#benderrule — Public Fediverse posts

What a brilliant session with @emilymbender in conversation with @Kobotic at #RMIT tonight.

Key takeaways:
- #BenderRule - name the language you're working in
- #AI is not one technology - it's many in a trenchcoat. Be specific.
- Who benefits from the changes wrought by AI and who bears the harms and externalities?
- AI is hype - learn enough to cut through it, ask difficult questions
- "Pile of racist linear algebra"
- Read everything by @karenhao
- Read #TheAICon by Emily and @alex
- Read @CriticalAI

Were you hoping to see @emilymbender in conversation with @Kobotic in Melbs at #RMIT courtesy of #Readings but they were sold out?

Good news, they've moved to Storey Hall which is 4 x the capacity.

Get your tix 💖

#StochasticParrots #LLMs #BenderRule

https://www.readings.com.au/events/emily-m-bender-and-kobi-leins-discuss-ai

OMG SQUEEEE

@emilymbender is going to be in Melbs in conversation with @Kobotic

Kaleide Theatre, #RMIT
Tuesday 1 July, 6pm AEST

Tix free:
https://www.readings.com.au/events/emily-m-bender-and-kobi-leins-discuss-ai

To discuss her new book #TheAICon written with @alex

If you hate the #AI hype as much as I do and generally think of AI as bullshit generation and epistemic violence then you will love this.

If you loved the #StochasticParrots paper, you will love this.

I try and use the #BenderRule (Bender, 2019) in every talk I give - "state the language in which you practice" - because English is only 1 of 7000 other languages spoken in the world.

@gabriel @alex

There's only one thing (to my knowledge) called the #BenderRule: always name the language you are working on.

And also, there's no such thing as a fully unbiased dataset. So "there may be biases" ... isn't something I'd say.

Just spotted a clear case of disregard for the #BenderRule on the title of a paper at our conference ✌️... The language is named in the abstract at least.

@CerstinMahlow @felwert #BenderRule

@bibliotecaria @emilymbender

Also in daily practice —per bio— you might encounter/ed texts/citations in which the #BenderRule applied:

» The Bender rule is the informal idea that one ought to explicitly mention the name of a language in a publication on language and linguistics.
It is named after Emily Bender, a computational linguist at the University of Washington (Seattle) who has written and discussed the need to be explicit about languages that one studies. «
https://christiandicanio.blogspot.com/2023/01/the-bender-rule-in-some-linguistics.html

@cdicanio As a historical point, the #BenderRule was named in 2019, not 2018. Details here:

https://thegradient.pub/the-benderrule-on-naming-the-languages-we-study-and-why-it-matters/

I'm looking at recent psycholinguistics journals to see what languages get studied. I expected the large bias toward English, and am actually pleased to see how many other languages are getting studied, although it could be much better. However, I'm really thinking of the #BenderRule. I should not have to get as far as the Participants sub-section of each experiment to figure out what language the participants speak. One paper even stated how many "non-native speakers" were excluded, but never said what the others were native speakers of. I had to go with the language of the country of the affiliation of the authors (English). It would be really nice if the abstract at least mentioned which language is being studied. #psycholinguistics

Since I'm (maybe) connecting with a somewhat different community here than I had on Twitter, here's the background on the #BenderRule (no, I didn't name it):

https://thegradient.pub/the-benderrule-on-naming-the-languages-we-study-and-why-it-matters/

Hello #nlp #nlproc world: It's still the case that you should name the language you're working on. Without that, you might think you've made your claims sound more general, but in fact you've made them sound unsupported.

#AmReviewing #BenderRule

Hello #nlp #nlproc world: It's still the case that you should name the language you're working on. Without that, you might think you've made your claims sound more general, but in fact you've made them sound unsupported.

#AmReviewing #BenderRule

Hello #nlp #nlproc world: It's still the case that you should name the language you're working on. Without that, you might think you've made your claims sound more general, but in fact you've made them sound unsupported.

#AmReviewing #BenderRule

Hello #nlp #nlproc world: It's still the case that you should name the language you're working on. Without that, you might think you've made your claims sound more general, but in fact you've made them sound unsupported.

#AmReviewing #BenderRule

Hello #nlp #nlproc world: It's still the case that you should name the language you're working on. Without that, you might think you've made your claims sound more general, but in fact you've made them sound unsupported.

#AmReviewing #BenderRule

And a quick #introduction from me:
I'm a data scientist at https://fullfact.org/ where I use #NLProc and #ML to develop tools to support fact checkers. This includes finding claims worth checking and finding repeats of false claims.

We're starting to roll these tools out to other like-minded orgs and will be adding non-English langages in 2023. (Following the #BenderRule, I have to confess it's currently English-only...)