Bio Professor, Linguistics, University of Washington // Faculty Director, Professional MS Program in Computational Linguistics (CLMS) // she/her
Tweets 18,3K
Followers 18,7K
Following 1,7K
Account created 29-07-2010 02:38:06
ID 172179385
TweetDeck : (((ل()(ل() yoav)))) This is a deliberately obtuse misreading of our paper. We aren't talking about using LMs to study the terabytes of data produced by Hemingway nor about LMs used to study reddit. We ground our discussion in the particular trend of large "general purpose" LMs overviewed in Sec 2.
TweetDeck : Edward Grefenstette 🇪🇺 Oh hell no. Research is a conversation, with papers as the speech acts. If they are constantly changing, then the conversation gets unmanageable. On top of that: researchers need to be able to *wrap up* specific papers and move on to the next.
TweetDeck : Joanna J Bryson DuckDuckGo The link in my original tweet is a bit.ly link, which points to a link on my publications page.
TweetDeck : Debjit Paul ACL Anthology For a specific conference, I would guess that the conference schedule on the web page would be one way to go.
Android : Suresh Venkatasubramanian Timnit Gebru Angelina (Angie) McMillan-Major Thank you for this.
TweetDeck : Emiel van Miltenburg There is some linguistic study!
twitter.com/tallinzen/stat…
But that's not what they're created for, to be sure.
TweetDeck : (((ل()(ل() yoav)))) And yes, his online behavior is very relevant --- it was precisely around promoting this dreck that he was harassing us.
TweetDeck : (((ل()(ل() yoav)))) Pseudo-academic, more like it & incredibly problematic. To the point that if there's something in there that you want to cite, and you don't want to be associated with the rest of it, you get very specific.
TweetDeck : (((ل()(ل() yoav)))) I am not taking your suggestions for editing our paper at this point. It is complete. We said what we said.
Go ahead and publish your own response to it.
TweetDeck : (((ل()(ل() yoav)))) I'd say we do. If you want to make an argument to this point that you find more precise, I think you'll find more success if you can do it without calling our work "political" & citing trolls like Lissack.
TweetDeck : Emiel van Miltenburg He seems to have backed off to "the data as it is", which isn't even "the language that was used" but "the language that was collected".
I totally agree that there are uses of LM technology in studying what people say, but that's diff to "general purpose" LMs.
TweetDeck : (((ل()(ل() yoav)))) As I said in my thread above (not to mention at length in the paper): size and quality are not independent.
TweetDeck : (((ل()(ل() yoav)))) Also: you're the one who chose to cite Lissack in your piece, despite the fact that he has been harassing Timnit and I on Twitter and that his premise seems to be that it is unscientific to leave it as unspoken that white supremacy is bad. So yeah, I'm not gonna let that slide.
TweetDeck : (((ل()(ل() yoav)))) So public attention on the fact that lots of tech that is being deployed in the world & affecting peoples lives is based on "all the data we can scrape" collection methodology ... is a bad thing?
TweetDeck : (((ل()(ل() yoav)))) But "the data as it is" is completely incoherent. Data sets only exist when they have been collected, representing a whole series of decisions in terms of what to collect & how to filter, etc. There's no there there to represent!
Twitter Web App : The claim that this kind of scholarship is "political" and "non-scientific" is precisely the kind of gate-keeping move set up to maintain "science" as the domain of people of privilege only. /fin
Twitter Web App : We draw on scholarship from a range of fields that looks at understanding how systems of power and oppression work in society. >>
Twitter Web App : And lastly, miss me with the claim that our work is "political" and therefore has a responsibility to "present the alternative views".
See also:
twitter.com/emilymbender/s…
>>
Twitter Web App : Furthermore, the "debate" you would like us to acknowledge is based on a false premise. As we lay out in detail in Sec 4, the training data emphatically do NOT represent "the world as it is". >>