The failure of a mathematical approach to Shakespeare’s authorship | Hacker News Reader

The failure of a mathematical approach to Shakespeare’s authorship

the-tls.co.uk

65 pointsHooke11 comments

Note: This is not about the "anti-stratfordian" conspiracy theories. It is about the well known fact that certain of Shakespeare plays was written in collaboration with other playwrights.

Scholars have discussed at length exactly what scenes are written by who. In recent years computer-based text analysis have been used to discern between different authors based on word frequencies and other quantitative characteristics.

This article posits that quantitative textual analysis is basically pseudo-science. It provides the example that two different quantitative analyses yields different attributions of scenes for Henry VI. So at least one has to be wrong, but possibly the whole approach is flawed.

A sort of aside: I'm no expert here, but every time I've run into the whole question of who authored Shakespeare's work the basis seems to be "I don't think this nobody could have read it, and if we assume he didn't all these other ransom theories sort of make sense!".

That is to say ... there's not much there to support much in the way of conculsions at all.

So I guess this makes sense that some weird math also doesn't help.

The "anti-Statfordian" theory say that William Shakespeare (who we know existed) didn't actually write any of the plays, but acted as a front for some other author. If this mysterious real author wrote all of the texts attributed to Shakespeare, there is really no way to prove or disprove this through any form of textual analysis.

But as noted, this is not the question the article at hand is concerned about. No legitimate scholars are taking that theory seriously anyway.

I think it is worth adding that the author is not making a simple "machines can never understand Shakespeare" claim, but that the techniques used here are being interpreted beyond what they can achieve. At one point he writes "it may well be that computerized textual analysis will ultimately go beyond treating word frequencies in isolation and will be able to cope with the full range of language. But that time has not come..." The article also quotes Pervez Rizvi, a software engineer who has collaborated with the author in using computers where appropriate, who makes several trenchant criticisms of Taylor's interpretation of computer-generated evidence.

Any editor of the New Oxford Shakespeare may make changes without needing the consensus of the others. It now seems to be experiencing some of the problems arising in Wikipedia from that policy.

Do these methods have any history of, say, being able to successfully predict things where we have a known "ground truth"? We have plenty of cases of famous authors writing under pseudonyms for positive examples, as well as having considerable ability to know that certain authorship is unambiguous (e.g. much modern writing may have involved multiple drafts being retained somewhere, or have been done with witnesses, etc). So we can probably safely say that most of the authorship of 20th Century novels is fairly well-known give or take a long-suffering wife or two ("who patiently typed up my genius"); do the same methods here actually work to predict things that we already know?

Or will they, with appropriate levels of massaging, ascribe portions of Dashiell Hammett's oeuvre to Raymond Chandler?

for example that j.k. rowling crime novel written under a pseudonym was discovered by just a random astute person who commented somewhere(maybe on the amazon reviews) that it felt like j.k. rowling wrote it.

it would be cool to try this method and see if it could detect anything.

> discovered by just a random astute person who commented somewhere

I don't see that this really counts as 'discovered'. I'm sure thousands of people have said It felt like it was written by Rowling about many different books.

That's not quite how it happened. People spotted that Galbraith and Rowling had the same agent and the same editor.

https://en.wikipedia.org/wiki/J._K._Rowling#Cormoran_Strike

Stylometry https://news.ycombinator.com/item?id=20621402

Amateur Shakespearean here. I've often wondered why these statistical approaches don't look for stylistic signatures. For instance, several plays are linked by their use of doubles (in Hamlet, this scales from "too, too solid flesh," up to the running joke on the interchangeability of Rosencrantz and Guildenstern). Shakespeare also seems to be especially fond of puns on "bear" and "born" -- the "to be or not to be" monologue has three or four of these. There are also ways in which scenes repeat their designs -- 1.2 of Hamlet is similar in some ways to 3.2 of Julius Caesar, etc.

I imagine two things that make simple analysis harder.

1) once an author has a sufficient library, hiring a team to emulate them is easier (Tom Clancy, Hanz Zimmer.)

2) if in a collaboration, a partner could write a skeleton or first draft, and the main author could edit and fill in the cracks. Conversely they could write scattered sentences and have their partner flesh out filler.

Reply on news.ycombinator.com