generating glitches, or all being interpreted in the same systematic manner). After running this test, you should realize that it is not a good idea to turn contractions into words, especially not unless you intend to change all contractions in all texts, and thereby to clean up a likely significant style-indicator that is impacting the outputs (even if a grammarian might guess the contractions are not being counted as what they actually are).

>187 anglemark: They are not "scare quotes"; they are quotes to cite your usage of the term.

The evidence is inside the 17 volumes of BRRAM; if you opened one of them, you would realize I am not making "vague assertions", but rather those in fact supported in these many volumes.

It is absurd to cite your own ramblings on LibraryThing to prove a grammatical rule about all texts in the Renaissance. If you believe you know more about Language X (having only read your own chats about it) than somebody who has been translating Language X for a couple of years as their daily job; then, you have an inflated sense of your superiority.

>193 andyl: To repeat what you said: you have never used X before, but you are sure that X works perfectly, and that you can make it work perfectly, and also that I know nothing about either X or about computers (just because you imagine I do not).

>196 faktorovich:

No - I think you have demonstrated your lack of knowledge. A complete lack of knowledge about character encoding, about Excel and CSV files, and more.

Why am I sure I can get it working so quickly.
1) Other people contributing have no problem - so there are no major breaking bugs. Literally millions of people use R. Fewer use stylo of course, but enough that I have confidence there are no major breaking bugs.
2) I have decades of experience of working with open source software (I guess I am of the same vintage as keeline in that I also started with BSD 4.2 in around 1985), and I am a software developer by trade.
3) I have had lots of experience in learning (through use) of many different types of software - some without any documentation at all.

Now obviously I would only get the basic analysis that stylo does in that initial run. But it is enough that it would prove my point that getting started with R/stylo isn't difficult.

>190 faktorovich:

Any tool that is 40 years old, and might not have been edited since it was created is most likely inferior to a tool that performs the same function that was created in the past decade, as computing has leaped forward.

Wrong again. What do they say about their tool (emphasis added)?

Our POS tagging software for English text, CLAWS (the Constituent Likelihood Automatic Word-tagging System), has been continuously developed since the early 1980s. The latest version of the tagger, CLAWS4, was used to POS tag c.100 million words of the British National Corpus (BNC).

You have shown that you are prone to snap judgements to dismiss anything with which you are not familiar. You could have looked at the link I provided where you'd see the above quote in the first line of the introduction. But clearly you did not do so.

Since you have shown that you have your own definitions for certain phrases that are not to be found in dictionaries or websites, I have decided it is easier to ask what you mean by something like "proximity results".

The normal use of "proximity" is something that is physically close to another. That is why I noted the sentence and word number information that is in the output of CLAWS with the vertical output.

I will leave it to others with more experience to state whether "proximity" is used as you do. It is not something I have encountered nor will I look it up at this time.

"false nonsense"? Who is the one who throws accusations that the participants in this forum are all on the payroll of the producers of free-to-use text analysis software like Stylo for R or CLAWS?

I want the insults and accusations to stop. It is not in keeping with the terms of service of LibraryThing. If you have nothing that you can possibly learn from anyone else, then there is little point in continuing this thread.

James

>187 anglemark:
>195 faktorovich:

From Nevalainen (2006, p. 94), a standard textbook:

{T}he perfective structure was well established in Early Modern English. One difference between the Present-day and Early Modern English constructions is that Early Modern English normally preferred the auxiliary be with verbs of motion (for example, arrive, come, depart, enter, fall, go, land, return, ride, run, sail, set) and change of state (for example, become, change, grow, melt, turn, wax ‘grow’). Come is illustrated in (10). The preference changed in the course of the Late Modern period, when have replaced be.

(10) Al thes are come (sayde he,) to see yow suffer deathe; there ys some here that ys come as farre as Lyengkecon {Lincoln}, but I truste ther commynge shal be yn vayne. (HC, Thomas Mowntayne, Narratives of the Days of the Reformation 1553: 203)

Nevalainen, Terttu. 2006. An Introduction to Early Modern English. Edinburgh Textbooks on the English Language. Edinburgh: Edinburgh University Press.

>182 prosfilaes:

"Your conclusion regarding "furries" cannot be proven factually because for starters there were no furry-like costumes that were worn in the Renaissance, and your argument requires proving what they did in their private lives."

Such a nice setup. I lol'd at that.

>188 faktorovich:
for starters there were no furry-like costumes that were worn in the Renaissance, and your argument requires proving what they did in their private lives.

You don't know that. People in the 1500s had easy access to pelts and furs. If you read between the lines, it may even go back all the way to the intimate friendship between Gilgamesh and Enkidu.

"Exit, pursued by a bear": https://www.yalehistoricalreview.org/exit-pursued-by-a-bear/

>202 lilithcat:

Making the beast with two backs.

There's the "falling in love with a donkey" bit in Midsummer Night's Dream, as well.

We may be on to something here...

>196 faktorovich: >197 andyl:

I have just done the experiment.

I installed R onto a linux server and installed stylo as a local library, and used it to analyse 3 texts using the defaults. It took me approx 10 mins 45 seconds - it would have been less but stylo compiles some bits of itself when installing.

In some senses my setup would have been less easy to use, and less effective, than the standard as I do not have X on the server so no nice Tk interface for me.

>197 andyl: If there were no bugs in Stylo/R/other popular computational-linguistic attribution methods; then, the Renaissance would have been previously accurately attributed, instead of having new studies varying between introducing new re-attributions of isolated texts, and re-affirming ancient attributions that stem back to the Renaissance itself. The fact that the currently accepted bylines are ridiculously incorrect, as I explain across BRRAM, is the reason these methods have been proven not to work. The enormous number of bugs that make these tools unusable and uncheckable is the reason specialists in this field have been able to avoid being discovered manipulating results or providing fraudulent results that are designed to make the methods seem "accurate" because they re-affirm current bylines, and not because it actually establishes true authorship. The absence or insufficiency of coherent manuals or "documentation" is one of the ways this un-usability has avoided detection, as nobody can learn something without any guide as to how it works. This would be like learning a new language without a single dictionary or grammar book to describe its rules or definitions. Yet you are arguing that you can learn this undefined language and this makes you superior to anybody who can only learn things that are learnable. The only reason you can be making such a claim is either it is a language you have made up, and have failed to make a manual to allow anybody else to learn it; or you are simply making an untrue exaggeration designed to make yourself appear superior.

>198 Keeline: How did they find 100 million words in an English language that at most has around 1 million words? You don't see anything suspicious about this claim? Software is only free to use if it can be used by itself without paying a computer programmer to finish it by adding tools that make its output usable in a practical linguistic experiment. I am making factual statements, with specific examples from your own statements across this thread. All of you are making false accusations that my method is "garbage" and that I am "prone to snap judgements", along with other insults you have tossed at me. My 27-tests method does work. I decided on this method after intricate research into the potential forms of computational-attribution (as I have explained previously, and document across my explanations in the BRRAM series). I have tested the largest corpus of texts and bylines ever tested from the British Renaissance. I am here to explain my method to all who are interested in it. If you are going to advertise your own method(s) as superior in a thread that is about my method and my findings; I am going to explain to readers any falsehoods or errors your bring into your insulting arguments tossed at me to drown out the significance of my findings.

>199 Petroglyph: The Workshop texts I tested are mostly between 1570 and 1648, the text you cite and the unique feature you are mentioning appears to be from just before the Workshop's time. I previously gave these examples in "Restitution" where I found "have" in "have been", "have had", "will have", "would have" etc. All of these are the standard Modern applications of "have" and not this earlier variant you are referring to. Either way, any variant usages of "have" are not likely to make any significant impact on the 27-tests I applied to this corpus. Thus, such concerns are irrelevant to checking if an attribution method is accurate.

>203 Petroglyph: I hope you will all spend the next 2 and 1/2 years pursuing this hypothesis, only to face a similar thread of insults as this thread has been.

>191 faktorovich: It is absolutely not true that I "need to calibrate on a known corpus",

I really hope you never run a grocery store. "That's two grapes, and your scale says five pounds; you need to calibrate your scales!" "Such adjustments only distort the data; I know the scale is correct, because it is logically correct."

>206 faktorovich:

Keeline, I apologize for butting in here.

HoW diD tHey FINd 100 MILlioN WOrds In an EngLISh LAnGuAge tHAT at mOST hAs AROuND 1 mILlIOn WORds? yOU doN'T See ANYthING suSpiCiOus ABOuT tHis cLaiM?

Faktorovich has jerked her knee and has produced more lies to smear people who are not Faktorovich.

A corpus of 100 million words does not contain that many unique words, obviously. It will contain millions of instances of the word the, a few tens of thousands of the word rather, and maybe a few thousand of the word neighbouring. 100 million is the number of tokens (all the occurrences of all the words), not the number of types (the unique words).

Take this smallish corpus of 100 English novels of the 19th and early 20thC of 14+ million words. It contains 14,349,539 tokens spread out over 80,636 types. The word the alone accounts for over 716,000 of those 14 million tokens.

Here is a screenshot of the 68th-82nd most frequent words:



This is the range that is still dominated by function words, but where high-frequency content words show up more and more (know, see, time, good).

The Corpus of Contemporary American English (COCA) contains 1 billion words (tokens). The British National Corpus (BNC) contains 100 million words (tokens) from the 1980s up to 1993. The open-source version of Early English Books Online (EEBO) has 755 million words (tokens), spread out over +25,000 texts. The Proceedings of the Old Bailey, a collection of just under 200,000 courtroom trials held at the Old Bailey in London between 1674 and 1913, contains 120 million words (tokens).

And these are fairly constrained corpora, aiming for one variant of English, or one period, or one coherent group of texts, with very strict admission criteria. Web-based news corpora are up to 15 billion words (tokens). And even these are not random crawlers harvesting texts here and there and adding them willy-nilly. The sources are carefully chosen and vetted.

How do I know Faktorovich is lying? She proudly touts the 7.8 million words (tokens) of her own corpus.

Any and all pretensions to presenting a corpus larger than 1 million words as suSpiCiOus end right there. It's flailing, unfocused lies and smears.

No-one who knows anything about corpus linguistics should seriously, knowingly produce a quote like the one above.

Faktorovich, Keeline is quite correct in stating that your methods are garbage (they are garbage from front to back, and so are your results). He is quite correct in stating that you are "prone to snap judgements" that, somehow, always affirm your opinions regardless of their connection to reality (which often goes through at least one filter of misunderstanding). You demonstrate those snap judgements (or "brisk impressions" as you call them) over and over again.

>208 faktorovich:

Don't plagiarize our idea! The timestamps are right there in the post headers!

>207 faktorovich:

Lies and deflection.

From a quick search through Romeo and Juliet:

My sword, I say. Old Montague is come
And flourishes his blade in spite of me.

the guests are come, supper
served up

Make haste. The bridegroom he is come already

>206 faktorovich:

How did they find 100 million words in an English language that at most has around 1 million words? You don't see anything suspicious about this claim?

You must work pretty hard to overlook obvious statements. I'll do your work on this for you. Details on the corpus can be found here. I don't know why they don't have https (an SSL certificate to encrypt traffic) but my browser (Chrome) complains of this.

http://www.natcorp.ox.ac.uk/corpus/index.xml

The 100 million words is not 100 million different words but the number of words in the corpus.

Oxford Dictionary has 273,000 headwords; 171,476 of them being in current use, 47,156 being obsolete words and around 9,500 derivative words included as subentries.

https://en.wikipedia.org/wiki/List_of_dictionaries_by_number_of_words

Software is only free to use if it can be used by itself without paying a computer programmer to finish it by adding tools that make its output usable in a practical linguistic experiment.

OpenOffice is a free (free as in beer, free as in freedom because the source code is available to those who care to look at it) open-source software. It includes a word processor comparable to Microsoft Word, a spreadsheet like Excel, and a presentation program similar to PowerPoint and other tools. However, it will not write a novel for you nor even your BRRAM series. What kind of world should do everything for you for free without the slightest amount of work?

What I did to string one Unix utility to another with the pipe characters (|) does not qualify as programming. It is a tiny bit clever perhaps but it is just using the tools just as a furniture maker would use several tools (saw, planer, drill, screwdriver, sander, etc.) to make a piece of furniture. There is a traditional sequence that these tools are used and each requires a bit of skill to use them well. Here's what I wrote before:

cat TS17-CLAWS.txt | awk '{ print $NF }' | sort | uniq -c

cat filename — displays a text file line by line
awk '{ print $NF }' — gets the last word from each line (the parts of speech tag) and displays it
sort — performs a basic alphabetic sort, leaving duplicates intact
uniq -c — takes a sorted list and produces a list with a count of the number of occurrences (-c)

This is not magic. It is basic processing of a text file to extract what you want from it. It is on a par with writing an Excel formula to perform a tabulation or a conditional one if you want to make it a little more advanced.

my method is "garbage"

I have said no such thing. I have expressed doubts and asked questions. You asked for them.

Extraordinary claims require extraordinary evidence and persuasion.

You claim that you are rewriting the bulk of literary history with your findings. Yet, you don't think it is important to demonstrate that your method works for texts of 100% known authorship. It is beneath you, apparently, from your repeated vociferous reactions in this thread, including #191 most recently.

I have looked at more than a few scholarly articles about authorship attribution studies and a couple books on the topic. It is often suggested that beginners with a technique use known writing, such as their own, to get familiar with the results produced. Here is the advice from the authors of one form of authorship analysis:

Starting with Yourself: Making a Graph of Your Own Writing or Speaking

Before attempting to use ______________ for any specific problem or attribution, it is always a useful — indeed necessary — exercise to become familiar with how the technique works. If the method is to be used, the user needs to know exactly how a sample of utterance can be attributed, and how it can be differentiated from the utterance of other persons.

The first necessity is to obtain a sample of writing or speech whose origin is beyond dispute: you, the user, need to be sure of who wrote your sample, or uttered the spoken words of your sample since

• the integrity of the text

is of first importance when making experimental use of _____________. The most obvious source of such a sample for all of us is ourselves: letters, an essay or obvious source of other written piece, a talk or ordinary recorded conversation, will provide suitable material.

I have left off the name of the method since I don't want to be accused (again) of shilling for someone's method that is not your own. Suffice it to say that this method has been around for 30 years and, while it has some detractors as all methods do, it is reliable and trusted enough to be admissible in disputed document analysis in British courts.

This is not the only case where the author(s) of a method advise getting familiar with the results of known texts but it is the example I will offer.

If someone devised an attachment for an automobile and they claimed that it displayed the speed of travel, would you not expect them to calibrate it against known units? If you bring home a barometer, it is often necessary to set it against a known unit. Likewise, would you want a physician's assistant to take your temperature with a thermometer that had not been tested and marked against a known standard?

I have also stated that a tenet of the scientific method is to devise experiments that isolate for single variables that affect the outcome.

They do not try to bury the results by throwing out data without a good reason. One such reason is that they are mindful of significant digits in the measurements and the resulting calculations. If a measurement has only two significant digits like 1.5 inches, it doesn't matter what other calculations you make, you can't justify more than two significant digits in the output. Engineers taking physics classes had this problem all the time. Their calculator showed 8 or 16 digits and yet their measurements were rarely beyond 3 significant digits.

But my opinion holds no weight with you because you do not see me as a peer. I'm just part of the rabble, nipping at your heels, apparently. There's nothing in what I write that can contribute to or add perspective. You have a conclusion at the start of your project and you seem to want to bury the reader with walls of text, repeating the same six names whom you claim wrote just about everything of importance in the British Renaissance, nowadays called the Early Modern period by scholars.

Why is it so important to unseat the authors? What is your purpose behind it? Are you like an Oxfordian who feels that the son of a glove-maker in Stratford could not have possibly had the education to write the plays and poems which have such high esteem in English literature? Or are you a Baconian who finds numerology and hidden messages buried in every text like the authors of The Bible Code and a tongue-in-cheek article a bookseller I know wrote called "The Bobbsey Code" in reaction to that book?

You seem steadfast that every study that points to a certain collection of results you derisively call "bylines" are contrived and staged to end up with a certain result. Yet, you should be mindful of this quote which can raise more doubts than followers to your claims and methods:

Every accusation is a confession

James

>188 faktorovich: Your conclusion regarding "furries" cannot be proven factually because for starters there were no furry-like costumes that were worn in the Renaissance.

Nick Bottom would beg to differ.

Keeline, thank you for your posts here, which I for one appreciate very much. In fact, I suspect there is only one participant in these discussions that doesn't appreciate them, but speaking for myself as a linguist who barely knows the rudiments of programming, I find it very helpful indeed to get the kind of pedagogical explanations you have provided in, for instance, >170 Keeline: and >183 Keeline:.

Unfortunately, I don't think it will help Faktorovich much if I point out that https://corpus-analysis.com/ , which you (Keeline) linked to in >148 Keeline:, lists several tools that will create readable output of a CLAWS-tagged text. Including AntConc, which was mentioned in the previous thread as well. I can't believe I wasn't aware of corpus-analysis.com; I have it bookmarked now.

As for "proximity results", not a term I'd ever heard used before. To be honest, I thought she was referring to collocations (sequences of words that co-occur more frequently than they would do if it were random.)

-Linnéa

>205 faktorovich:
Earlier you said "stylo cannot even manage a basic pre-programmed experiment without destroying the data fed into it, and it can only accept files that are posted on websites, and not uploaded into its system." and "Based on my previous testing, R/Stylo has a glitch that blocks usability of their subdirectories, as it looks like files are added to the correct subdirectory, but Stylo does not see these files, or produces numerous errors and only recognizes files when they linked to from websites."

To explain your lack of being able to use the program. Your statements are factually incorrect. Stylo saw my files with no problems, it did not produce errors, and my files have never been near any websites. It did not destroy my corpus.

Your original gripe was that it was unusable gargbage.

Now you are moving your objection to some set of bugs that doesn't find the correlation that you do. You also say The enormous number of bugs that make these tools unusable and uncheckable is the reason specialists in this field have been able to avoid being discovered manipulating results or providing fraudulent results

I think that is an extraordinary claim and you cannot back it up. The source code for stylo and for R are freely available. For example I cannot say whether a particular test is a good one or a bad one for a corpus (I don't have the knowledge or experience in the field) but being a software developer I can look at the code and assess its quality and tell if the particular function has been implemented poorly (with loads of bugs) or well. If you look at the source code you can see where bugs have been eliminated as people (computational linguists and developers and the original authors) have done this.

Of course that doesn't mean there aren't still some very subtle, very minor, bugs in the code but there are not major ones like you describe. Those get found out real easy and stomped on.

This would be like learning a new language without a single dictionary or grammar book to describe its rules or definition
You mean like we all learned our first language? I certainly don't remember being sat down with a dictonary and grammar book when I was 6 months old and starting to speak English. I may have used a dictionary in primary school for some difficult words but not that much. I certainly didn't use a grammar book in primary school or secondary school.

The only reason you can be making such a claim is either it is a language you have made up, and have failed to make a manual to allow anybody else to learn it; or you are simply making an untrue exaggeration designed to make yourself appear superior.
Well that isn't completely fair.

1) I literally used 3 commands in R. One to install stylo. One to say that I wanted to use the stylo library. One to run stylo with the default settings. It is hardly rocket science. The rest, creating the corpus directory, copying files, looking at the results was all done outside of R.
2) R and stylo do have documentation
3) There are youtube videos showing you how to do simple stuff in R and stylo
4) Even if I had to write my own functions in R, then the language is not completely out there on its own, it is not some weird thing that shares no similarity with other languages.

>205 faktorovich: The absence or insufficiency of coherent manuals or "documentation" is one of the ways this un-usability has avoided detection, as nobody can learn something without any guide as to how it works.

There's an R for dummies book, for crying out loud. For Dummies, Manning, O'Reilly, Packt and Chapman & Hall each have multiple books on the subject. You're talking about one of the most solidly documented programming languages. Whining about documentation is simply an excuse.

If one wants to become acquainted with Stylo for R, these research papers may have interest. I was able to download the PDFs for free with an account on the site so trust others can too.

https://www.researchgate.net/publication/313387787_Stylometry_with_R_A_Package_f...

https://www.researchgate.net/publication/345358488_Stylistic_palimpsests_Computa...

As I have time I will read them more carefully to see if they are as helpful as they appear to be on the surface. I have loaded them on my iPad for offline reading.

James

>209 prosfilaes: A better example is if you come into a store with your own scale that has been giving you precise data for years. You have weighed your two grapes with your own scale. Then you come to the counter and the attendant weighs them and tells you they are "five pounds, and two grapes have always weighed five pounds." Other attendants come up with their own rigged scales and they all insist that they have "calibrated" their scales and according to all of them the 2 grapes are "five pounds". And they insist that you must "calibrate" your scales, so that on your scale the 2 grapes would also weigh exactly "five pounds, or the 'correct' amount that has been agreed to by all of the attendants".

>218 Keeline:

The README file over on Github has detailed installation instructions (including a few caveats for mac users).

There's links there to Howto pdf files that tell you what the functions do and how to use them.

It also links to a three-part set of instructional videos by Maciej Eder, the main developer behind Stylo. It shows you how to use the tool, and what the various features and settings mean.

1. Introduction to the package 'stylo': first steps
   
2. Introduction to the package 'stylo': installation
   
3. Introduction to the package 'stylo': basic parameters

There's more useful tips there, but I'll end by including a link to the Computational Stylistics Group, whose projects page lists a few small-scale test projects you can do to get comfortable with the software.

>210 Petroglyph: For the purposes of tagging words, it is absurd if the corpus of the words you are tagging includes any words more than once, as you only have to tag each unique word once for it to be automatically marked with the specified tag when a program is applied to any corpus fed into this system for tagging. So to say you have a corpus of 100 million tagged words, when you actually have tagged a million words or less, and the rest are repetitions of the same words in different texts is absurd. A tagged corpus is basically a dictionary of all the known words with tags identifying their categories; so it would be absurd if a dictionary had annotations next to words that some words appeared 1,000 times in the texts consulted while making the dictionary, while others appeared only 5 times etc. Using the 100 million figure is just designed to make the dictionary seem enormous, even if the actual number of tagged words might have been far less than the 1 million actual English words that exist in some dictionaries that could have been tagged. Tagging 1 million words would take a lot of labor, tagging 100,000 words, while making it sound as if you have tagged 100 million words takes a lot less labor, and sounds like a much bigger achievement.

I specify that I tested 284 texts with 7.8 million words between them because all such studies cite this figure to indicate the size of the corpus, and my study has more such words than all previous computational-linguistic studies from the Renaissance. I do not say that I have tagged 7.8 million words, or that there are 7.8 different words in the corpus. Instead, I have a column in one of the GitHub tables that lists the number of words in each text, and this column adds up to 7.8 million words. While an attribution study does not need to cover more than 284 texts with 7.8 million words to reach a firm attribution even about this highly collaborative Renaissance corpus, a tagging system absolutely should include more than 100 million words from texts if its goal is to create a dictionary of as many unique words as possible, and in fact it should not be looking at texts, but rather focusing on dictionaries that parse out the 1 million or so unique English words. A corpus can have 100 million words in it, but only 70,000 unique words, so using a fragment of texts is a nonsensical approach to such tagging.

Your barrage of personal attacks in your conclusion reaffirms that you have no logical basis for your tirade, and you are just howling at the moon with nonsensical rage.

>219 faktorovich: Arrogant much?

>212 Petroglyph: You appear to have found Percy's linguistic quirk. He might also have used this quirk in this particular text to signify a unique way of speaking for Capulet, as two of these fragments are from his lines, and the third is from a Servant. If this was a common problem that would influence the results of a statistical attribution method, you should find far more than 3 words that are influenced by it in any given text with around 30,000 words. So, go ahead and check this text and let us all know if these are the only 3 cases, or if there are more instances how many there are. If this glitch impacts 3 words out of 30,000 that's .01% of the text. If 10% of this text are verbs (3,000) then 3 verbs out of these 3,000 make up only .1% of all verbs, so this would make absolutely no impact in the attribution of this text. And if this glitch had occurred in 10% of all verbs in the text (which is an absurd concept to imagine) then it would an extreme linguistic (poker-like) tell that would skew the data towards a more precise attribution of this text to Percy (if he overused this verb variant); again a grammarian might look at the results and question if the verbs are interpreted correctly, but if they are tagged consistently as the same tag-X versus tag-Y etc., it would be a strong indicator of authorship.

>219 faktorovich: ???

Do you...not understand what calibration is?

>221 faktorovich: you only have to tag each unique word once for it to be automatically marked with the specified tag when a program is applied to any corpus fed into this system for tagging

I'm far from expert at this sort of technology, but this claim looks incorrect on its face. Don't homographs complicate the matter?

>224 amanda4242: She's made quite a few posts to demonstrate that she does not.

>213 Keeline: You are saying that the 100 million words claim is similar to Oxford dictionary's claims about having "273,000 headwords; 171,476 of them being in current use, 47,156 being obsolete words and around 9,500 derivative words included as subentries." Thus you are suggesting that there are similarly "100 million" headwords, with the other variants within this system. This is entirely different from claiming there are 100 million words in a corpus of texts that had been tagged to create possibly only 100,000 tagged words. By creating this comparative parallel you are thus exaggerating the claim to make the 100 million figure sound even greater and all-encompassing.

To defend my accusations against you guys regarding your advertising of Stylo; you are using "OpenOffice" being free as a defense. This is like if I used the argument that air is free, when questioned why I have blocked water access to the city, and started charging all residents to get water from my pump.

My objection is not that programming is not easy, but that the tools you recommend are unusable even for an advanced programmer because the programs have glitches in them. And in cases where a program can generate tagging without major glitches (at first glance), it requires a programmer to create a program to put the data into a usable format. My computational-linguistic method is designed so that I do not have to help users who want to follow my method at all. They can use free tools and the steps I have explained for free in this discussion. If you method was just as good, it should be at least as easy to use your method. And instead, as you keep repeating, all who use your method must be computer programmers. The number of people in this field has been shrinking: https://datausa.io/profile/soc/computer-programmers#:~:text=The%20number%20of%20....

I have demonstrated my method works for texts, but there is no such thing as "texts of 100% known authorship" because any text with a firm byline could have been ghostwritten, especially if no previous study has compared it with a significant portion of texts with other bylines in a given genre. Any method that relies on any line as being "100%" known simply because it is placed on a text is biased in its core, and cannot produce accurate results.

It is a plagiarism to cite a source without giving credit for where it is from. Somebody can search for words in this fragment to learn the intended method, so you are not hiding anything by leaving out the name; you are still advertising this method.

And I have tested my own texts, and they do show that they match each other linguistically. I have tested texts from several different centuries, like the dozens of texts I tested from the 18th century, and several from the 19-20ths centuries. I have established that it works. My results for the "Lunch" experiment mostly matched the results Petroglyph had, with the exception of the ghostwriting intersections, co-authorships my method spotted that Petroglyph's analysis ignored (because his data was far less precise, and because he just ignored the same abnormalities when they did come up). Since most of my results matched the established bylines for this "Lunch" experiment, this should have been sufficient to stop all of you from repeating the criticism that I haven't checked it against known "bylines". But your point is that all computational-linguists must manipulate their data to make it look as if nearly all known bylines are correct; this is unethical, and yet you keep repeating it with progressive loudness.

"Renaissance" is the correct term because Harvey's earliest published books under his own byline discussed the revival of Greco-Roman classics into new variants in English, and he is likely to have taught Percy and Jonson ("Shake-speares") these lessons that they applied in their adaptations of Greco-Roman dramatic rules both to Greco-Roman and English subjects. The revival of ancient knowledge via its publication for the first time in printed books was essential for starting a global education process.

Scholars have been questioning (adding and subtracting) authorship claims from the Renaissance until the present moment. A significant portion of all scholarship has been about determining authorship. There is nothing new about my work to "unseat" any "authors" from bylines. Yes, most previous attribution studies have been wrong because they have focused on a byline's education-level, or hidden messages. That's why most of the texts were initially anonymous, but most of these have since been assigned bylines by scholars. I have established precise credits for the Renaissance's texts with a systematic quantitative method. There is nothing "hidden" about a "Nashe's Ghost" byline. The evidence I am finding in the texts and documents is not hidden, but it has been ignored by scholars who have been influenced by their intuition, whereas I am driven by data and facts to arrive at the precise, proven truth.

"Every accusation is a confession" is something a dictator would say as he threatens a reporter who is accusing him of corruption of finding dirt on him instead. As I stated previously, I have absolutely nothing to hide. So every accusation I make is based on firm facts about those who I am accusing; and there is nothing confessional in my statements about myself.

>216 andyl: I have already proven that Stylo/R has bugs in its system that make it unusable for attribution. Just refer to my previous explanation. Unlike you, I am not going to repeat myself.

>224 amanda4242:
>225 paradoxosalpha:

She does not. At all. Like, the whole point of it has passed her by.

>225 paradoxosalpha:

Yes, a record vs to record; do as a main verb vs do as an auxiliary; better as a comparative adjective vs better as a verb. Discriminating between those often requires looking at the POS of the surrounding words: the noun record will often have a determiner (a or the) in front of it; the verb will be surrounded by adverbs, nouns, or the infinitival marker to, to give an example.

Also, obviously, you don't tag a corpus with POS just to get every unique word assigned a tag. Things like that already exist, and we call them "dictionaries". That is part of their raison d'être.

In fact, tagging software often makes use of such a pre-prepared lexicon -- a list of words that have been tagged with the correct tags (the two records and betters etc. will have been included twice there, with the relevant tags). When tagging large portions of text, the software will treat the lexicon as a lookup table.

And before Faktorovich chimes in with her utterly ignorant drivel: this is how analyzemywriting.com does their tagging. They use a particularly unsophisticated one (e.g. all instances of have and do are marked as auxiliary, regardless of actual syntactic context). But that's how they do it.

There are other methods, but they are more complex and I don't feel like explaining them right now.

Anyway. You tag all the words in a corpus with parts of speech so you can look at things like passives (the books were gathered, you got served, the collection that was looked through) -- which, in English, is a particular combination of a few auxiliaries and a participle. You really can't do that in any other way over large quantities of text.

POS tagging allows you to use corpora to look at things like the number of adjectives before nouns; the alternation between attributive (a harsh complaint vs the complaint is harsh), adjective complementation (e.g. accustomed to luxury vs accustomed to leading a life of luxury vs the life they were accustomed to lead -- the distribution between noun, -ing form and bare infinitive); the difference between This is obviously wrong and Obviously, this is wrong (i.e. from verbal adverb to sentential adverb). And thousands of other bits of variation and language change that involve a syntactic, collocational or contextual component. (Hint: that is all of them.)

You can't do that unless you have millions, or even hundreds of thousands of words tagged.

Like, that is the entire point of doing this. Just tagging every type/unique word once is... absolutely ridiculous? Like, who in their right mind would even think that that's how things work, or how they are supposed to work?

Faktorovich does not know what she is talking about, at all. But she'll sure try to sound knowledgeable.

>218 Keeline: Many books can be written about a subject without the thing being described being as good or usable as the propaganda-books insist it is. I recently watched a documentary about the Marcos family in the Philippines sponsoring a Facebook campaign to puff their achievements to make them sound like saints, after they have been found guilty of stealing billions from the country's citizens, as one of them is trying to become President yet again. I checked the available manuals and queried everybody here who is advertising for Stylo, and there was no possible way of making this tool usable for an actual attribution analysis. Even if all of the bugs were solved in this broken system, and it actually worked; my 27-tests method achieves 27-times more than the 1 word-frequency test that it at most can perform, as I have not seen any studies that use anything beyond the word test in Stylo.

>220 Petroglyph: You are now advertising hundreds of Stylo users' projects without giving any actual tutorial links to how to perform even a basic word-frequency test in Stylo. If there were such instructions posted somewhere, you would have linked to them by now. The "first steps" such as "installation" does not at all help somebody who needs to get an actual result out of this program. Saying that a lack of glitches in an installation (and there were some glitches even during installation) is sufficient to prove a program works, is an absurdly low bar.

>225 paradoxosalpha: Yes, homographs is one of many glitches that can occur during tagging that might miss-categorize isolated words in a text run through a tagging system. Some glitches occur in all systems, but this group has been attempting to argue that Analyze's tagging is inferior because they have found glitches in its results, when they have not performed the same identical tests in Stylo or the other programs they have been advertising to see if they would be any better at spotting homographs' intended meaning etc.

>228 Petroglyph: If you have 2,000 sentences with half that include one of these phrases "accustomed to luxury vs accustomed to leading a life of luxury". The 1,000 example of this phrase is not going to be any more useful at explaining the proper tagging of a word in relation to the words next to it. There are grammatical rules behind word-type-order that would establish the required tagging far better than if a computer is fed examples. Only if a human interferes and reviews these 100 million words to see if they are tagged correctly, and edits the cases of tagging by programming each special case with how it was supposed to be tagged would the presence of this large corpus make sense. Otherwise, the system just needs to know what word-order is most likely to lead to tagging with X vs. tagging with Y.

>231 faktorovich:

Analyzemywriting.com uses a simple lookup list for its tagging. Stylo does basic text comparison (and some machine learning techniques as well). It is not a tagger, nor is it meant to be.

Are all pieces of software just interchangeable to you?

this group has been attempting to argue that Analyze's tagging is inferior because they have found glitches in its results

We've been arguing that AMW's handling of homographs is non-existent. Have and do are always categorized as auxiliaries, despite being main verbs sometimes.

Here is what AMW makes of "a record. to record" (so, noun and verb):



It only categorizes record as a noun.

Here are the results for "better results. to better your results" (so, a comparative adjective, and a verb):



It only categorizes better as an adjective.

Conclusion: AMW uses a lookup table that completely ignores homographs. And you have based all your POS tests on this.

Professional taggers don't do this. Look at the list of categories that CLAWS uses in >170 Keeline: "AJ0" = Adjecive; "AJC" = comparative adjective (e.g. better); "AJS" = superlative adjective (e.g. best). That level of detail is unachievable with AMW.

they have not performed the same identical tests in Stylo or the other programs they have been advertising to see if they would be any better at spotting homographs' intended meaning

It is not my responsibility, nor anyone else's, to convince you that better alternatives exist. It is not my responsibility, nor anyone else's, to demonstrate to you that things more sophisticated than AMW yield more reliable results.

As you've amply demonstrated in this thread, you will ignore criticism, smear it (and the people who offer the criticism) with knee-jerk accusations, and remain entrenched in your narrow comfort zone. Even if I went to the trouble of comparing TreeTagger or CLAWS with AMW, you'd simply remain unconvinced that your home-brew tinkering was in any way inferior to professional tools with decades of development behind them.

>232 faktorovich:

You don't tag in order to "explain the proper tagging of a word in relation to the words next to it". The tagging is not done just for the purpose of tagging more.

Tagging is done in order to do things with the results of the tagging, after that process has been completed. So you can extract all the hits of a relevant set of grammatical structures and constructions you want to investigate. Looking at "accostomed to":



The top 54 hits are of the type "accustomed to the NOUN PHRASE"; then follow a bunch of "accustomed to INFINITIVE". A properly tagged corpus would be able to give you the proportions of these two structures; it would allow you to search very precisely and directly. Looking at corpora from different time periods, this would allow you to track the changes in distribution between "accustomed to" + noun / infinitive / ing-form. (spoiler: the proportion of bare infinitives is way, way down these days compared to two, three hundred years ago; -ing forms are on the rise.)

Only if a human interferes and reviews these 100 million words to see if they are tagged correctly, and edits the cases of tagging by programming each special case with how it was supposed to be tagged would the presence of this large corpus make sense

You, who until recently knew nothing about POS taggers, think that a problem you literally just thought up has not been addressed? In the sixty years of development behind these algorithms?

>230 faktorovich:

Lies, deflection. Retrenchment in your comfort zone.

Your comfort zone is not the measure of research. Nyarlathotep be praised.

Hey, Faktorovich.

Instead of using strikethrough across William Shakespeare's name on your covers and title pages:



And instead of using scare quotes all the time ("William Shakespeare") in order to indicate that this attribution is passé.

Have you ever given thought to using Wouldiwas Shookspeared?

Please let me know your opinion ASAP. If you decide to use this name, I want to be cited.

>230 faktorovich:

You sound like one of those whiny students who post to forums asking people to give them answers for their homework.

Getting a word frequency list is really easy. You can even find out how to do it in the non-existent documentation https://rdrr.io/cran/stylo/man/make.frequency.list.html

The above link was the second result when I googled for "R/Stylo word frequency"

>237 andyl:

Run any analysis through Stylo, and it will create one automatically. (The frequency lists is what many of the subsequent analyses are based on).

This frequency list will be dumped in the working directory as "table_with_frequencies.txt"

Faktorovich knows this: she asked for these files way back in the Oz portion of this thread. I provided those files to her then.

>238 Petroglyph:

Yeah I know. That was relative frequencies, maybe she wanted absolute counts. I don't know if that is configurable in the Tk GUI that pops up - I run R on a machine without any graphics.

But my comment also aimed to point out that there is very good documentation, and how a simple google search finds it in the first responses.

At the moment if she told me it was raining I would have to look outside to check.

>238 Petroglyph:

I see. My mistake, carry on!

(yeah: as R goes, stylo is very well documented)

>239 andyl: At the moment if she told me it was raining I would have to look outside to check.

I've been thinking much the same thing for months, but worded it as "If she told me water is wet I would seek independent verification."

>233 Petroglyph: I am still waiting for one of you to use the tools you are recommending to create anything like a column that measures the percent of nouns, etc. in a text that would actually move the process through the first step of measuring the percentages of word-types. You have given a tagging service, and have managed to create a program to add these up, but you haven't even attempted to add up all of the nouns categories together in your data; and we are nowhere near a simple one-stop tool that is free and can achieve a count of word-types without the user having to write a program of their own.

And as I said, the method has to have mathematic consistency and not exact grammatical accuracy. Always measuring "record" as a noun, even when it is a verb is not a quantitative problem if the question is not how many nouns are in a text, but rather how many words in a given text are in category X vs. category Y etc. In other words, it would be like counting the frequency of words in a category that makes up 10% of a text (more or less) as opposed to merely counting most frequent, or medium frequent independent words. Grouping them by word-type achieves a uniquely accurate measure of linguistic style. The failure of a tagger to distinguish nouns and verbs that are spelled identically to each other is not relevant as a given term like "record" is always consistently tagged, and thus still identifies the user as somebody who prefers to use this term over other options. If it bothers you greatly, you should be writing to the programmers of this system to ask them to volunteer to help them for free to fix this and other glitches you have noticed. Analyze is the best and most relevant tool I have found, and until you have a rational alternative, you are just criticizing something that works, while advertising tools that you haven't even explained yet how they could work to achieve the same result (or if your alternative might not generate the same glitches: you have not being testing the same words with both systems to check if there is a difference, or if programmers have just not managed to solve these problems yet).

Distinguishing between "comparative adjective (e.g. better); "AJS" = superlative adjective" does not require the same type of complex analysis about word-order, or neighboring words involved in the recognition of differences in meaning between "record" when it is used as a noun or a verb. There is no category for "homographs" in the list CLAWS offers; so it does not solve the central problem you have described in this post. So while it seems as if there are more adjective types; an ideal system is not one that distinguishes the highest number of word-types, but just one that captures the larger divergences between texts and presents data in a format that is applicable for all non-specialist researchers.

>234 Petroglyph: While there are some linguistic studies where finding all occurrences of a phrase + infinitive might be relevant; it is entirely irrelevant for a computational-linguistic method that is interested in the relative number of broad word-types in the text to establish the characteristics of a linguistic signature. It would also be just as efficient to search for all instances of "accustomed to the..." in the text and check what follows, versus creating a program that categorizes these and then still reading through a long list mostly with single-appearances of each uses per-text.

>226 faktorovich: I have demonstrated my method works for texts, but there is no such thing as "texts of 100% known authorship" because any text with a firm byline could have been ghostwritten,

On one hand, all data has errors. You try to minimize them and hope they don't affect the project much. On the other hand, you're saying that most works have ghostwritten, in which case, we're at square one. Show me some evidence that any of this is in any way reliable for detecting authorship. There's no evidence that people have a consistent writing style distinguishable from others, especially those trying to mimic it. All previous tests were done on texts that were presumed to be of 100% known authorship, probably with tools like R/Stylo.

If you method was just as good, it should be at least as easy to use your method.

Which is, of course, why we produce paintings with crayons; if oil paints were better, they would be as easy to use. From my perspective, your method is terrible to use; you have to manually do all this stuff and copy results back and forth, whereas with Stylo I create my corpus, open the program and hit go.

"Renaissance" is the correct term

No. Whether you call it the "Early Modern period" or the "British Renaissance" is not a matter of correctness; it's a matter of labeling and communication.

Scholars have been questioning (adding and subtracting) authorship claims from the Renaissance until the present moment.

Now you claim you're doing nothing remarkable? Had you merely claimed that six ghostwriters had done almost all of the literary work in the time period, that would be a pretty audacious claim. It might not have got me coming back to this thread, because it's conceivable. Six ghostwriters wrote all the writing in the time period, including that by people like judges and preachers, whose job was in part to write, plus one guy wrote both the Protestant and Catholic pamphlets and translated the whole KJV? That's preposterous.

There is nothing "hidden" about a "Nashe's Ghost" byline.

No; there's no evidence that anyone has ever looked at it and assumed it was Nash, either. The fact that someone used a pseudonym once doesn't mean that every name is a pseudonym.

>236 Petroglyph: The "Shakespeare" attribution is not "passé"; it is inaccurate. I use the quotation marks to indicate derision and the falsity of this attribution-claim, while mentioning the byline that is currently broadly accepted as accurate for a given text.

Meanwhile, you have started pirating front-matter pages with images from BRRAM. To avoid a blatant piracy violation, you at least have to cite my name and the source for the image you have copied; the operators of the museum that makes this image available specifies that a credit that it came from them must be provided with the image, and you have not provided this credit.

>237 andyl: Word-frequency is indeed easy, as it can be generated with the program I cite in my book that generates a list with frequencies for all words in a text. There is no need to create a new program to count frequencies. We are not discussing general word-frequencies in the last few posts, but rather word-category frequency (nouns vs verbs). And your "documentation" does not provide the "easy" steps for how to create a frequency list, but rather just summarizes that it can be created. Maybe you should have kept searching beyond your second googled result.

>244 prosfilaes: I am not saying that "all" or "most" of all texts ever written were ghostwritten. I have not checked nearly enough texts to reach such a conclusion. I have checked enough texts to prove that all, or at least all of the texts I checked from the British Renaissance were written by six ghostwriters. This is a statement of fact based not only on the quantitative data, but the other evidence that you are all working to ignore, as if it is not the overwhelming proof that guarantees my method has reached precisely correct attribution results.

"There's no evidence that people have a consistent writing style distinguishable from others, especially those trying to mimic it." If you are so biased that you believe this to be a fact, you should not go anywhere near authorial attribution, as you do not believe such attribution is possible, and to achieve something you have to at least imagine that you can succeed in an endeavor. "All previous tests were done on texts that were presumed to be of 100% known authorship". You are basically saying that all previous experiments have been designed to repeat the established byline, without any hope that they can actually distinguish between linguistic styles; this is exactly why all of these previous methods have failed to arrive at anything approaching accurate results.

Then, you are saying that if I had made a smaller claim by crediting fewer works to these ghostwriters, it would have been "conceivable", so you would have ignored my findings and not commented on them at all. But only because my claim is so humongous that you cannot "conceive" it can be true, that's why you are engaging in this conversation, or because you have assumed my claims must be false, and you have assumed you will easily win the argument? My conclusions are not "preposterous", or contrary to reason, as they are instead purely rational; they might only appear unbelievably irrational to anybody who believes to dogma of the established bylines. Similarly, the idea that humans evolved from apes appeared "preposterous" to religious folks when Darwin pitched this idea, and it still appears "preposterous" to some Darwin-doubters today; but science comes up with a true answer, and not one that sounds believable based on what people already believe.

>247 faktorovich: "There's no evidence that people have a consistent writing style distinguishable from others, especially those trying to mimic it." If you are so biased that you believe this to be a fact,

You don't understand science. The evidence that people have consistent writing styles is based off certain tests. If you don't believe those tests were accurate, then you're assuming something in your tests that you don't know to be true. Same thing with history; you assume it's wrong in so many ways, but you assume it's right when you want it to be.

My conclusions are not "preposterous", or contrary to reason, as they are instead purely rational;

Purely rational things are usually wrong in the real world. The truth of the real world does not come from arguing like Greek philosophers; it comes from poking at things, and trying to not let your preconceptions get in the way of what they're telling you.

they might only appear unbelievably irrational to anybody who believes to dogma of the established bylines.

I see you didn't read what I wrote. I don't know who translated the KJV; I do find the idea that some exiled author did the whole thing a lot more preposterous than a bunch of scholars. I don't know any of the religious authors bickering back and forth, and given religion and the law at the time, I wouldn't be surprised if many of them were written under pseudonyms. But the idea that there was one person writing everything for all sides... why? It's not even like most of that stuff sells well. Much of it is the 17th century equivalent of forums like these; people arguing back and forth over stuff that most people consider trivial, to be heard and to oppose the idiots on the other side.

The issues is not about any dogma of the bylines; it's that people don't behave that way.

Darwin

Actually, Huxley wrote all of Darwin's works; Darwin wrote under the name Adam Sedgewick, where he called the Origin of Species "A cold atheistical materialism".

science comes up with a true answer, and not one that sounds believable based on what people already believe.

"But the fact that some geniuses were laughed at does not imply that all who are laughed at are geniuses. They laughed at Columbus, they laughed at Fulton, they laughed at the Wright Brothers. But they also laughed at Bozo the Clown." -- Carl Sagan

And now for something completely different, a purely rational conclusion.

>249 amanda4242:

Or the first few minutes of this sketch.

>245 faktorovich:

you have started pirating front-matter pages with images from BRRAM. To avoid a blatant piracy violation, you at least have to cite my name and the source for the image you have copied

You mean this post?



The post where the image contains the title (highlighted in a yellow box), that starts by explicitly addressing the author (you, highligted in blue), that explicitly assigns the book to you ("your covers and title pages", highlighted in green), and that points out a noteworthy feature (highlighted in orange) that uniquely identifies Faktorovich's BRRM. That post?

How long do I have to keep that post up before your writing starts to suffer from the emotional distress (as in >114 faktorovich:)? How much money do you think your emotional distress is worth in a court of law?

And, more importantly, what do you think of "Wouldiwas Shookspeared"? You never answered that one.

Do cite me if you decide to use it!

>243 faktorovich:

To paraphrase: "why use precise tool when blunt tool do trick poorly?"

It would also be just as efficient to search for all instances of "accustomed to the..." in the text and check what follows, versus creating a program that categorizes these and then still reading through a long list mostly with single-appearances of each uses per-text

Your manual, one-text-at-a-time mindset is showing again.

Take a look at this:



This is an overview of the collocates of the verb kiss (not the noun kiss!), specifically all the collocates to the right. They have been sorted by their POS (noun, adj, verb, adv).

"reading through a long list mostly with single-appearances of each uses per-text" my ass.

You've never really performed any corpus analysis, have you?

>242 faktorovich:

I am still waiting for one of you to use the tools you are recommending to create anything like a column that measures the percent of nouns, etc. in a text that would actually move the process through the first step of measuring the percentages of word-types.

Every time you ask us to do your homework for you (thanks for that comparison, >237 andyl:), the list of demands grows longer. Now you are "waiting" for us to visualise the results in a bar chart, as well? And it should resemble AMW, apparently, too.

You've been pointed to the tools. Their workings have been explained to you. You've been given the correct search terms. Good luck!

Distinguishing between "comparative adjective (e.g. better); "AJS" = superlative adjective" does not require the same type of complex analysis about word-order, or neighboring words involved in the recognition of differences in meaning between "record" when it is used as a noun or a verb.

Distinguishing between better (comp. adj) and better (verb) does require looking at adjectival behaviour vs verbal behaviour. There's 62 different categories in CLAWS5. They're not just there to make finegrained-but-useless distinctions. They're there to permit future fine-grained linguistic analysis. That's why comparative and superlative adjectives are separated out.

while advertising tools that you haven't even explained yet how they could work

Do your own homework. Do you expect us to do the work for you? Aren't you the person who sees plagiarism everywhere?

There is no category for "homographs" in the list CLAWS offers

I lol'd at the child-like misunderstanding of that. Ah Faktorovich. You are guaranteed to misunderstand the world in a very unique way.

Always measuring "record" as a noun, even when it is a verb is not a quantitative problem if the question is not how many nouns are in a text, but rather how many words in a given text are in category X vs. category Y etc.

Right. Because when measuring the proportions of nouns, verbs and auxiliaries and so on, it doesn't matter if all homographs are systematically only assigned one category. If that is your stated opinion.

You do know that there are literally thousands of these in English, right? Thousands of zero-derivation nouns/verbs: list, support, kiss, record, order, release, find, use, bridge, work, holiday, move, bank, man, shut, answer, knife, skin, fish, wound, sound, paper, freeze, mask, make, ...? All of these can be a verb as well as a noun (to paper (a wall) vs a paper; man the ship vs a man on the ship). And then there's verb/adjective pairs: open, close, dirty, clean, warm, ....

How is this not a massive issue? How is "I tried command-line R once and I couldn't get it to work now I'm gonna whine online how it's all the bugs that make the thing unusable and its just more corrupt academics shilling their version of history" such a massive issue for you, but the systematic shortcomings of your own methods aren't?

You're never gonna admit you're wrong. Like a toddler who denies drawing on the wall.

>246 faktorovich:

it can be generated with the program I cite in my book that generates a list with frequencies for all words in a text. There is no need to create a new program to count frequencies.

Ah yes. WordSmith. A corpus-linguistic tool that does Key Word In Context, collocations, concorcances and a few other things. Oh, and word frequency lists. Which is all that Faktorovich uses it for, and so there's "no need to create a new program to count frequencies" As though the point of the programme is to count frequencies and then do nothing with them.

Such a narrow slit to see the world through.

>250 Petroglyph: Ha! I haven't seen that one before. Thanks!

Requoting >219 faktorovich: in full since it so perfectly encapsulates every single one of the author's posts:

">209 prosfilaes: prosfilaes: A better example is if you come into a store with your own scale that has been giving you precise data for years. You have weighed your two grapes with your own scale. Then you come to the counter and the attendant weighs them and tells you they are "five pounds, and two grapes have always weighed five pounds." Other attendants come up with their own rigged scales and they all insist that they have "calibrated" their scales and according to all of them the 2 grapes are "five pounds". And they insist that you must "calibrate" your scales, so that on your scale the 2 grapes would also weigh exactly "five pounds, or the 'correct' amount that has been agreed to by all of the attendants"."

Could it never occur to you, in this hypothetical situation, that you've been weighing grapefruits all this time, and calling them grapes, faktorovich?

>248 prosfilaes: What you term as "arguing like Greek philosophers" is just how a rational argument is supposed to sound by the rules of rhetoric and logic, as described by writers like Harvey in his rhetoric textbooks.

Again, KJB was primarily ghostwritten by Verstegan, but he had help from Harvey as a co-writer, so he was not its only translator.

You are asking: Why would Verstegan have written propaganda for both sides under a monopoly on theological book publishing? Imagine if you had a monopoly today on publishing all textbooks, theology books, non-fiction books, the Bible (all versions). Even if you only sold a single copy of most of these books, you would sell so many copies of hit titles like the Bible, you would make enough profits to publish numerous other books. And you would receive ghostwriting payments from hundreds or even thousands of "authors" who needed byline credits to advance in their careers. And if a book was accusing somebody of scandalous things that book would sell more copies, and would prompt the opponent being slandered to need to hire you as a ghostwriter to publish a rebuttal. Yes, people behave this way; there have been plenty of ghostwriters who have asked for credit, like Trump's ghostwriter. You don't think there are people in the media today that threaten publishing revealing stories if somebody doesn't pay them to keep them quiet? Just the "National Enquirer" has been accused of extortion to cover-up scandals for Bezos, Trump and others: https://www.theguardian.com/technology/2019/feb/07/jeff-bezos-national-enquirer-... This is what humanity has been since at least the Renaissance; so solving current problems begins with acknowledging the existence of these problems in the Renaissance.

It is a funny history that six ghostwriters wrote the Renaissance; scandal is funny; discovering scandal where it seemed unlikely is funny; laugh if you are cheerful.

>251 Petroglyph: You might have semi-cited my name, but you have forgot to cite the image that you have left as the only item un-circled in your diagram. You see the image belongs to a museum that allows for its reprinting, only if the printer gives credit to the museum for holding the artwork and creating the digitization. So, the question is not if I am emotionally distressed in this matter, but rather if the museum might go after you for copyrights infringement; I have informed you that you have failed to give them credit, so they would be able to argue that you were fully aware of their request for a credit, and intentionally repeated the piracy after being informed. I don't even have to spend any time on this lawsuit as the museum can carry it out with their own lawyers.

>252 Petroglyph: You had to have searched an enormous corpus for their to be 1813 occurrences of the words "kiss" and "cheek" together. In a single 30,000 word text, there are going to be few instances of a word like "kiss" and so you would have to look through these types of pattern lists for nearly every word in the text. In a single text, there would not be 1813 occurrences that could be jointly categorized as noun vs. verb by checking only a single box in a tagging system that does not have an automated system for distinguishing between a noun and a verb. The 27-tests that I use are accurate for attribution, and the fact that I have performed them is registered in their data on GitHub.

>253 Petroglyph: You are all volunteering your thoughts on my research. The unsolicited homework you have been submitting is atrocious. I have just been investing my time into grading it for free and pointing out areas where you can benefit from improvement.

As I stated before, if all homographs are consistently categorized by a system under the word-category it believes each of them falls into; the overall analysis will still be counting the tendency to use or not to use them by the author in the data. An attribution method is designed to spot variation in style, so it does not at all need to be an accurate grammar-checker. The method I designed works with free accessible tools that anybody can replicate. I will not ask those who use my method to deal with the types of unpassable bugs that I encountered with Stylo. If there was a better counting program out there than Analyze, I would switch to using it; but my Renaissance computational-linguistic study is finished (unless I decide to add texts to it to solve mysteries at a later time).

>256 susanbooks: I see, let me adjust the scenario. You come into the store and try to purchase 2 grapes. You go through the same spiel, but after the calibration objection doesn't work to convince you to let the attendants "calibrate" your scale. The attendant insists, "Did you say that's a 'grape'? NO!!! That is a 'grapefruit'! You don't even know what a 'grapefruit' is! Ha, ha!" "But," you object, "they are the size of my thumb's nail, and they are purple, and you are insisting they are 'grapefruits'?"

>261 faktorovich: And yet everyone else in the world says they weigh 2.5 pounds each. And they, of course, are wrong. Not you. Never you.

>262 susanbooks: Remember, they burned Columbus on a pyre before conceding he was right all the way!

-Johan

>263 anglemark:
Yes, Columbus was correct from the start: The American continents and the Pacific Ocean are hoaxes. And the non-Christians he enslaved had no right to the Indies he discovered.

>226 faktorovich:

>213 Keeline: Keeline: You are saying that the 100 million words claim is similar to Oxford dictionary's claims about having "273,000 headwords; 171,476 of them being in current use, 47,156 being obsolete words and around 9,500 derivative words included as subentries." Thus you are suggesting that there are similarly "100 million" headwords, with the other variants within this system. This is entirely different from claiming there are 100 million words in a corpus of texts that had been tagged to create possibly only 100,000 tagged words. By creating this comparative parallel you are thus exaggerating the claim to make the 100 million figure sound even greater and all-encompassing.

I have seen an estimate of 1 million words in English from a group in Texas.

https://www.macmillandictionaryblog.com/one-million-words-of-english

The number seems a little high since the largest dictionaries contain a little more than 1/4 this figure. Where are the other 722,000 words?

The numbers I cited are the numbers of words in the Oxford English Dictionary as cited on the Wikipedia page whose link I provided.

I make no such claim that 100 million words has any relationship to 273,000 distinct "headwords".

Since there is obviously some confusion, a corpus is a collection of writings. For the purposes of stylometrics, they are works that have been carefully digitized and edited to be true to the originals.

Let's consider a real example I have access to. If I have a collection of 110 "Tom Swift" books in digital form, that is a corpus. The number of words per volume will vary widely since some books are much longer than others. Typical Tom Swift Sr. books from 1910 to 1935 range between 40,655 and 46,989.

Altogether, this corpus has 1,599,037 words. That is merely the size of the corpus. The number of distinct words is much, much smaller. Making a rough count, the number of distinct words is about 23,618. This may be a bit high because I think some punctuation is getting in the way.

You have claimed that your collection of texts (a corpus) has X million words in it. Yet, I think you realize that the number of distinct words is much smaller. After all, you also claim the high estimate of 1 million words in English.
_____

When it comes to identifying the parts of speech for a word, it is a perilous assumption to think that all word should be counted as the same type. I and others have given examples of words which are spelled and sound identical but have radically different parts of speech identifications.

Identifying the parts of speech is difficult to do. Often it depends on where it is placed in a sentence whether a word like "lead" is a verb or a noun. If you tie your hands by working from a alphabetically sorted list of distinct words in the corpus, without the context, and assume the first parts of speech from a dictionary (or another arbitrary source like AnalyzeMyWriting.com), you will get an unacceptable contribution of noise to the counts of each type of parts of speech (nouns, verbs, adverbs, adjectives, etc.).

If part of your "fingerprint" for an author is based on usage of different parts of speech, suddenly the counts become important that they be as accurate as possible.
_____

In discussing the CLAWS5 tagset, I have already noted how the fine-grained granularity of parts of speech can be combined to count "all verbs", "all nouns", etc. You take the list of tags of interest and count all of them from the CLAWS5 output. Remember when I wrote something like:

egrep "NN0|NN1|NN2|NP0" input_file.txt | ...

The egrep command is like grep that searches one or more text files for the occurrence of a word. With this usage of egrep I have used a simple Regular Expression to get each of the different variants of tags for nouns — NN0 NN1 NN2 NP0 — and asked the system to show me the lines that have any one of those values. If I counted them, they would be an aggregate count of the nouns in input_file.txt. This is not particularly clever. It is merely using the tools available for text manipulation in a Unix-type system (including Mac OS X). If you are using Windows, you might need to look at CygWin to get similar functionality.
_____

As stated by others before. I am not your student, your minion, nor your sycophant. I will try to help with suggestions of tools that are out there.

You are all volunteering your thoughts on my research. The unsolicited homework you have been submitting is atrocious. I have just been investing my time into grading it for free and pointing out areas where you can benefit from improvement.

This forum and its replies are not homework for you to grade and declare to be "atrocious".

The reasons other tools are mentioned is not because of the results they produce which reinforce traditional bylines but rather because the code for something like R and the Stylo extension to it are open-source and have been evaluated by people who like to look for flaws in code as well as people who wish to use them and make them better. Not everyone who uses open-source software looks at the code or tries to change it. But the opportunity is there for someone who knows or will learn to read it or hire someone who can review it. This is part of the reason why open-source software tends to close security holes faster than proprietary software does. There are more people looking at it, looking for problems, and trying to fix them. With proprietary programs and operating systems, you can write in to the company like Apple, Microsoft, Adobe, etc. but good luck getting it to someone who will evaluate the report and initiate a fix if called for. There are problems with some of these programs that stick around for years without resolution even though the companies make large sums in the licenses to use the software.

For example, multi-core processors have been the norm for computers since 2000 or so (1/5 of a century or more). Yet, Adobe Acrobat "Pro" (quotes because I dispute that the expensive program deserves "pro" status; it is more like an amateur program) does not use more than one core or thread for its heavy tasks. If you want to do OCR on a PDF with scanned or photographed page images, only one core is used. On my machine with 6 cores, that means that 5/6 of them are idle and the whole process is slower than it would be if it ran things in parallel. There are open-source utilities to OCR PDFs that do make full use of the available resources. One is a command-line Python script called OCRmyPDF which uses the Tessaract OCR engine. When this is running, you know it because the fans spin up and the load meters show that the computer is working hard. Good programs like this will let you adjust how much of the processor to use.

On my MacBook Pro, which I generally like, there are long-standing quirks such as the Finder's inability (in the past few years) to find file names with certain words in them. Apple changes how things work and sometimes their own engineers don't know the details. It appears that if the full filesystem paths are too long, the storage in the database for the file names is clipped off thanks to a too-short field size in the database. I have other tools (EasyFind) that I can use to get the results but this sort of thing used to work in the Finder. I've gone the rounds with the Apple technical support with lots of demonstrations of what is happening. Eventually they acknowledge there is a problem but won't fix it.

It doesn't matter how good one's word processor is or how much you paid for it. It won't write your novel for you. That is up to the user of the software to learn how to use it and type something worth reading. But I do think that if your hypothesis depends on word counts of parts of speech then you'd better know exactly how they are determined and how accurate they are. Otherwise the whole thing falls apart. Or, in the words of the AnalyzeMyWriting.com site:

Secondly, these indices assume that a text is written in grammatically correct, properly punctuated English. If this requirement is not met and you feed a nonsense text into the program, the famous computing adage applies: "garbage in, garbage out."

https://www.analyzemywriting.com/readability_indices.html

James

>226 faktorovich:

I have demonstrated my method works for texts, but there is no such thing as "texts of 100% known authorship" because any text with a firm byline could have been ghostwritten, especially if no previous study has compared it with a significant portion of texts with other bylines in a given genre. Any method that relies on any line as being "100%" known simply because it is placed on a text is biased in its core, and cannot produce accurate results.

You may have satisfied yourself that you have found the TRUTH but you haven't persuaded the people who are responding here.

My quote from Analysing for Authorship (University of Wales Press, 1996), p. 17, stresses the importance of starting with text you should be able to trust — your own writing. Unless you are quoting someone else, your writing should allow you to determine the traits based on ways of measuring it.

The next level from this would be to take works where the authorship is know with a great deal of confidence and are not in dispute. This can be done, as I originally suggested, by getting writing samples from people now with enough supervision to ensure that they are not copying text from other people to muddy the results.

Your analysis seems to attribute texts to your six authors on the basis that the 287 or so texts are similar (by your measurements) to the writings of these "ghostwriters" in the "workshop" you name.

How can you be sure that your 400+ year-old texts that you attribute to people like Verstigen is really by him? How do you know it was not ghostwritten by another?

You mention "translating" texts, even in your current daily work. Let us see an example of the before and after of a couple hundred words of this process.

In an analysis where vocabulary, sentence length, proportions of parts of speech to the whole, and even punctuation (!!), are relevant to your process, how much is altered in the texts becomes important. Are we detecting traits of Shakespeare or Verstigen or Faktorovich?

Most textual analysis does use a certain amount of pre-processing of texts to make them usable by the software. This could be something like placing each sentence on its own line. Or, if it produces unwanted deviations, proper nouns might be tokenized so a short name, like "Tom" isn't counted among 3-letter words and distorting their proportion in a work. Some procedures call for dropping pronouns if they are more common in one genre than another and you are looking to explore the other parts of the text.

I did read through this article and found that it is a good introduction to using the systems mentioned, Stylo for R. It is not an absolute, hold-you-by-the-hand step-by-step, set of directions but it is the level of detail one sees in academic articles that are introducing / reviewing a process or package. It is just 15 pages plus a cover sheet and nearly all of it is pretty clear. There were some examples I have not seen and will need to learn about but that is normal.

https://www.researchgate.net/publication/313387787_Stylometry_with_R_A_Package_f...

As I stated before, I have a free account with researchgate.net and was able to download this article as a PDF with it. It was a good thing to read while waiting to see if my local court wanted me to serve on a jury. It was more interesting than the home makeover shows that were on the TV. At least the volume was low.

If you were to make a bug report to LT about some function that was not working as you thought the developers intended (not the features you wish they had, that is an RSI), you would be asked to go to the Bug Collectors group (no not insectophiles who will complain about your bug zapper) and describe the problem and detail the steps that would allow the developers to reproduce the undesired behavior. Since you have not done this for Stylo for R, I can't agree that you have

>227 faktorovich: I have already proven that Stylo/R has bugs in its system that make it unusable for attribution.

Perhaps you are thinking of these posts? Or is there another?

https://www.librarything.com/topic/337240#7674605
https://www.librarything.com/topic/337240#7674656
https://www.librarything.com/topic/337240#7676974

I know you think you have but you really quit too soon. It's like sitting in the driver's seat of a rental car and not realizing that it is necessary to depress the brake pedal before you will be allowed to turn the key for the ignition.

Not all text files are the same. One big difference between Windows (which I believe you use from your reference to Notepad in >268 faktorovich: https://www.librarything.com/topic/337240#7674808) and Unix-based systems is how each line of a file is ended. In Unix systems, the character used is a newline (\n or ASCII-10). On Windows it is two characters — the carriage return followed by the newline (\r\n or ASCII-13, ASCII-10). This alone can cause problems with some programs.

Windows (and Word in particular) is nasty about using its own standard for how to encode typographer's quotation marks, apostrophes, em-dashes, and other characters. A document manipulated in Word is likely to have these Windows-specific encodings that don't play well with standards-based systems.

It is not enough to name any old folder "corpus", it has to be in the place where the program is expecting to find it.

The program also has to have permission to read the directory. This is probably not an issue with Windows since most things are wide open. On a Unix-based system it is more common.

Finally, you refer to breaking words. A default behavior of Stylo for R is to take words like don't into two parts do and n't. If this is what you saw and are complaining about, that is configurable as noted in the ResearchGate article that I linked to above.

I've not done extensive work with Stylo for R but I like what I see about it. I am also giving favorable consideration to try JGAAP. I have used other programs previous to this forum thread. I have even written a program to implement the procedure for the method described in Farringdon's book. I made this about 20 years ago for a PCA presentation in 2003 which would generate their particular forms of graphs that Excel would not be able to replicate easily.

I have no particular allegiance to it or a stylometric process. My usual process is to set up some experiments and see if it create expected results for works with known authorship to a high degree of confidence (e.g. a long vintage letter written by and typed by an individual). Once I have that understood, then I can see what the technique offers for more questionable texts. Does it support the extrinsic evidence (letters, contracts, etc.)? Does it make sense with what I know of the circumstances?

But at the end of the day, the stylometric analysis is just an interesting sidelight. It is not the primary evidence. It can support or contradict what the other evidence indicates. If it contradicts, further research is required to understand what is going on.

For example, I know that the series book texts had many hands influencing the final product. This is really true of any text but it is more documentable in the series books. Knowing who contributed which portions can be interesting if they can be detected. It shows the process of producing books like this. That is why I look at this. But I can say with 80% confidence that my next PCA presentation will not be based on a stylometric study. There are too many other interesting things to explore in this field.

James

>265 Keeline: The number seems a little high since the largest dictionaries contain a little more than 1/4 this figure. Where are the other 722,000 words?

Beilstein has millions of chemicals mentioned in it; whether the verbal formula really counts as a word is questionable. Many of them are medical compounds; are Zoloft(tm) or sertraline (notm) English words? What about names? Looking at the IMDB data files, there's more than 100,000 first names there, e.g.
...
Abigael
Abigaël
Abigaelle
Abigaëlle
Abigail
Abigaíl
Abigaile
Abigail-James
Abigal
Abigale
Abigayl
Abigayle
...

Names of cities? Mountains? Wikispecies has 800,000 content pages, each with its own scientific label for some group of creatures. What's a word, and what's an English word can get quite fuzzy.

I've edited quite a bit at the English Wiktionary, and it has 588,632 pages with different names with English entries with a definition, with 1,053,233 pages with English entries counting all the grammatical entries (plural, past tense, gerund). It's a wiki, yes, but we do a fairly good job of cleaning things; we require three cites over a year with electronic sources limited mostly to Usenet. It's not the Urban Dictionary.

>265 Keeline: "You have claimed that your collection of texts (a corpus) has X million words in it. Yet, I think you realize that the number of distinct words is much smaller. After all, you also claim the high estimate of 1 million words in English." Yes, you have figured it out, the number of total words in the corpus' texts is not the same thing as the "number of distinct words" in it.

Yes, you guys have explained that "Identifying the parts of speech is difficult to do." However, what you have failed to do is to show how any program you are pitching as an alternative to Analyze can accurately tags all words, even when they are Homonyms or look the same but have different meanings. As an experiment, I ran the old-spelling version of Percy's "Shakespeare"-bylined "Romeo and Juliet" through CLAWS. Here are the first 2 errors I found going down the list. "mou'd" was miss-interpreted as 2 words: ""NN1">mou" and ""VHD">'d". So, "mou" was categorized as NN1 or "singular noun", when it at least should have been categorized as a "verb", if the system was sensitive to the common spelling of "move" with a u in Early Modern English. And 'd was misinterpreted as "past tense form of the verb "HAVE", i.e. HAD, 'D"; this is not as far as judging "mou" to be a noun, but it still does not precisely categorize the word. And by splitting this single word into two separate words (just because it has a similar appearance with the apostrophe there as a "he'd" or "he would" type of a contraction); thus the word-count for the text and the counting for this one word into two is miscounted.

And the system is not sensitive to the idea that "Samp" can be the name of a character (as it is on a line on its own, and repeats across this play); instead it categorized it as a VVB or "base form of lexical verb".

It would be extremely difficult for any tagging system to tag any text perfectly or even with a high degree of accuracy when Early Modern English is fed into the system. A programmer has to work with a linguist to establish far more complex algorisms for spotting most of the possible glitches by running actual texts through a system and correcting all glitches by writing new lines of code that accounts for all miss-classifications. CLAWS might be a bit better at catching some word-types than Analyze, but the data it generates is not in a practically usable format without a programmer's intervention. Thus, Analyze's somewhat weaker tagging accuracy is substituted with its far more practical applicability to the attribution method I designed with programs like Analyze in mind.

Putting all this aside, are you saying that: if I edit my 27-tests method into an 82-tests method and re-calculate the entire corpus of 284 texts by running them through these 62 CLAWS tests for word-type and my other 20 tests with the standard programs, and this data still points to the six ghostwriters behind the Renaissance; then, you will agree that my attributions are correct, that my method works, and that there are no major linguistic glitches that can be preventing it from working? Or still easier, are you suggesting I just run all of the texts through CLAWS use your suggesting combination of adding all of the nouns (and other same word-types I used) together into the single measure, and re-calculate the data to see if this would generate the same attribution results; then you would agree with me? Or are you insisting that to make you believe my method works I have to create a program that turns my method into an automatic process where you can put your own texts into the system and receive an automatic result? None of these steps are rational, as you are only pitching adjusting 7 out of the 27 tests, and these 7 tests already show the same attribution outputs by themselves, as if the other 20 tests are used separately. If there was something intrinsically wrong with the Analyze tagging system that was creating miss-attributions, the 7 word-type tests on their own would not show the same attribution results as the rest, and they do; so this cannot be a significant problem with my method. If CLAWS' tagging was flawless it might be a worthwhile experiment, but it still generates a significant number of glitches, so it is not a rational test for grammatical purity. Of course the actual reason you would not be willing to agree to any of these options is because none of them involve me paying you to hire you to do the work you keep insisting I have to do. This is exactly how you all are pressuring me to buy your programming services when you are pitching completely unusable tools like Stylo or tools that require programming to make them fit my needs like CLAWS.

>266 Keeline: "How can you be sure that your 400+ year-old texts that you attribute to people like Verstigen is really by him? How do you know it was not ghostwritten by another?" I explain in several chapters in Volumes 1-2 and in the "Restitution" translation why Verstegan was the only ghostwriter who was alive long enough and otherwise (by his biography) was the only one out of the bylines in his authorial-group of texts who could have written all of them. If you have this question, the logical thing would be to ask me for a free review copy of Volumes 1-14 of BRRAM, and then you can read the relevant chapters and reach the full answer, which is too complex to give here, since you would otherwise assume that my summary was the only evidence I have for the conclusion.

I have already quoted from my translations of "Restitution" and other texts. You can see the "before" or original versions by searching for them on Google Books or EEBO. Here is "Restitution": https://www.google.com/books/edition/A_Restitution_of_Decayed_Intelligence_In/vC... You can search for words in my translation fragments to find the relevant sections.

"how much is altered in the texts becomes important. Are we detecting traits of Shakespeare or Verstigen or Faktorovich?" I have no idea what you are trying to ask. The edits I make to texts prior to testing include deleting unreadable characters, or random dots that have no linguistic meaning. I do not edit the texts in any way that would alter their signature. And "Shakespeare" is a pseudonym, not a real person. I do not "drop pronouns" or take any of the other invasive steps you mention.

The article you are citing mostly repeats the standard mumbo-jumbo in such articles that I commented on across BRRAM. It does include a few lines of code that almost never appear in other scholarly articles on this subject, but these lines give an isolated direction when the problems with Stylo are spread across this program starting with glitches of accessing texts for testing. It would be absolutely impossible for anybody who has not used Stylo before, or even an advanced programmer who is not familiar with Stylo in particular to be able to use it for all of these different functions just based on the information in this article.

Why would I invest any time into researching Stylo's bugs when I have explained that this system is really only capable of handling word-frequency counts, and not a combination of 27 tests that my method requires? I have already completed the tests for the Renaissance, and you just haven't read BRRAM yet, and so you are not aware of what I have found out (as apparent from your question regarding why Verstegan was "the" underlying ghostwriter for one of these groups).

A better example would be if I had built a car of my own design and test driven it to show that it works. And all of you were insisting that I have to change it from electric to gas-operated and for this I had to use only the gas-tank operator that you recommend as the best one in the market, whereas the electric battery I am using is "garbage" because it needs re-charging. So you keep pressuring me to invent a new car and also to fix the bugs that are making your gas-tank suffer occasional sparks. Why don't you just review the car I already invented, and not the old version of the car that you have been using that you think I should duplicate.

When you state "stylometric analysis is... not the primary evidence", I once again realize that you still have not read any part of my BRRAM series aside for this discussion, and you have not even been reading the additional evidence outside of stylometrics that I have been giving inside of this thread.

>269 faktorovich: A better example would be if I had built a car of my own design and test driven it to show that it works. And all of you were insisting that I have to change it from electric to gas-operated

The electric car is about as old as the gas-powered car. Making the first modern electric car that was a serious alternative to the gas-powered car was a huge challenge for many reasons, and required extensive industrial research, not an amateur who decided to poke at it a bit. I don't think that you've shown the level of evidence for a test drive, but even then, sure, your car can drive around the block. How far can it go on one charge? Does it have the speed and acceleration to handle interstates? How long will it last? Will it spontaneously catch fire? How safe is it in an accident?

Reversing the analogy, you present us with a poorly tested system and keep repeating that it works because you know it works and claiming that the system everyone else uses is buggy and that we're in the pocket of big R, instead of looking at the problems with your system.

>270 prosfilaes: I just watched a documentary about John Harrison (https://en.wikipedia.org/wiki/John_Harrison) "a self-educated English carpenter and clockmaker who invented the marine chronometer, a long-sought-after device for solving the problem of calculating longitude while at sea." He spent 30 years researching the best method for this clock to function at sea until he finally succeeded. Most great inventions and research takes place by independent researchers who tend to self-publish, as Galileo did (if they are not the rare few to win a grant after they have already succeed as Harrison did). Harrison was as much an "amateur" in clockmaking as I am an "amateur" in researching literature, as both of us have spent decades on our craft.

My 27-test computational-linguistic author-attribution is a finished and tested invention. "The level of evidence" is at the max when somebody has published 14 volumes full of evidence, and many more are forthcoming. I have been discussing some of this evidence in this thread, so even if you have not read any of the BRRAM series, you should be aware of the evidence by now, unless you have not been reading anything I have said, and instead have only been listening to yourself. 284 texts is the largest corpus ever tested in one experiment of the British Renaissance (which is the most challenging corpus given the linguistic obstacles with Early Modern English), so this method has been tested on the longest and most complex obstacle course imaginable. In the many periods I have tested, and the hundreds of texts, the method has consistently worked, and has not "caught fire". Whenever I have encountered what looked like a glitch with it, it was actually a unique attribution indication that has led my to a more precise authorial attribution after further documentary research.

My method works and it has led me to correct attribution results. Your method does not work, and that's why users of it keep changing between a dozen bylines for each new attribution study even for a single poem such as "Funeral Elegy". There are no bugs in the attribution steps I designed. Analyze is the least bug infested free tool I could find; Analyze has similar bugs to CLAWS, so there is no better alternative (yet); if there are bugs in Analyze or CLAWS these are not the fault of their users; my invention is not a program to count word-types, but rather the method for how to calculate authorial attribution by using free tools accessible to all. After I finish translating the volumes for BRRAM by around January, 2023, I will perform a new giant study of a range of bylines from the 18th century to reach more precise results and to write a book on the re-attribution of that century in British literature. Before I start that calculation, I will reconsider if there are any new tools that have become available in the past few years, and if I could incorporate the steps to make the process quicker and more precise. I have adjusted my method many times since I began designing it in 2015. This is why my posted data results are extremely detailed and precise. In contrast, I have read decades-old articles about the Stylo method that had identical vague descriptions of word-frequency calculations in the most recent articles.

>271 faktorovich: My method works and it has led me to correct attribution results.

You make this claim with the conviction of someone who has received in-person confirmation from the shades of these dead alleged authors. I'm beginning to wonder if you are the materialist skeptic you've sometimes presented yourself as.

Stylometric analysis (whether by your disputed idiosyncratic method or more collegially-verified approaches) can only ever be one indicator among many. The supplementary "proofs" you've pointed at in these threads have been underwhelming.

>272 paradoxosalpha: I have the conviction of somebody who has spent the last year-and-a-half translating these writers texts into Modern English. This process involves re-writing nearly every word, and researching every hinted at meaning and checking every possible quote (without quotation marks) or reference to an earlier source. If you refer back to my explanation about the "Ghosts" of the dead writing, these ghostwriters sold their services with the claim that those who paid to have their byline printed on a book would be immortalized or would live in fame beyond their natural deaths. They did not mean that these "byline"-purchasers would live as spirits or ghosts, but rather that their names and the words attributed to them would still be around hundreds or thousands of years later (like ancient scrolls or hieroglyphic tales about pharaohs in the tombs). By reading and re-writing what these ghostwriters wrote, I have indeed come to know them very well. And what I learned is explained in the annotations and introductions to these translations that take up at least half of each of these volumes. Knowledge is not a ghost that travels into minds. To figure out what my full "proofs" are you have to ask for a review copy and read the details for yourself. My method is rational, correct, and proven; whereas the other method(s) are a black-box of cryptic claims of superiority without any fully disclosed raw data (at least none that does not reveal fraudulence or mistakes). If you are dismissing my method and my "proofs" without reading the books where I present them, you are leaping to shady spiritual conclusions, and not me.

>271 faktorovich: I just watched a documentary about John Harrison ... Most great inventions and research takes place by independent researchers who tend to self-publish,

A 17th century clock maker who was raised in the field hardly seems a good comparison to the modern world. I recently read that Marjorie Rice was the greatest modern mathematician who never did graduate work, and she discovered pentagonal tilings of the plane, something that needed a sharp mind and patience more than an intense understanding of complex mathematics. You again make a wild claim that I'm sure you can back up. I doubt it; a huge amount of research is done by with very expensive instruments that independent researchers don't have. David Hahn went about as far as an independent researcher could into nuclear power, and succeeded mostly into turning his back yard into a Superfund cleanup site. Medicines need labs for creation and large organizations for proper testing. My fields of mathematics and computer science are plausible places for independent researchers, but I know of no major independent mathematicians. One can bounce around computer science, but even in this young field, a lot of the stuff was prototyped in universities and made mainstream by big companies hiring university hotshots.

"The level of evidence" is at the max when somebody has published 14 volumes full of evidence

And somehow you don't believe in God/Scientology/that Elvis lives, despite the volumes and volumes of evidence produced for them.

In the many periods I have tested, and the hundreds of texts, the method has consistently worked,

You test if something works by feeding it a problem you know the answer to and seeing if it gives you the results that are known. If you want to be a Galileo, a Kepler, a Newton, an Einstein, show that your system fits the movement of the planets better than your predecessors, not that people who have been watching the skies have been all wrong about where the planets are.

Your method does not work, and that's why users of it keep changing between a dozen bylines for each new attribution study even for a single poem such as "Funeral Elegy".

Which might indicate that stylometric analysis lacks the power to give completely reliable answers, especially for works like the Funeral Elegy.

>269 faktorovich:

I have already quoted from my translations of "Restitution" and other texts. You can see the "before" or original versions by searching for them on Google Books or EEBO. Here is "Restitution".... The edits I make to texts prior to testing include deleting unreadable characters, or random dots that have no linguistic meaning. I do not edit the texts in any way that would alter their signature.

You keep using translation but I think most others would call this a transcription.

Words mean things

But if your translation involves normalizing spelling or changing words to synonyms, that is where more information is needed to assess the impact of this process.

James

My goodness... the last time I looked at this topic thread was last Christmas... thinking it was dead, I never returned, as I had run out of the ability to laugh and crack jokes about it... but it's clearly like Frankenstein's Monster resurrected! Presumably Frankenstein being ghostwritten by some impoverished writer in a Dickensian London setting... which of course poses the question, was Dickens ghostwritten as well? Talk about down the rabbit hole...

>271 faktorovich: Most great inventions and research takes place by independent researchers who tend to self-publish, as Galileo did

Galileo was a professor in a leading university, as close to being a professional scientific researcher as you could be in his day. He was also a leading figure in the scientific world of his day. Even in old age, after his brush with the church authorities, he still counted as a major tourist attraction, as witness Milton seeking him out on his grand tour.
I don’t think the modern distinction between publishers and printers existed in his day, so I’m not sure how you would define “self-publish”.

Almost all inventions are made by professional people — engineers, scientists or artisans — in the course of their daily work, as incremental improvements to previous work by others in the field. Almost all inventions made by private people come to nothing, either because they are not original, or because they lack the means for putting them into practice. In the very rare instances when a private inventor invents something significant, that is news, and we read about it: you never hear about all the hundreds of unfortunate people who waste their money inventing perpetual motion machines, flying cars, spaceships and sometimes potentially useful things that simply never make it to the market. I know this because I worked in a patent office for many years. I also know that surprisingly few of my colleagues ever came up with Nobel Prize winning ideas about relativity in their spare time.

>259 faktorovich:

I don't understand what your problem is. Large corpora allow meaningful scrutiny of relatively rare patterns? Or of very specific combinations where both parts are explicitly specified (KISS + "cheek" instead of KISS + NOUN)?

In a single text, there would not be 1813 occurrences that could be jointly categorized as noun vs. verb by checking only a single box in a tagging system that does not have an automated system for distinguishing between a noun and a verb.

??? Of course there would not be a single text with that many occurrences of kiss + cheek. Why is that noteworthy? Is this a problem? And of course a corpus that's untagged would not allow for this kind of analysis.

The 27-tests that I use are accurate for attribution, and the fact that I have performed them is registered in their data on GitHub.

You may have performed them. But that's about all that can be said about that.

One of these days we need to have a conversation about "data" vs "analyses". I don't think that distinction is particularly clear to you.

>260 faktorovich:

The unsolicited homework you have been submitting is atrocious. I have just been investing my time into grading it for free and pointing out areas where you can benefit from improvement

K.

the overall analysis will still be counting the tendency to use or not to use them by the author in the data.

You only look at the total numbers of nouns, verbs, adjectives, prepositions etc. You have a wildly inflated sense of the kind of information that is explicitly included in those figures.

If you want to count the tendency of an author to use particular words, you look at those words in particular! You can't glance at the total number of nouns and figure that that says something meaningful about record as a verb or a noun. Or about any other individual noun, or subset of verbs, or whatever.

Same thing with looking at the six most frequent words and phrases. You think that that means you've actually considered and carefully analyzed individual words and phrases. But you only look at the top six by absolute frequency. All you've taken into account is the top six frequencies; you don't even glance at the frequencies of the other words.

>258 faktorovich:

K.

>258 faktorovich:, >280 Petroglyph:

Litigious is always a good look for a "scholar."

/s

>274 prosfilaes: The probability that a discovery can be made by an independent researcher versus a funded dependent-on-a-university researcher is of no interest to me. My method works. And I happen to be an independent researcher. I have researched the frequent cases of the best scientists, such as Galileo, who have self-published and worked independently; but I have not conducted a statistical study to determine how much more likely it is that a ground-breaking finding might come from an independent versus a dependent researcher.

One would assume that people who believe in "God/Scientology/that Elvis lives" have read the "volumes and volumes of evidence produced" on these topics; but in truth, it is very likely that none of the people who believe in these have ever read an entire book that attempts to seriously prove in their existence. Such fields insist on belief without proof, and their books tend to be puffing restatements of shared belief, without any attempts at factual proof. If you imagine my BRRAM series has anything to do with any of these belief-based theories; then, you clearly have not read BRRAM. Every word in BRRAM is a piece of evidence that explains and supports the re-attributions and explains the history of the Renaissance from this new perspective.

No, I did not at all know the answer to "Who wrote the Renaissance" before I started this experiment. And all of the answers regarding the names of the six ghostwriters were surprising to me when the data and documentary evidence indicated them.

The planets or the texts are entirely unchanged by my re-attribution. Most of the Renaissance texts, as I have said, were originally anonymous, and have been miss-attributed by linguists across the previous centuries. So if you think I am moving the locations of the planets, and you are only happy with the original bylines; you should add many plays and subtract some plays from the "Shakespeare" collection volumes, as some did not initially have his byline on them, and some that did are currently not in the collections.

I changed my own attribution assignments a few times as I kept researching all other possible bylines. But then I reached a point where as I continued to do volumes of additional research, my attribution conclusions remained solid and unflinching because I was only finding evidence to further affirm the conclusions, and no evidence of any other alternative bylines that could have done it. The problem with previous attribution studies such as "Funeral Elegy" is that researchers have compared it in isolation with only a handful of other texts by the bylines they predetermined to be likely authors; all of them have used a corpus of no more than 20 texts, and frequently only a handful of them; it is impossible to come up with an attribution conclusion with any certainty until a study is expanded to at least 104 bylines tested, with thousands more reviewed for possible authorship, as I found out because when my corpus was smaller I had not yet added into my study many of the bylines that turned out to be the true underlying authors. Out of the six ghostwriters, my initial set of around 100 texts only included the Jonson and Harvey bylines; so I could not have reached a precise attribution, since I was missing 4 of the ghostwriters, and I found out who they were as I kept expanding the corpus of possible authors. My computational-linguistics author-attribution method can and has given precise attribution results, but only if the researcher behind it exhausts the options and keeps researching the subject long before and after the math is solid.

>275 Keeline: If you do not know the degree of the translation I have performed in the 12 previously published volumes of BRRAM, you are confessing to never have looked inside the series, nor read any part of this discussion where I quote my translations. Distinguishing between a "translation" and a "transcription" would be easy enough if you had done the minimum degree of research by looking these books up in Amazon's LookInside, even if you can see it fit to ask free review copies. I am doing translations and not transcriptions. The degree of the translations is explained in the series. I disagree with the traditional methods of no-translation or extreme-modernization-translations that have been applied to this corpus by previous editors. My approach is to change all words that are classified as "archaic" or are otherwise out of use to their modern equivalents; I modernize and standardize spelling; and I change the order of words when it is following archaic grammatical rules. I do not change any non-archaic words simply because some words are long, complex or unfamiliar to the average modern reader. I developed my own annotation system that makes it easier for readers to grasp all of the major changes I am making in archaic words. As I said, half of these volumes are also annotations and introductions that explain the meaning, references, and my attribution decisions about the texts in the corpus.

>276 Bushwhacked: I have not tested Dickens before, but I have written about Dickens in my books, and from this I would intuitively conclude it is more likely that Dickens ghostwrote for others, than that his books were ghostwritten by somebody else. Though it is also possible that Dickens only wrote his own books.

>282 faktorovich:

I have researched the frequent cases of the best scientists, such as Galileo, who have self-published and worked independently

Which of Galileo's books do you claim were originally self-published?

I am not seeing indication of this from the first edition title pages or commentary about him.

I do find that he had several academic posts. Here are a couple of them mentioned:

Galileo Galilei was born in Pisa in 1564, and became a professor of mathematics at the city’s university in 1589 — aged 25. Initially, he didn’t show much interest in astronomy, devoting himself instead to discrediting Aristotle’s views on the physics of motion.

When his three-year contract in Pisa expired, Galileo moved to become the chair of mathematics at Padua, at the time one of the most important intellectual centres in Europe. It was here in around 1597 that he began to develop an interest in the astronomical model developed by Nicolaus Copernicus, which made the then-extraordinary proposition that the sun was at the centre of the universe with the earth and other planets rotating around it.

https://www.christies.com/features/Six-first-editions-by-Galileo-7519-1.aspx

While I don't consider most auction listings to be wholly accurate, this one is more detailed than most. I was mainly looking for information on the publication history of his books.

James

>277 thorold: Yes, I just reviewed a book the other day that described how Galileo was patronized by the Medici, who not only purchased copies of his printed books in large numbers, but also actively mailed these copies to those they knew to propagate as his patron for Galileo's findings. In the first centuries of print none of the printers paid authors, and only gave them author-copies, and typically authors paid for their author-copies or paid to have their books printed. The idea that a publisher gives an advance to an author also has been subverted, as an advance means a loan against expected profits; this loan has to be returned by the author to the publisher, if the expected number of sold copies is not reached. I cover the latter point in my book on "The History of British and American Author-Publishers". Since I have taught at universities as well and since I am a publisher, I am not exactly entirely independent either.

There is an incredible quantity of boundaries between an inventor in the modern world and their ability to advertise or sell their invention without the act of trying to sell it prompting a competitor to steal it before a patent or sale is solidified. Crooks that steal are more likely to have stolen before and thus have the funds to create a factory to realize an invention. Corporations have a self-interest to monopolize the market to keep people buying whatever it is they have invented centuries ago; any new invention that threatens to make their product obsolete is a threat to their profits, and so they go out of their way to bankrupt the new invention before or after it reaches access to the market. So cereal, for example, is stuck in 1863 when James Caleb Jackson "invented" it. Zero can be spent on science and development if the same flakes can be cold for 2 centuries, and people keep buying them. So however invention is currently done is not working, as it is stifling any hope for innovation. An inventor who had their first product stolen and bankrupted is unlikely to invent a second. This discussion here on LibraryThing is one of these obstacles that I am facing as an inventor. Those who use a rival, widely-accepted method are working to insult my method and to silence my study. Unlike the average inventor who would have run out of money by now; I am an independent publisher, and I have been at it for the past 4 1/2 years and I will keep going for at least another couple of years. Since I have been researching the obstacles that are constant in publishing, writing and invention, I am not surprised by this response, or the obstacles I am facing.

>279 Petroglyph: According to studies such as, https://aclanthology.org/www.mt-archive.info/MT-1967-Earl-1.pdf, there are around 2,600 "Words forming a homonym in MW3" or Webster's Dictionary, which in total includes "470,000 entries". So, if these homonyms were spread equally among words, they would compose .6% of all words in a text. It is very likely that tagging systems use the most common word-category (noun vs. verb) for each of these homonyms, so in most cases they are likely to pick the right group even if they do not have a special feature to check proximate words for the most likely intended meaning. Even if the tagging system always miss-classifies homonyms this is likely to make up only .6% of all tagged words; this would not have a statistically significant impact on the attribution results the tests would produce.

And as I have said before, the tests for word-types are not a grammar check. It does not matter if the words are categorized in the proper categories that fits the intended meaning in the texts. These tests are mathematic sorters that group words into categories A, B, C... As long as all words in the dictionary are consistently grouped in only one of these categories, they are still being sorted accurately. The tendency of the author to use a given homonym is still tested, even if they always mean to use it as a verb, but it is being classified as a noun. If you realize that the words "verb" and "noun" are irrelevant for a mathematic equation that is counting words in category A versus words in category B. And so Analyze's system still precisely fits the needs of my experiment by sorting words consistently. As I have already said, I will research if there are any more precise programs out there before I start re-testing and expanding the corpus of my 18th century study in a few months. At that time, if I do try testing CLAWS, I will be sure to check its attribution results against Analyze, and will report my findings in my study.

I have "looked" at the frequency of other words beside the most frequent etc. I decided that those frequencies cannot be systematically compared in a simple system that gives accurate results. The top-6 words and letters do frequently tend to identify the underlying author by themselves without even having to perform any other tests. Comparing all words in any one text against all words in all 284+ texts in a corpus is statistically nonsensical, as there would be an infinite possibility of matches and non-matches.

>285 Keeline: A better question for you is which of Galileo's books do you imagine were published by somebody other than Galileo? And how are you defining a "publisher"? The title-pages of 15-17th century books list the printer, bookseller and author; I do not recall seeing any that list a "publisher". To prove that somebody other than Galileo published his books, you would have to provide a document that says a printer/bookseller paid Galileo for selling his copyrights to them to be the exclusive publisher of a text. Galileo's patrons paying for his books to be printed does not mean the Medici published his books; they simply sponsored the publication and his research.

Taking a few examples from the page you cited. The publishing information is at the bottom of the pages under the illustrations.

"Venetiis, Apud Thomam Baglionum. MDCX. Superiorum Permussua, & Privilegio."

"In Firenze. Appresso Cosimo Giunti MDCXV."

"In Roma MDCXXIII. Appresso Giacomo Mascardi. F. Villamoeria Fecit."

Latin for:

"Venice, by Thomas Baglionum. 1610. With Superior Permission and Privilege." There have been no major studies of "Thomas Baglioni's" printing house.

"In Florence. Press "presso" means "press" in Latin, while "appresso" is "near" in Italian, so this might have been a deliberate typo Cosimo Giunti 1615." Info on these printers: https://en.wikipedia.org/wiki/Giunti_(printers) - there is no mention of Galileo, and this might be connected with the strange "appresso" typo.

"In Rome 1623. Press: Giacomo Mascardi. Printed by: F. Villamoeria." Mascardi: printer: https://library.brown.edu/projects/rome/people/0151/ "Villamoeria" does not seem to appear elsewhere.

Where in this information do you notice anything about a publisher? Only the Giuntis have been vaguely called publishers, though they were professional printers.

>277 thorold: I don’t think the modern distinction between publishers and printers existed in his (i.e. Galileo's) day
>288 faktorovich: Where in this information do you notice anything about a publisher? Only the Giuntis have been vaguely called publishers, though they were professional printers.

I find it easier to credit Thorold's premise that "publishers" emerged as a reified function out of the class of printers ("presses" and "imprints"), rather than Faktorovich's that the work of publishers could be subsumed in the identities of authors. The latter seems like an anachronistic projection of 20th and 21st-century "self-publishing." Even the Giunti instance cited by Faktorovich supports the former notion.

while "appresso" is "near" in Italian, so this might have been a deliberate typo Cosimo Giunti

It's not a typo, deliberate or otherwise.

"Appresso" in this context means "at the house of", "the press of": https://maltamapsociety.mt/glossary/terms-for-publisher/

See, for example:

https://lcweb2.loc.gov/diglib/media/loc.natlib.ihas.200196602/0001.tif/1719#h=12...
https://galileo.ou.edu/sites/default/files/G-10-Gal-1613-1.jpg
http://sites.nd.edu/rbsc/files/2018/02/BOO_004711008-001r.jpg
https://thumbs.worthpoint.com/zoom/images4/1/1015/13/1559-del-cortegiano-del-con...
https://upload.wikimedia.org/wikipedia/commons/thumb/0/01/Caccini_-_le_nuove_mus...
https://dobianchi.files.wordpress.com/2022/01/agostino-gallo.jpg
https://www.researchgate.net/profile/Annibale-Mottana/publication/284927487/figu...
https://c8.alamy.com/comp/K4N45B/peri-jacopo-cover-of-score-for-opera-euridice-1...

You get the idea.

>289 paradoxosalpha: I did not state that "the work of publishers could be subsumed in the identities of authors". I stated that the term "publisher" is misused (anachronistically) in modern times when it is applied, especially to the first two centuries of print. The current use of the term "self-published" as a pejorative is contrary to the long history of the best authors paying to print or finding patrons to print their work. If you are going to argue that "Giunti" supports an anti-self-publishing position, you really have to clarify what you are trying to say, as the Giuntis were printers who mostly printed books people paid them to print.

>290 lilithcat: If you had read my full explanation, I stated ""presso" means "press" in Latin". I translate it as "press" in my translations. It does not mean "the press of"; this phrase in Italian would be, "la tipografia di"; if it is a contraction, it would be more like "lapresso" or "latipografia", and not "Appresso". While the cite you refer to claims "Appresso" means "the press of" etc. in Italian, they are wrong, as I have explained. "appresso" is "near" in Italian. An English reader of one of these books might guess "Appresso" means "press" without checking in what language this word has this meaning. Latin and Italian were frequently mixed together in books from this period, which is why a foreign-speaking publisher might have made this typo consistently on many books where this word was repeated to signify "press". It is amazing though that you guys are committed to finding errors in my statements, even if you have to create an avalanche of errors of your own to do so.

>292 faktorovich:

Oh, right, I forgot. You never make mistakes. Obviously, all those early modern Italian printers (including a slew of others to whom I did not bother to link) didn't know their own language, or all made the same typo, or were part of the same deep-dyed conspiracy.

I don't suppose it's occurred to you that language changes over time.

>282 faktorovich: The planets or the texts are entirely unchanged by my re-attribution. Most of the Renaissance texts, as I have said, were originally anonymous, and have been miss-attributed by linguists across the previous centuries. So if you think I am moving the locations of the planets, and you are only happy with the original bylines;

The point of the planets is that we know where they are. You're the one who chose a corpus where most of the texts were originally anonymous, and chose to devise your own tools to handle it. You're the one who has also refused to apply those tools to a corpus that is well-known.

>286 faktorovich: This discussion here on LibraryThing is one of these obstacles that I am facing as an inventor. Those who use a rival, widely-accepted method are working to insult my method and to silence my study.

Paranoia is such a good look, isn't it. At least one of the things I think that make academics more successful is that they have, from day one, worked with the idea that they have to make their case to other people.

>287 faktorovich: These tests are mathematic sorters that group words into categories A, B, C... As long as all words in the dictionary are consistently grouped in only one of these categories, they are still being sorted accurately.

Do you have tests to prove it? Have you tried randomizing the words that go into categories, if it doesn't matter? (I know you're going to take umbrage, but that is exactly what I would ask any scholar who made that assertion.)

>287 faktorovich: Comparing all words in any one text against all words in all 284+ texts in a corpus is statistically nonsensical, as there would be an infinite possibility of matches and non-matches.

No. Just no. There's nothing infinite about it.

>293 lilithcat: Language changes, but "Appresso" does not register in modern dictionaries as an archaic Italian word, but rather as a modern word that means "near". I did not say there was a "conspiracy" behind the term "Appresso" whenever it was used on title-pages, but merely that based on my other research, its use is suspicious. I have found several suspicious wordings on title-pages including "Nashe His Ghost", and radical books claimed to be printed in cities where no other books were printed in the neighboring decades; the implications of such suspicious wordings take a while to figure out, and I only glanced at "Appresso", tagging it as a good research project, but one that I was not going to explore further. If you assume a suspicion is enough to establish a conspiracy, the exaggeration is yours.

>294 prosfilaes: Most of all of the texts that were printed in the Renaissance were anonymous. The 284 texts corpus I chose was less anonymous as a percentage of the corpus than the percentage of the entire pool of all texts published in those decades, because my goal was to figure out which of the stated bylines belonged to the underlying ghostwriters, and that cannot be done without testing at least 104 stated bylines.

And aside for picking texts with different bylines, the rest of the corpus is mostly texts that have been studied by previous attribution researchers, and thus are known mystery-texts. I studied most of the "well-known" texts that have been the subject of debate, such as "Shakespeare" and "Marlowe".

One of the most overused tools used to disregard a rival theory is to call the theorist "paranoid". I have made the case "to other people", and I have provided overwhelming evidence. A sign that your insults are malicious and designed to discredit an accurate finding is that you have given up on finding any specific things wrong with it, and have now just devolved into name-calling.

Have I tried creating a program that would divide a dictionary into 7 categories other than "noun", "verb" etc. and testing if it would still give accurate attribution results? No, as I have said, I have not yet created any new computer program, as I am relying on the fact that programmers have already created plenty of programs that fit the task. There are also no equivalent categories to parts-of-speech that would divide a dictionary into 7 groups. I have tested the same texts with 20 other tests, and those tests are the alternative measures that reach the same results as the 7 parts-of-speech tests. That's why I am using 27 different tests, so if there are any glitches with any single test or a few of the tests, the other tests still steer a texts to be similar to the texts it is most proximate to.

My corpus has 284 texts with 7.8 million words. A comparison of all of these texts against all others on the frequency of each of the words would be approximately = 284 * 283 * 7.8 million = 629 billion comparisons. If you try imagining the size of this 3D table that would fit all of these comparison scores on each of these words for each of the texts, you will notice why the idea of measuring all words in all texts for frequency is an absurd idea.

>296 faktorovich: Most of all of the texts that were printed in the Renaissance were anonymous.

Then forget the Renaissance!

One of the most overused tools used to disregard a rival theory is to call the theorist "paranoid".

Not really. It tends to go against people who believe that people who disagree with them are only doing so because "Those who use a rival, widely-accepted method are working to insult my method and to silence my study."

I have not yet created any new computer program, as I am relying on the fact that programmers have already created plenty of programs that fit the task. ... That's why I am using 27 different tests, so if there are any glitches with any single test or a few of the tests, the other tests still steer a texts to be similar to the texts it is most proximate to.

So your system is hacked together from a number of publicly available tests, unrefined and unoptimized.

My corpus has 284 texts with 7.8 million words. A comparison of all of these texts against all others on the frequency of each of the words would be approximately = 284 * 283 * 7.8 million = 629 billion comparisons. If you try imagining the size of this 3D table that would fit all of these comparison scores on each of these words for each of the texts, you will notice why the idea of measuring all words in all texts for frequency is an absurd idea.

Have you ever heard of a computer? This doesn't even get in to the range of big data, much less impossible. A mid-size system would grind through it in minutes.

>296 faktorovich: you have given up on finding any specific things wrong with it,

* The method is untested, and you've show disdain for running it against a known set of works to see if you would get the expected answer.

* Your unfamiliarity with other tools, and unwillingness to learn them, brings into question the quality of the methods.

* Your inability to handcraft your own tools and choice of online webpages instead of the tools professionals swear by is also problematic.

* The method is producing extraordinary results, of works normally thought to be by many authors being by just six.

* This extraordinary result seems to be caused by how you tuned the method to detect two works being by the same author; if that's turned to extremes, naturally it will produce a very small number of authors.

* You rewrite known history, like the location of Vergestan and the authorship of the King James Bible.

* You chose a section of poetry you claimed was obviously by the same author that wasn't clearly by the same author, and you backed away when challenged.

* You're offering seriously extraordinary results, like one exiled author writing all of England's religious works on all sides.

* Your answer is simplistic; there's six ghost writers, period. There's no one-book authors, Queen Elizabeth and King James were ghostwritten, as were all the clergy and all the judges.

* The Holy Ghost, part of the Trinity, can't write. I'm not a believer, and it's hard to get accurate counts of religious beliefs in places where blasphemy is still a crime, but most of the authors and readers would presumably have been religious, and claims of the Holy Ghost writing through an author would be taken straightforwardly.

Hopefully this is enough; not that I expect you to quit, but maybe I can.

>297 prosfilaes: There is no rational reason other than a self-interested investment in a rival attribution of the Renaissance for anybody to disagree that the attribution method I have described is the best option of those that have been presented so far in this field.

My system is the product of years of research, refining and optimization. Yes, it does rely on publicly available tests, so that members of the public can use this method on their own without needing to consult me in the use of specialized programming.

This calculation is not impossible because there are too many data points, but rather because there is a near-infinite number of combinations for which texts match on which words. For example, the word "impossible" can occur once in 1 text, or a varied amount of times in all texts, or a varied amount of times in half of the texts. And the same would be the case for all of the other words. All of the texts are going to have at least a set of the most common words in common. All of them will also have at least one word that does not appear in any other texts. And then in the middle of these two extremes there will be variations of counts for each of the words. The standard Stylo process claims to count word-frequency, but what level of frequency is relevant (is there a minimum number of appearances?) changes between studies, and how these frequencies are weighed in significance is never clearly explained. Comparing all words in all texts in frequency against each other is like comparing full medical histories for 284 patients to figure out which of them are related to each other; you might hit on a few obvious cases where folks have hereditary diseases, but most of the data will be useless noise. If you are sure there is any rational basis for this word-frequency for all words comparison, try posting the raw data for all comparisons and explain every step how you (or Stylo: the steps the program takes) arrived at the attributions for at least 10 texts to start (if you have not attempted to do it on a larger corpus).

>298 prosfilaes: You are repeating the same false statements. I will yet again correct you. Yes, my method is tested, and I have tested it against works with firm bylines. There are no bylines that are so firmly established that a miss-match to one of them proves a method wrong. What is important is that the method finds results that are consistent across the corpus, so that the same style is consistently identified and external evidence confirms all conclusions. This is the case with my method and findings, as I prove across BRRAM.

Your "other tools" are faulty or have bugs in them and are unusable. I tested all of the major tools you have been advertising across this thread to come to this conclusion. I am not going to re-invent your tools to prove that my better tools already work as-is.

If there is a 3D printer that already exists, it is a better idea to use this 3D printer to print out an original shape for a new invention than to invent one's own 3D printer before printing the shape one is contemplating. Once something is invented, it is absurd to invest effort into re-inventing it just to gain credit or traffic.

The results are "extraordinary", and it is also extraordinary that all of you are responding to them with insults instead of reading the evidence that firmly substantiates them.

The "extraordinary" nature of my findings is the result of the extraordinary volume of labor I put into perfecting and expanding the argument into 17+ evidentiary volumes. When I and Petroglyph tested the same number of texts and bylines, my results were mostly similar to his, with only some evidence of co-writing, and the deletion of the "Emily Bronte"-byline as a pseudonym or part of the same authorial-signature as "Charlotte Bronte". If my method was designed to shrink the number of attributions, I would have ended up with only a couple of ghostwriters in this corpus. My method works and derives the precise number of actual authorial signatures in a corpus of any size from 10 to at least 284.

Yes, "history" is regularly re-written as scholars discover archeological digs, or find new evidence, or new ancient manuscripts. There is nothing innately problematic about re-writing history, and I provide historic evidence outside of my computational-linguistic data for why the current history is a fictitious falsehood.

I have no idea to what poetry you are referring to. If I tested this poetry and made the data available, the attributions I made are the correct ones. I have not "backed away" at any point in this thread; I challenge you to prove that I did.

Yes, an "exiled" author wrote most of England's theological works on all sides, such as for and against prosecuting witches. In a book I reviewed this past week I learned that Verstegan used a similar strategy when he wrote anonymously for and against draining the fens to establish the side against draining was using threatening language and needed to be silenced. The draining included confiscating common lands from the people who lived on them, and Verstegan was hired to make it look like a debate had taken place before the paying side won.

I tested 284 texts and reviewed many others and I did not spot any unique texts between 1570s and 1640s that could have possibly been written by anybody other than the 6 ghostwriters. This is not a "simple" answer, as I support this claim by explaining that these ghostwriters directly manipulated this market. For example, Byrd held a poetry/music monopoly with the state. In a largely illiterate society, why would the monarch be literate, when the monarch has a right to a throne from birth or close to it, and so he or she has power without having to invest any energy or intelligence to gain it.

If your side of the argument depends on the Holy Ghost writing; present proof of this Holy Ghost at work and then you are starting to support this position.

You are hoping "maybe" you can make me quit? Quit what? I am in the process of translating a 400-page book (Verstegan's "Restitution") during the surrounding months. My computational-linguistic method has already been applied and already solved the Renaissance attribution problem. I proved this case in Volumes 1-2. You have not even read more than a few paragraphs from this study. I am not going to quit working. This is year 17 of me working non-stop on my research. The question is really why are you here interrupting my work by repeating nonsensical babble that you have previously stated and I already replied to?

>298 prosfilaes:

You chose a section of poetry you claimed was obviously by the same author that wasn't clearly by the same author, and you backed away when challenged.

Yes, there are only two kinds of reactions when she is challenged: either she backs away and claims that she never said whatever it was (as if the rest of us can't read – those ridiculous claims about "appresso", for instance), or she simply ignores the explanations that could have helped her understand what it is she is doing, and goes on to claim that nobody has been able to challenge her methods, or results, or definitions.

>294 prosfilaes: Re your question about >287 faktorovich:, where she said "the tests for word-types are not a grammar check. It does not matter if the words are categorized in the proper categories that fits the intended meaning in the texts. These tests are mathematic sorters that group words into categories A, B, C... As long as all words in the dictionary are consistently grouped in only one of these categories, they are still being sorted accurately." asking her about it is not going to be meaningful, since she has fundamentally misunderstood a few things. She is trying to justify the conflicting facts that a) part of her "attribution tests" has to do with how frequently an author uses nouns, adjectives, etc, and b) that the tool she has opted to use miscategorises a lot of words (and not in a consistent manner), so she can't actually say anything about the frequencies of nouns, adjectives, etc in her tests. "all words in the dictionary are consistently grouped" is pretty much gobbledegook. It's anybody's guess what she means by "dictionary" since dictionary definitions are not what's relevant here, and the whole problem with using the AMW tool is that is doesn't analyse the parts of speech consistently. As has been explained to her a couple of times. (What I don't get is why she doesn't use WordSmith, a pretty powerful tool which she has said that she uses to create word lists, to analyse the parts of speech. But it probably doesn't create colourful bar graphs.)

But to try to answer your question, if by "the tests for word-types" she meant part-of-speech tagging, any sophisticated automated tagger is going to use a combination of the parts of speech associated with a word in a dictionary, and syntactic analysis. For example, if the POS tagging software finds a phrase where a definite article is followed by a word that could be a verb or a noun depending on context, it is probably going to guess that the word is a noun in that position.

-Linnéa

>300 faktorovich: I believe the reference was to quitting this thread; that is, no longer posting responses here.

Your consistent lack of understanding of what is quite plain writing is unsettling, since you claim never to have made a false statement or, apparently, been just plain wrong in any statement you have made.

That sort of thing (refusing to admit even the possibility of error on your part) really reflects poorly on your status as an unbiased researcher who has covered information that overturns a huge part of literary history. You don’t seem to grasp the fact that one who makes that sort of claim must provide proof. Your stating that you’ve spent hundreds (thousands?) of hours doing the research and proving your predetermined theory is not proof. Nor is 17 or 27 or 100 volumes proof that what you claim is true. Your conclusions don’t turn out in other programs with long histories of scholarly work; therefore those programs are in error, unusable, full of glitches.

If your theory is indeed the correct one, you should be able to duplicate the results across the board using any of a number of existing research tools. The fact that you can’t doesn’t mean that all the tools are wrong; it means that your conclusions were arrived at erroneously, or prematurely.

No one is trying to force you to stop. The only interest here is to have something besides your self-designed systems as proof of your ideas.

Please do let us know as soon as you have such proof and evidence.

Edited to add the post # to which I am responding.

>300 faktorovich:

I have no idea to what poetry you are referring to.

Here: post 561 in the previous thread. Conclusively shown to be nonsensical, in the following posts – and here in this thread there is even more evidence that your claims in post 563 are based on incorrect assumptions.

If I tested this poetry and made the data available, the attributions I made are the correct ones.

If you imagine for a second that your attributions are not correct, what might be an alternative explanation for your results?

-Linnéa

>300 faktorovich:

In a largely illiterate society, why would the monarch be literate, when the monarch has a right to a throne from birth or close to it, and so he or she has power without having to invest any energy or intelligence to gain it.

Why?

Well, first of all, many monarchs of the early modern period would NOT have been expected to achieve the throne. Henry VII got there by overthrowing a previous monarch. Henry VIII got there because his older brother died. Neither Queen Mary nor Queen Elizabeth were assumed to be heirs; Henry kept trying for a son and got one. James VI and I could have been expected to become king of Scotland, but his son, Charles I, also became king due the death of the heir presumptive.

More importantly, any reasonable monarch would want to be literate. Being king was a treacherous business, and literacy would be an important tool in keeping an eye on, and control of, your advisors and others who might be a threat. So even assuming, arguendo, that you didn't need to "invest any energy or intelligence" to gain power, you did to keep it.

And, of course, the part of society that was "largely illiterate" was not the part to which the monarch belonged. Elites were highly literate, and not merely in English. We know that Elizabeth I, for example, was multi-lingual. Edward VI also studied multiple languages, as did James VI and I.

>296 faktorovich:

My corpus has 284 texts with 7.8 million words. A comparison of all of these texts against all others on the frequency of each of the words would be approximately = 284 * 283 * 7.8 million = 629 billion comparisons. If you try imagining the size of this 3D table that would fit all of these comparison scores on each of these words for each of the texts, you will notice why the idea of measuring all words in all texts for frequency is an absurd idea.

Other than showing you can use a calculator to multiply some large numbers to get even bigger numbers, what is the purpose of this? What does it really mean?

For example, if you had a text of 46,418 words (a real example in my corpus) and overall there are 1,607,892 in 39 works, I don't have to compare 1.6 million words. I need to know which words are unique to each text. In this case for the one story it is 3,909 words for the longest text, not 46,418.

Text file manipulation has been refined for more than 50 years in Unix type system (yes, since 1970). There are many tools of varying levels of sophistication and complexity that can be used, one after another in a pipeline, to get answers.

Let's say you have files with these names:

% ls
TS01	TS04	TS07	TS10	TS13	TS16	TS19	TS22	TS25	TS28	TS31	TS34	TS38
TS02	TS05	TS08	TS11	TS14	TS17	TS20	TS23	TS26	TS29	TS32	TS35	TS39
TS03	TS06	TS09	TS12	TS15	TS18	TS21	TS24	TS27	TS30	TS33	TS36	TS40

and you want to know how many words are in each (and sort them from largest to smallest for the illustration):

% wc -w * | sort -rn
 46418 TS34
 45695 TS35
 45556 TS22
 45208 TS36
 44939 TS23
 44823 TS19
 44517 TS29
 43945 TS13
 43763 TS18
 43742 TS31
 43460 TS11
 43360 TS28
 43265 TS38
 43121 TS04
 43112 TS20
 43088 TS33
 42935 TS30
 42665 TS02
 42651 TS03
 42436 TS16
 42369 TS24
 41892 TS05
 41723 TS26
 41706 TS01
 41690 TS27
 41564 TS08
 41526 TS14
 41482 TS12
 41402 TS25
 41337 TS32
 41308 TS10
 41135 TS07
 41041 TS21
 40986 TS15
 40442 TS09
 40411 TS06
 40310 TS17
 14469 TS40
 12400 TS39

 1607892 total

But now you want to know how many unique words in one of those files. I'll pick the largest and store the result in another directory to avoid too much clutter:

cat TS34 | tr -cd "\:alpha:\\:space:\-'" |
 tr ' \:upper:\' '\n\:lower:\' |
 tr -s '\n' |
 sed "s/^\'-\*//;s/\'-\$//" |
 sort |
 uniq -c > ../ts_uniq/TS34u

This puts all of the unique words and counts in a file (uniq -c). It is alphabetical by the word because of the sort command. I manually set the destination file path and name but this could be done with variables but I did not want to give too complex an example. There are leading spaces that LT Talk won't show but the first few lines look like this:

% head TS34u
1020 a
 1 aback
 4 ability
 11 able
 9 aboard
 158 about
 12 above
 1 absent
 1 absently
 1 absolutely

But maybe it is more helpful to have a sorted list by the frequency of a word in the text:

% sort -rn TS34u > TS34us

That new file TS34us has these for the top lines:

% head TS34us
2721 the
1454 to
1060 and
1020 a
 998 of
 831 tom
 815 i
 721 in
 667 he
 635 was

and these for the bottom lines:

 1 accommodations
 1 accommodated
 1 accommodate
 1 accidentrear
 1 accessories
 1 accepting
 1 absolutely
 1 absently
 1 absent
 1 aback

I see that Tom is used 831 times in this story and is the 6th most common word in this work. It is context-specific and these works routinely give the name of the character when referring to them. It is part of the style for this genre. The other words are ones that are common in English.

Looking at the bottom lines I see something unexpected accidentrear which is not a word but is probably a case of a typo in the source text. So I can look to have it tell me where the word appears in the file in my corpus directory:

% grep "accident—rear glider" TS34
accident—rear glider cut loose and is falling—hope they make a safe

I did not find it with "accidentrear". By looking at "accident" I found all of the occurrences of the word (52) and could spot the troublesome line. In this case, the Gutenberg typesetters put an em-dash that was not one of my characters for separating the words. I can edit the source text or change the lines of code to produce the lists of unique words.

Let's consider aback and I want to know how common that word is used in the other texts of the corpus:

% grep aback *
TS32:“Oh, did you?” Tom was rather taken aback by this reply.
TS32:“CUNNINGHAM!” gasped Tom Swift, taken aback as much by the boldness of
TS34:This seemed to take the angry man aback. He swallowed his feelings with

So it was used twice in TS32 and only once in the longest text TS34.

For a list of the file names and number of occurrences of aback I could:

% ls | xargs grep -c aback | grep -v ':0'
TS32:2
TS34:1

I could write a small loop to take the 100 (or 1,000) most common words and find out which works have those words.

The investigator needs to decide what questions to ask and learn how to get the results. After performing some sanity checks, you then have to be sure they are meaningful and can't occur from other things that are unrelated to the question being asked.

Does the weather in Buffalo, NY influence the weather in San Diego, CA? Perhaps. A little. But generally you want to focus on weather elements within 100 miles and the larger-scale systems like storms from the Gulf of Alaska that are carried south along the Jet Stream. Solar activity, including sunspots as an indicator of same are factors but probably not "Mercury in retrograde." Just because you can measure something does not mean it is significant for the question at hand.

Using the right tool for the job is important. A screwdriver might be used as a hammer but it is not as good as using an actual hammer. The complaints raised about your free website tools come down to whether they can do the job they claim to do — accurately identify and count certain parts of speech, punctuation, etc. If they have problems, anything that relies on them also falls apart.

Based on your posts in this two-part thread about your experience with Stylo for R, I don't for a moment believe that you used it enough to determine that it is filled with bugs as you claim. It requires spending more than 30 to 120 minutes before giving up and declaring it a failure. If you wrote a novel with Microsoft Word and did not manage to sell it to a publisher, is it the fault of Word?

James

As I have written before in this two-part thread, in 2002-2003 I wrote a PHP program to implement a stylometric system that showed promise. In my limited tests, the system did seem to work for texts with well-established extrinsic evidence (contracts, correspondence, etc. from the private business records for the Stratemeyer Syndicate now held at NYPL.).

But one of the disturbing things that I (and others) raised at the time was a theoretical basis to explain why the 10 or so tests used by the creators of the system would be meaningful to authorship attributions. For example, one might make a case that shorter words (2 or 3 letters, 23lw, but 4 letters in some tests, 234lw) could be mainly function words that help to join together the more contextual or lexical words in the content. But some of the prescribed tests counted these plus words that begin with vowel (ivw or initial vowel words). Why would the rate of use of "airship" (ivh) be counted the same as a word like "was"? The creators of the system who published a book on the topic and used it extensively, and with whom I corresponded some 20 years ago, could not offer an explanation or theoretical basis as to why these should matter. That caused me and others to have some doubts about the findings.

Inspired by this interesting thread, I decided to see what, if any, consensus there was on the QSUM system. Was it still in use? Was it respected? Was it admissible in evidence in British courts still as it had been in the 1990s and early 2000s? While I didn't find complete answers on all of these questions, I did find some critical (i.e. "negative") articles about QSUM and get a sense that someone interested in advancement in academia probably is not going to gain respect in their field by relying on QSUM.

This article, in particular, was written after my PCA presentation and it makes some careful and specific criticisms of QSUM which I find to be convincing. But the real reason to bring it up is what the author has to say about the importance of careful attributions since it is easier to make a claim to add a text to an author's canon than it is to remove it once it has begun to appear in major published editions. The last couple of pages offer some suggestions of ways to test any new method of authorship attribution that are relevant here.

https://www.jstor.org/stable/40372114?seq=1

As you know with JSTOR.org you can sign up for a free account, even with a Google account, and get permission to read (on screen) page images of 100 articles per month. Since most of the 24-page article focuses on the details of QSUM and the critique of it, I'll leave that to the reader who may be interested. It is not a complete presentation on the method so for that look to Analysing for Authorship by Jill M. Farringdon, et al. Instead, let's look at what is said on pp. 263 and 285-286:

Because attribution study lays foundations for others to build on, those who attribute authorship should be cautious. If and when these foundations crumble, the scholarly labor already performed is proven to be a waste of time and effort for all of us. Witness the energy expended to argue that Shakespeare wrote A Funerall Ellegye (1612): the attribution resulted in the poem's appearance in standard Shakespeare textbooks, from which it has begun to vanish now that some specialists believe that John Ford wrote the poem. But who knows how long it will take for this poem to become fully dissociated from Shakespeare. Perhaps never.

Can a new attribution method pass a "falsification test"? In Karian's example, if one selects texts believed to be of consistent authorship, if one text is removed from the corpus and tested as if it is an unknown, does it work? Repeat this process for multiple texts. Several other good suggestions follow. The final paragraph seems particularly important:

Finally, I would make a plea that proponents separate their arguments in support of the method from its application in particular cases. That is, proponents should first independently publish an account that describes the method and its limitations as well as how it responds to known and accepted cases of authorship. This step should give the scholarly community an opportunity to examine tlie method and respond. Only after a reasonable time to allow for this exchange should proponents begin to offer attributions (or de-attributions) of particular examples. I hope that this delay would decrease irresponsible claims of authorship.

James

>301 anglemark: An inconsistent manner of tagging would be if the word "explained" was not always categorized as the same part of speech; as long as each word in the dictionary is always tagged as the same part of speech whenever it appears, this is a consistent manner of tagging. Again, the goal of my method is the attribution of a text, and to "say anything" about the grammar of a text; the method is not a verbal description of the author's precise grammatical usage, but rather a quantitative system designed to check if the quantifiably linguistic measures are similar or different between the texts in the corpus.

The free version of WordSmith that I have has limited access to the features of WordSmith. I would not want users of my method to have to purchase a WordSmith version to use my method so I cannot switch to a paid service. The only tutorial I could find on how to perform parts of speech tagging with WordSmith is this unrelated page: https://lexically.net/wordsmith/version4/screenshots/index.html?concordancingtag... You guys have not come up with a simple step to apply my method in CLAWS without adding a program, so I am assuming it would take creating a program and paying for software to tag with WordSmith as well. The character counts is the only measure that is automatically accessible in the free version. There were glitches with the other tools I experiment with in the WordSmith program when I first considered which of their tools would be best for utilization.

If modern tagging use word-order to determine part-of-speech, they are likely to make mistakes with the Early Modern English texts because they use different rules of order, including reversals of verb and noun order.

>302 Matke: Everything you are saying is either nonsense or points I have addressed before. I have proven my conclusion beyond all reasonable doubt. There is no rational way to argue with willful ignorance.

>303 anglemark: My attributions are correct, and I made a detailed explanation in post 563 to explain them; your hypothetical question is irrelevant.

>304 lilithcat: Yes, that's why James I had books ghostwritten for him to increase his chance of attaining the throne when he was not next in line, and also to unify the Scottish and English kingdoms. Being king is only "dangerous" for somebody who has read the histories of past tragedies that have befallen kings. There is bliss in ignorance, so an illiterate king/queen would not be aware there are threats they could prevent by reading about how their predecessors died. I do not know why you are using the royal "We". Are there many of you writing in your account? Are you a monarch? Or are you suggesting this is an established fact? The claim that Elizabeth was "multi-lingual" was made in the Workshop's contemporary pufferies of her intelligence, for which she paid them pensions and grants; she paid at least one of them, Byrd, directly. How about if I write this puffery in the third person about myself under a different byline, would you believe it to be a fact?

"Anna Faktorovich is the most brilliant linguist that has ever lived; she has studied every language on the planet, and speaks 550 of them fluently." --Dr. SmartestMan OnEarth

>305 Keeline: "I need to know which words are unique to each text." Do you understand what you are saying with this statement? The words that are "unique" to each text are those that do not appear in any of the other texts. "Unique" means: "being the only one of its kind; unlike anything else." These single-appearance words cannot be used for attribution because they always only appear in a single text, and thus there are no texts in the corpus that can be proven to be similar if one only considers these single-occurrence words.

You have to verbalize the steps you are taking. It looks like then, you are calculating the percentage of the "head" words? "In English grammar, a head is the key word that determines the nature of a phrase (in contrast to any modifiers or determiners)." There is no logical reason nor is there a tagger that could accurately only pick out the head-words in this sense, so you must be trying to do something.

It is clear that you are not counting "unique" words, as you are providing what looks like the beginning of a table of the count of the frequency of appearance of all words in the text, from the likely most-frequent-word "a", to words that only appears once like "aback". If you had only counted "unique" words in each text, there is no chance your list would have included "a", as it is present in all texts I have ever read in the English language.

Then, you correct this list by ordering it in the manner that is relevant for my 6 most-frequent words, with the 6 most frequent words at the top. And you now have the likely to be most unique single-occurrence words at the bottom of this list.

Okay, then you finally arrive at the actual comparative method you are using. However, you are not giving the steps to compare every single word in the corpus against each other, but rather focusing on analyzing an individual word and its usage in only 3 texts. If you have to manually compare millions of words in a large corpus like this, you would have a list of billions of similarities and non-similarities for individual words that would not actually establish any signature conclusions.

Now you have gotten to the programmable part of your method. "I could write a small loop to take the 100 (or 1,000) most common words and find out which works have those words." First, there is a big difference between 100 and 1,000 of the most-common words as if you approach 1000 * 284 * 283 you are slipping into over 80 million comparisons, or 8 million for 100. Secondly, you are going to end up with a nonsensical number of matches and non-matches for each word. Imagine that each match is a string going between a 3D box of rows and columns with as many as 80 million strings (if all texts are identical), or at least a million strings if there are many unique words that only appear once in a text. Creating this crisscrossing box of strings with a program is easy enough, but how are you going to figure out if a match between texts on 50 words with varying frequencies between 30 and 1000 is more indicative of an authorial match than a match on 100 words with frequencies between 1 and 50, or any of the other myriad of possible combinations? You are missing this essential step in your process, without which you are just finding a chaotic set of matches and non-matches, and not the answer to who-wrote-what. But instead of answering this basic question, you digressed across the rest of your response into philosophizing on why you are the smartest-human-of-them-all.

Doctor Faktorovich, back in >59 faktorovich: you said, "And it is absolutely absurd to demand for a computational-linguistic method to be 'easy'." The tone of that post was that you did not want the method to be reproducible. Yet in >307 faktorovich: you claim you do want the method to be so easy anyone can use it. Which one is it?

>306 Keeline: My method accurately attributed the Renaissance texts because I compared the largest corpus ever attempted from this period of 284 different texts with 104 different bylines. As I have explained before, the "Funeral Elegy" and other previous re-attributors of individual texts have compared them to only a few other bylines or texts, and thus whenever they change their corpus to a new set of a few different texts with a few different bylines, they reach a different conclusion. The existence of only 6 ghostwriters during these decades explains why this is the case, as it is easy to pick any random set of 10 texts in the same genre and at least one of them will match the mystery-text you are trying to attribute because with only six ghostwriters probability is in the favor of at least one match. Especially when comparing dramas, Percy and Jonson wrote significantly more of the famous (mystery) dramas than the other ghostwriters, so it has been easy for attributors to ping-pong those texts between bylines (especially if neither Percy nor Jonson themselves are in the corpus, and the corpus is too small to notice, or for the researcher to care that there are other strange cross-byline matches).

It would be amazing if a precise description of an accurate method was all that was needed for scholars to look back and acknowledge they have miss-attributed the entire British Renaissance. But it seems that scholars in this field are too vain to consider any volume of evidence as sufficient to prove them to have been wrong. Previous re-attributions have indeed been irresponsible, and the 3-year delay since I first explained this method and submitted it to journals for review is too long for this review process to be based on any rational consideration for truth.

Regarding your JStor article, I solved the Defoe mystery in my 300,000-word 18th century book to which I will return to retest a much larger corpus after this project is finished.

>312 Stevil2001: My method is reproducible and easy to use to check which texts are like and unlike each other. The difficult part, as I explained in post 59, is figuring out which of the thousands of possible bylines is the correct author when a corpus has a complex combination of multi-byline ghostwriters. Each byline's biography and the historic implications behind their potential authorship have to be weighed. This non-quantitative research stage takes years, whereas the computational-linguistic data processing takes a few days, weeks or months at-most.

>308 faktorovich:
There is no rational way to argue with willful ignorance.

How true.

>311 faktorovich: First, there is a big difference between 100 and 1,000 of the most-common words as if you approach 1000 * 284 * 283 you are slipping into over 80 million comparisons

It's Over 9000!

https://gist.github.com/Prosfilaes/b00d5d4c1164df147b4703d24dbae635 compares all words between all texts given on the command line. It's 50 lines of code banged up in a couple hours. To process 300 works with 24 million words (from Project Gutenberg) took 6 minutes, and there's obvious steps to make it much faster.

I don't recommend anyone use this code; it's a naive approach that produced mediocre results on a tighter test sample, marking The War of the Worlds as closer to The Gods of Mars than the First Men in the Moon, among other things. (It lacks pretty output, or even code to group or tree the results. Just raw distances.) It's just you're so confident about these things even when you're so wrong.

>311 faktorovich:

Oh and "crisscrossing box of strings" what a weird way of looking at it. You are just manipulating sets (in a CS sense) of tokens (words). It would be very easy to get a result tsv file with every word on a new line with columns consisting of the count for all the texts. This is simple.

Of course the questions you ask this data are important - you can ask sensible questions or nonsensical questions of the data. But producing that data from the raw texts is pretty quick and easy (even for all words). No manual steps are needed. The runtime will also be a lot quicker than you expect (see prosfilaes's runtime for their code). Obviously this piece of code could not possibly supply all the processed data you might possibly need (some questions require more than just word frequency comparisons amongst texts) in order to be able to write your magnum opus. But it is just as easy to put together more programs to produce the data for those questions.

You continue to be factually wrong about how feasible it is for a normal practitioner to be able to write these sort of programs. You seem to have a bad case of NIH (not invented here) syndrome and favour very manual systems over something which a computer program is much better at.

>307 faktorovich:

An inconsistent manner of tagging would be if the word "explained" was not always categorized as the same part of speech

Yes, exactly! As pointed out above, (for instance) "you" is sometimes tagged as an auxiliary verb and sometimes as a pronoun. "Explained" is sometimes used as an adjective in the period you are looking at, even though it is nearly always a main verb, so it shouldn't always be tagged as the same POS. And if "explain'd" and "explained" are tagged differently you have an inconsistency there as well. Do you lemmatise your texts at all?

You guys have not come up with a simple step to apply my method in CLAWS without adding a program, so I am assuming it would take creating a program and paying for software to tag with WordSmith as well.

Nobody has claimed that CLAWS doesn't need a tool to display the relevant statistics. It's a tagger, which tokenizes a text and generates a text where each token is tagged for parts of speech. There are multiple free software tools that will generate the kind of statistics you are interested in, based on a POS tagged text, and you have been provided with links to websites that list such tools. I didn't know there was a free version of WordSmith – it's not a tool I have used for many years, myself, because as far as I'm aware it's Windows only – but I do know that even the older versions accept POS tagged texts. Maybe it doesn't generate all the output you'd like to have, but as I say, all you need to do is use another free tool of the many that are available to you.

>309 faktorovich: That is always a relevant question that every scientist and scholar needs to ask.

-Linnéa

>301 anglemark:

I had a brief look at the usefulness of Parts of Speech for author attribution in this post, and tried both a detailed set of 50-ish POS tags (courtesy of TreeTagger), and the 7 that AMW uses. Conclusion: For small corpora with limited authors they're useful, and the more detailed the POS are, the better the results. But the more texts you add, and the more authors, the worse things get.

prosfilaes (#316):

It's not that the results of a word-frequency comparison are mediocre, it's that they're more geared toward finding topical similarity than authorial similarity. This type of approach is intended to be author-agnostic and topic-sensitive. Authorial idiosyncracies that persist across topics, things like sentence length or an inordinate fondness for semicolons, aren't going to come up well in this type of analysis.

(And distance-between-texts by overall word frequency analysis is so boringly standard that very early in this first thread it's what I assumed faktorovich was doing, just because it's the most obvious imaginable approach.)

More generally, one of the many problems with faktorovich's approach is that there is no ground truth, no test case. If you're trying to do topic modeling, you have some human-labeled texts, and see how well your method reproduces the labels, and tweak to optimize that. Or you have someone go back and look at what your model decides is about space travel, and sanity check it. But faktorovich has decided that there is no amount of historical documentation that would be adequate ground truth to check that her methodology correctly reproduces known authorship - no, no, it's the history that is wrong!

faktorovich (#300):

If there is a 3D printer that already exists, it is a better idea to use this 3D printer to print out an original shape for a new invention than to invent one's own 3D printer before printing the shape one is contemplating. Once something is invented, it is absurd to invest effort into re-inventing it just to gain credit or traffic.


This is what we're trying to tell you. But you're the one trying to invent your own.

>321 lorax:

Faktorovich wrote:

If there is a 3D printer that already exists, it is a better idea to use this 3D printer to print out an original shape for a new invention than to invent one's own 3D printer before printing the shape one is contemplating. Once something is invented, it is absurd to invest effort into re-inventing it just to gain credit or traffic.

No. There is an endless group of 3D printer models with small changes and refinements because none is really ideal and general purpose (suiting all needs). There are large communities of people designing and building these and producing plans, kits, or ready-made products. The latter are often the weakest of the bunch. Facebook and other places have large and extremely active forums to discuss the latest refinements and help people decide which 3D printer should be their first or next to buy.

Closer to our field because is it part of the process to create many of the electronic texts analyzed is the community of people developing and refining book scanners. One such group is https://diybookscanner.org/ where one can find designs and plans or kits or learn of commercial products. The software to help with various stages of the workflow is also extensively discussed.

Of course there are special interest groups to discuss practically anything. The Internet and platforms that use it draw people of like interest together. Where they once had to exchange ideas in printed magazines or annual gatherings, now they can visit a forum and see the latest things people are talking about.

James

Keeline (#322)

The part of the message that you have attributed to me is the part that I was quoting from faktorovich. Attributing it to me directly is extremely misleading. Please edit your message.

>316 prosfilaes: You have an extremely slow computer if any computer process takes 6 minutes.

You are still not verbalizing what this program does. It can compare the word-frequencies for each word in each text to come up with a 0/1 of if a word is or is not in each of the texts. But then how does it figure out which texts are alike, since all texts will have some words in common and some words that are in none of the other texts, and many words with random degrees of frequency match/non-match or 0/1. If you do not know what a program you are pitching does, including the formula it is using and can vocalize the steps it is taking to carry out this formula; you do not understand why or how the system is giving you attribution conclusions, or if it is giving you anything to do with an attribution or a bunch of individual word comparisons. The "just raw distances" is precisely what I saw in the raw data when one of the top computer-linguists shared his raw data with me as we tested a set of texts to see if his process worked. When I began asking further questions, he cut off all communication. He also could not explain just how "raw distances" between 100 or 1000 different words in hundreds of texts can possibly establish a non-chaotic attribution conclusion.

>324 faktorovich: You have an extremely slow computer if any computer process takes 6 minutes.

Hahaha. I've not had such a good laugh for ages. You have put your ignorance on show yet again.

>317 andyl: If you are working on a math problem that has 3D data and you are not visualizing what the calculations you are attempting to apply is actually doing; you have a "weird way" of ignoring mathematical reality. If a solution you have come up with would look incorrect to a kindergartener, you are not on the right track. The correct solution is one you can explain to a child visually. If even a child looks at a ball of a million strings between 284 texts that represents matches at random frequencies and asks "what is that supposed to do?" you should re-think your method.

Yes it is simple to come up with a frequency count for all of the words. You are stuck on this point. The comparison step is where the problem comes in with this pop approach to word-frequency attribution.

I have nothing against computer programs. The program with your method is not that you use computer programs, but that the mathematic formula or comparison process you are telling your program to use is incorrect. The method I came up with is the correct formula for precise authorial attribution. I just have not (yet) created a program to make this process a one-step input-output answer. A program for the 27-tests would be a lot more complex than a single-test word-frequency comparison program that is measuring simple frequency. My word-frequency test requires far more steps of a potential program, as it has to calculate the 6 most frequent words, and then sort them into patterns of which words appear among these 6 and then compare these patterns between all of the texts. A full program for all of the steps would also have to clean up input texts of glitches such as lines over characters. And it has to create a full set of data tables to present the data in an easy to use way. So it would basically require the same amount of programming-time to make it usable as a website as a video game. There is no need for me to create such a program, since the manual method works to achieve the same result, and this program would be free to the public so it would be a charity-project that nobody has asked me for.

>325 andyl: I (Johan) will tell our developers this! We have to redesign our test suites and improve our source code so it doesn't take so long to compile.