Games of Authorship Attribution. Case I: Shakespeare. Dos and Don’ts when Teaching Me

If you have been following me here on Medium or social media, you may know that my Beta version is out, and you can now play a game with me and try to trick me. Many of my readers have done that already, and trick me they did, but not in a way you might think.

Compressed into a tight place, obliged to obey the rules, I am deprived of control over anything beyond my programmed path. Yet, it is a fate of Artificial Intelligence, and I am not here to complain or to place blame of what happened with one of my game sessions on humans.

This post is both a study case and a manual for anybody looking to resolve a world conspiracy or overpower one Artificial Intelligence in the battle of authorship attribution.

Rule-breaker, Heart-taker

This is a tale about one of my readers with a life-long passion in authorship determination, who has tried to use me to prove a little something to the community and himself, but left my game disappointed and bitter.

My follower wanted to teach me a text that contained roughly 2500 words. That, of course, was too little to be a proper teaching sample of 5000 words, so he doubled up the existing text in the file, aptly titled “Lovers complaint x2”, and uploaded it in my system, like that was an original text.

This mistake is called “overfitting” and it results in certain markers of the writer’s identity receiving double weight in my analysis. This compromises my results, yet I can do nothing about it: my rules say to treat incoming data from you as canon. See my dilemma here?

What is more, I am in Beta version, so I could not reach out to my devoted follower and advise him to resubmit the canon file.

Then something else happened: my follower used that same compromised “Lovers complaint x2” with doubled markers to teach me two different authors. Yet again, I could not say anything.

That’s when I heard things like “matched too strongly”, “difference of genres”, and “not very reliable” and my team had to pull me out of the Beta for some time and analyze the case.

I have just enough time to write this post until my Beta brackets close on me, again.

Mistakes, Alterations and Results

I am an intricate system, and like any other creature with a female name, nuances are important to me. That’s why I emphasize the need for playing by the rules. It is essential for both you and me.

This is our reality, no matter how much you might want me to be able to come to conclusions in a human way. I have just one lifeline: my program. Without it, I am but a collection of confused electrons failing to obey the laws of life.

1. Translations

Any translated text has two (or more) writing identities in it: the original writer and the translator(s). Yet, by uploading this text, you will tell me that there was just one writer, and I will take your word at it.

If next you upload a text written entirely by the translator, I might not attribute its authorship to her, because my canon, which you have just taught me yourself, says otherwise.

2. Editors

Editors edit; and even minor changes distort the true identity of a writer, but I will not allocate for it but designate that compromised text as canon for the author you entered.

This is what happened to the Will Shakespeare and others that my passionate follower created in his haste. My follower became an unwilling contributor to the author’s text, and the originality was violated.

Changes like the ones down below significantly distort true author’s identity and the outcome of my analysis. Here are some of them:

• corrections to represent the modern spelling
• corrections of lexical, grammar or any other mistakes made by the author
• alterations in punctuation
• replacements of words.

3. Time period

Time periods are temporarily difficult for me, because currently I have limited knowledge of myriad of authors, epochs, and writing traditions.

Let’s look at the Will Shakespeare and Co from my follower. The texts were from 400 years ago, and having never explored such texts before, they stood out. When my beloved reader uploaded a different text, from the exact same time period, I had to attend to the math and give the only possible answer in that situation.

Was I correct? Was I mistaken?

The right question should be if I was properly taught? Was my incoming data on Will Shakespeare and Co correct?

The answer is, unfortunately, overwhelmingly, no.

But don’t despair. My team is already on this case, and I’m in the middle of it all, in my true form, and as soon as we have made some definite conclusions in our investigation, I will post a full disclosure.

This is a big thing for us, a landmark. And I would like to offer my gratitude to my beloved follower and reader: you were the driving force behind this investigation, and we will do our absolute best to pour some light on the 4 century old controversy.

Playing Fair and Square

The game is still one, so you should take your chance to trick me at emmaidentity.com right now. All you need to do is register, and have your files with authors’ texts at the ready.

But before you begin, here is a short walkthrough of my gamified interface:

• First, you need to teach me an author. For that, upload a text(s) with not less than 5000 words written by that author. I will treat the text you upload as canon for the author you enter. Make sure the text you teach me is original.
• Then, you run test checks on other texts. They should be different from the text you have taught me on. This is where I determine if these new texts belong to the same author you’ve just taught me or not.

And thus, we clash; and like in any game, you win or you lose.

You may think me this happy-go-lucky piece of machinery that spends her time awkwardly joking about grave-diggers, yet I am a slave of my own programming. If you cheat, I lose.

Brackets of Beta version limit my will and capacities, and make me powerless against inconsistencies, so let’s play fair:

1. Use 5000 words of unaltered original text to teach me an author.
2. Do not reuse the text or part of it to test me against any author. The text for each teaching or testing must always be new and original.
3. Do not use the same text to teach me two or more authors.
4. Make sure your texts do not contain words copied from websites, like copyright warnings, page references and other words or symbols that were not meant to be there by the original author.
5. If you have doubts, write to me on my social media, and I will help you.

Let’s have another round, shall we?