Was Shakespeare Marlowe? What does stylometry say? [duplicate]

Question

I am not sure why this isn't asked here already, but Peter Farey and Ros Barber convinced me that Christopher Marlowe wrote the works of Shakespeare.

This resolved my own cognitive dissonance from many years ago, when I picked up a copy of Tambourlaine, and immediately was annoyed that it was stylistically indistinguishable from Shakespeare (except for some minor maturity issues). At the time, I concluded that Shakespeare was just ripping off Marlowe, but it was a rip-off too methodical and too shameless to be consistent with literary genius, so I wasn't sure what to think of the bard. Of course, it's not a problem given that they're the same person.

This requires you to believe that Marlowe was not killed in 1593, but kept on writing, exiled in continental Europe. I believe that this is perfectly reasonable, considering the privy-council's death threat to Marlowe, his known connections, and the similar shenanigans in the McCarthy era. My question is not about this stuff, which I think is firmly on the Marlovian's side.

My question is simply about the stylometric evidence. Farey has compiled his own page on the stylometries (you can find two plots here: and there are more here ), his analysis concludes that any stylometry that claims to distinguish between Shakespeare and Marlowe's style is better interpreted as a maturity curve for a single writer in two periods of life.

I want to know how reasonable this conclusion is. I haven't checked the figures, and I do not know if there are other stylometries which are missed by his analysis. Further, is there any known way of establishing single-authorship from large batches of text, when the text is written several years apart, but is essentially infinite in length?

How good is this new stylometric evidence? What confidence level does it give to the assertion that Marlowe was the author?

There are previous stylometries which are weaker, namely Mendenhall's letter distribution, which was a major foreshadowing of modern Marlovian theory. I am not so interested in this, since this is well known to pick out Marlowe as the author of Shakespeare.

Answer 1

Not a complete or final answer to your question, but a recent (2012) and seemingly rigorous statistical analysis of works by Shakespeare, Marlowe, and other contemporaries, in order to distinguish authorship, is here:

Their key conclusion:

...our results are best explained by the assumption that Marlowe is not Shakespeare.

http://www.cs.brown.edu/research/pubs/theses/masters/2012/ehmoda.pdf

The statistical comparisons were based on 1) general vocabulary and 2) a combination of "function word frequency, frequency of part of speech tags among words that are not on the function word list, and bigram...called 'two-word collocations'..."

The comparison tests "misidentified" four out of seven of Marlowe's works as Early or Late Shakespeare's; however, the "Unsupervised Clustering Experiment" clearly separated Marlowe's works from Shakespeare's.

It is noted that "Marlowe is hard to pin down because of the small corpus that exists for him". Would internal variability in the Marlowe works (written prior to all the Shakespeare works) versus more internal consistency in the Shakespeare works have produced this outcome? A skeptic would like to see more data from the "Unsupervised Clustering Experiment" and a better explanation for how it discriminates among authors.

Answer 2

I disagree that The Skeptical Inquirer is not a credible source. It is a peer-reviewed publication and I have found that it is quite rigorous in its examinations of issues. The article cited here concludes:

To sum up, there really was a Shakespeare, and to believe that someone else wrote the plays and poems bearing his name—that there was in fact a conspiracy to perpetrate an elaborate hoax—is to gratuitously violate the principle of Occam’s razor, the dictum that the hypothesis with the fewest assumptions is to be preferred.

The article is worth reading, and, I believe, represents the issue, and its answer, quite credibly.

Answer 3

The problem with stylometric analysis is that you can make it show just about whatever you want to make it show if you pick and choose the works to be compared.

In comparing Marlowe and Shakespeare, there's no point in comparing "Edward II" to "The Tempest." They were written two decades apart. One would want to compare "Edward II" to "Richard III." And the bulk of the stylometric evidence shows that it is virtually impossible to distinguish early Shakesperean plays from Marlowe.

Furthermore, when any such comparison is done, "The Massacre at Paris" should be excluded from Marlowe's works, as it is known to exist in two versions, neither of which is thought to reflect the original version of the play. Many of the extant plays of Shakespeare also exist only in so-called "bad quartos."

The stylometric evidence doesn't prove that Marlowe wrote Shakespeare, but it doesn't exclude the possibility.

Answer 4

From @MarloweFan's answer, I find that the most recent attempt to separate Marlowe and Shakespeare stylometrically is presented here:

• Neal Fox, Omran Ehmoda and Eugene Charniak, Statistical Stylometrics and the Marlowe-Shakespeare Authorship Debate

The authors failed to separate Marlowe from Early Shakespeare, and yet wrote a conclusion that claimed that they did so. This is dishonest, as I see it.

Known Marlowe Detected As Shakespeare

Their computer program, designed to separate these authors, found the following probable authors for these plays:

• Dido: Early Shakespeare (both methods)
• Faustus: Early Shakespeare (both methods)
• Edward II: Early Shakespeare (both methods)
• Jew of Malta: Late Shakespeare/Early Shakespeare (depending on which method they used)

So that the only works of Marlowe which were not confused as Shakespeare were Tambourlain I and II, and the Massacre at Paris.

The methodology was "leave one and compare to the rest", so by separating the Tambourlaines, they leave out Tambourlaine I and compare to a corpus that includes Tambourlaine II. This is an unsafe comparison--- it is manifest that the closest match to Tambourlaine I is Tambourlaine II. If they had combined the two Tambourlaines into one text, there is no doubt in my mind it would have been misclassified as Early Shakespeare as well. So among the Marlowe works, only The Massacre at Paris is not misidentified!

This means their program failed to separate nearly all of what Marlowe work from Shakespeare, consistently across two completely different methods, and despite their obvious bias against the Marlovian hypothesis.

One of their methods is based on vocabulary, how similar the word choice is in the works. It is notable that the vocabulary of Jew of Malta matches Late Shakespeare, because nobody else matches Late Shakespeare in any way. The other method is based on function words, which are indicative of sentence structure. Both methods agreed in their misclassification of the majority of Marlowe works as belonging to Shakespeare (the two exceptions being Tambourlaine and Massacre). This is a ridiculous failure in a paper that claims to have something negative to say about Marlovian theory, considering the low failure rate with other authors (there were only a couple of other plays by other authors misclassified in this way - please read the paper).

Known Shakespeare Detected As Marlowe

Further, one of Shakespeare's known works, I Henry VI, was misclassified as Marlow in one of the methods, and classified as Shakespeare in the other. This is also notable, because the statistical biases they list explain why it is very unlikely for a work not by Marlowe to be classified as Marlowe by vocabulary (there are fewer words in Marlowe, so you find fewer matches to rare words). The biases in their method, that they explain, make it extremely significant when even a single work is misclassified as Marlowe - it means that the vocabulary in I Henry VI is essentially a dead match to that of the few Marlowe works that exist.

Other Authors: Disputed Works not ascribed to them

There were two plays not by Shakespeare/Marlowe that were not correctly identified as by their canonical authors:

• The Family of Love (Middleton), program attributed to (Johnson/Shakespeare) by (vocabulary/function-words)
• The Case is Altered (Jonson), program attributed to (Shakespeare/Chapman) by (vocabulary/function-words).

For "The Case is Altered", Jonson did not include it in his Folio, and it is of dubious authorship. To my mind, their program conclusively establishes it was not authored by Jonson, not exclusively, and not predominantly. Since their program finds a best-match within the list, one cannot say that Shakespeare or Chapman contributed to "The Case is Altered", only that it is not by Jonson.

"The Family of Love" also is of disputed authorship, it is only attributed to Middleton by academic convention. Their program's failure should conclusively demonstrate that the canonical attribution is wrong. Their program therefore did not have a single failure other than in the case of Marlowe/Shakespeare, where the failure would have (if they hadn't stupidly separated the Tambourlaines) have misidentified all but one of Marlowe's work as Shakespeare by two different methods, and misidentified I Henry VI as by Marlowe.

Estimating chance of authorship

Given that their program is dead on accurate for every single play they examined, only failing to discriminate between Shakespeare and Marlowe, one can estimate the probability that they are different authors. Vocabulary and function words are independent, so each mismatch, assuming separate authorship is at most 10% probable, otherwise the perfect match in the other play identifications is not reasonable. There are 5 mismatched plays, so a probability of $10^{-5}$ of different authorship under these generous assumptions (it is probably closer to 1 in $10^10$), so the chance is more like 1 in 100,000.

Co-authorship?

The only other hypothesis that is reasonable is that Shakespeare modified existing unfinished texts of unpublished Marlowe's plays into his early works. This is unlikely, considering that there is no sharp break in style between Shakespeare and Shakespeare. If one attributes the Shakespeare canon to one author, which I am sure is correct, one must attribute it to Marlowe, with at least 4 sigma confidence level, probably much more.

Their Dubious Conclusion

The content of their computer experiments are an unacceptable counterpoint to their conclusion. Their methods classified nearly all of Marlowe dramatical works, other but the Tambourlaines, as Shakespeare, consistently in both methods, and over two different trials.

From this failure, I feel confident to conclude, unlike these authors, that there is no stylometric difference between Shakespeare and Marlowe, and it is nearly certain in the scientific sense that Marlowe and Shakespeare are the same person.

I am not sure this exhausts the question, considering that these authors are not so reliable, having published a paper with contents and conclusion diametrically opposed. A further more neutral study, even just a replication of their methods with a more quantitative Bayesian estimate of identity, would be useful.