magistrate: The arc of the Earth in dark space. (Default)
Because definitely what I need is more webapp ideas that I don't have time to develop.



Anyway, following on from a Twitter conversation, I'm wondering how it would work to make a writing program which could track the genders of a number of characters and then arbitrarily shuffle them. What I'm picturing is, simplified, something like this:

• At the top of the document are a number of fields which ask for a character name (or a list of character references, such as name and nickname and other variations) and pairs the name with a gender (and its associated set of pronouns).

• Each character you add is arbitrarily assigned a color (or icon or other distinguishing visual marker).

• As you type, a parser will keep track of which name (or referent) has been typed last for each of the original genders. When you type a pronoun, it will look at the last character reference matching that pronoun's set, and highlight the pronoun (or assign it the correct icon) to associate it with the specific character. It'll also have some kind of (mouseover?) menu to allow users to correct its assumption about which character it refers to.

• When you finish writing, each pronoun will be associated with a character. So you can hit a shuffle button, and then the characters' genders will be shuffled, and each pronoun can be brought back into compliance with the character's gender.

Needless to say, this would fail in a lot of situations. Take, for example:

• Dialogue. "He's not coming today," he said. (I mean, I guess I could set up a sub-parser which kept track of the last character reference inside a set of quotes?)

• Ambiguiety. We'll just call this the Randall Munroe exploit. I guess people would just have to make close, personal friends with the drop-down menus?

• Gay porn. I am reliably informed by people who have tried to write gay porn that pronouns are a nightmare anyway. And humans are better at parsing language than computers are.

• Unexpected cases. Language is complicated, yo!

I feel like there should be a way to handle this, and that it probably involves algorithms. I'm a bit worried that trying to write a general-purpose pronoun shuffler would actually require re-inventing Google Translate. Any computational linguists out there who want to point out things I'm missing?

Date: 2015-01-18 06:54 pm (UTC)From: [personal profile] zeeth_kyrah
zeeth_kyrah: A glowing white and blue anthropomorphic horse stands before a pink and blue sky. (Default)
It took me a few moments, but I realized that the rules you placed at the top don't say whether the original writing should have replacement genders, standard genders, ambiguity, or what. Perhaps most would write with standard genders, but I would have used single-gender replacements or ambiguous references and then tried to shoehorn gender into the piece in the second stage.

Some people use "thon" as a gender-neutral or neuter pronoun; I don't usually, as it doesn't flow for me. I prefer "su" or "na" as non-gendered or ambiguous pronouns for English usage, but that's a personal preference, and doesn't seem to hold in all cases for me. Using non-English pronouns doesn't otherwise seem to work for me unless it's a language I already use a lot.

Some people just get stuck on certain thoughts of "how things are", sort of like how some people are so fixed in their handedness that trying to use the other hand ends in confusion to the point of frustration.

The algorithmic form of what you're saying you want fails in ambiguous or same-gender situations quite a lot. The computer has to understand enough context to do pronoun association as well as a human reader in these cases, and that means understanding who is speaking, who is interacting, who is addressed, and who is the subject. Humans are a lot more relationship-focused than machines, and I think it's an NP-open problem to follow ambiguous referents in a flattened context like text.

prior art, kind of

Date: 2015-01-18 07:38 pm (UTC)From: [personal profile] brainwane
brainwane: My smiling face in front of a brick wall, May 2015. (Default)
I feel as though there ought to be a way to reuse bits of Jailbreak The Patriarchy to get 60% of the way there.....

Date: 2015-01-21 06:39 pm (UTC)From: (Anonymous)
As an author, I absolutely totally love this idea. And I am a computational linguist (who coincidentally just happened across your site via the QDSF email that just went out). I haven't looked at coreference resolution for a couple of years, but basically, the pronoun labelling stage works *reasonably* for English and it's probably "good enough" if you're happy that people can correct it's labelling as they write. As you identify, it's going to need extra training for speech. That doesn't begin to cover the gendered nouns (she's X's daughter or an actress; he's Z's husband or headmaster of the school... etc), but again, it's a closed-class problem so probably tractable. If you want to open source this I'd contribute... drop me an email if you want to discuss it more?

Rachel
http://rachelcotterill.com (having problems with openid for every site I own!)

Profile

magistrate: The arc of the Earth in dark space. (Default)
magistrate

January 2017

S M T W T F S
1234567
891011121314
15161718192021
22232425262728
29 3031    

Style Credit

Expand Cut Tags

No cut tags
Page generated Jul. 21st, 2017 06:40 pm
Powered by Dreamwidth Studios