Interested in the LLM vs. linguistic theory debate? My just out book will tell you (almost) all about it. mitpress.mit.edu/9780262052009/…
English
yosef grodzinsky
16 posts
yosef grodzinsky
@YGrodzinsky
neurolinguist @HebrewU & @FZ_Juelich
Israel Katılım Ekim 2018
158 Takip Edilen215 Takipçiler
@lenelid I am sad to see that you’re not identifying me…pls retry
English
@lenelid Well I was surprised to see that so many years later you’re still very much engaged in ME affairs
English
Elizabeth Tsurkovs historie er lite omtalt i Norge. Anbefaler hennes avslørende historie om torturen hun opplevde i Irak.
Elizabeth Tsurkov@LizHurra
Throughout my 903 days in captivity, I reflected on this odd juxtaposition: my torturers were incredibly ignorant, yet highly skilled at inflicting harm on others. theatlantic.com/international/… I wrote the first draft of this article while still in captivity. Now it's in @TheAtlantic
Norsk
@dudiamsalem בושה! לך תעבוד בדואר חבילות. אפס אפסים שמביא רק חורבן
עברית
ב-48 השעות האחרונות פעלתי באופן אינטנסיבי, יחד עם רה"מ נתניהו, להקמתה של ממשלת החירום הלאומית.
זהו צו השעה!
אני שמח ומברך על הקמתה, ובטוח שיחד נצא מחוזקים למשימה העליונה שלנו-
ניצחון מוחץ על כל ארגוני הרשע והטרור! 💪
רק ביחד ננצח-
עם ישראל חי ❗💙🇮🇱
עברית
@rothmar תתבייש לך! במקום לקרוע קריעה גדולה ולהכות על חטא אתה ממשיך להתחכם ולשקר. אין לך תקומה. ישר לפח הזבל של ההיסטוריה המשיחית יחד עם שבתי צבי ודומי
עברית
עכשיו כולנו ביחד!
כל שיחה שלי עם אנשי הפרקליטות, משרד המשפטים, הייעוץ המשפטי, החברים המשפטנים, כולל גם כאלה שהפגינו נגדי עד לפני שבוע.
הכי ענייני, הכי מקצועי, הכי מהר שאפשר.
כולנו מגוייסים למען מטרה אחת.
נשבענו כולנו, עם כל העולם, לאחר השואה:
"לעולם לא עוד"
רק ביחד ננצח!
עברית
@_jasonwei Don’t recall such hyperbole: you can cherry-pick a success among failures, but not vice versa. A failure case is a failure case. No cherry-picking.
English
@emilymbender An important part of the current hype is due to the use of the wrong benchmarks. Crucially, hardly any is linguistically motivated.
English
1. The general operating mode is "make shit up". Sometimes it just happens to be right.
2. "Make shit up" is actually giving too much credit, since the LMs are only coming up with sequences of linguistic form & it only means something when a person makes sense of it.
English
@neuranna @wmatchin @ev_fedorenko Unfortunately, neither work uses stims that test syntax or semantics by the standards of any reasonable linguistic approach
English
@wmatchin @ev_fedorenko has shown beautifully that syntax and semantics do not dissociate *within the language network*. Whether structure and content are processed separately in other networks (including these semantic demand regions) remains to be determined
English
Would it be fair to say this represents a syntax-semantics distinction in brain networks?
Anna Ivanova@neuranna
A set of brain regions - distinct from the language network - responds to semantic tasks for both sentences and pictures of events! Presenting this work tonight at #SNL2022
English
@sleepinyourhat The successes of NLP are very partial ENGINEERING successes. Benchmarks are mostly off the mark linguistically, bearing no similarity to human behavior on any serious test.
English
If progress extends all the way to near-human performance on language and reasoning tasks, the consequences are likely to be transformative. Quite possibly the most impactful technology humanity will ever build.
English
I’m starting an AI safety research group at NYU. Why? (🧵)
English
@haspelmath @davidadger @EvelinaLeivada @SashaWilmoth You don’t need me to tell you that in science you make bets. Generative or not, there is no alternative to theorizing. Track record may be ambiguous, but in the absence of a theory there’s no track record at all.
English
@YGrodzinsky @davidadger @EvelinaLeivada @SashaWilmoth No, I was not talking about „linguistic theory“ in general, but about the many different and mutually incompatible generative theories. How do we know whether the hard work will pay off? The track record looks ambiguous.
English
Little note from my thesis about word order in noun phrases (Pitjantjatjara has N-Dem-Adj-Num which is rare). It strikes me as very strange scientific practice, that if your theory doesn't account for the data, the best explanation is that the data is wrong.
English
@haspelmath @davidadger @EvelinaLeivada @SashaWilmoth Unglaublich: a serious linguist complaining that linguistic theory is too hard to understand? Work harder, my friend, and you’ll get it. Would a physicist ever think about saying that about her field?
English
@davidadger @EvelinaLeivada @SashaWilmoth Yes, this is my view of generative grammar, too - but the main problem I have is that the theories are too hard to understand. And very often they seem to be both too restrictive and too permissive. The latter is certainly a bigger problem.
English
@lfernandino @SfNJournals Nice. Though the “language areas” are THE LANGUAGE AREAS, once language is viewed as a rule system, not a collection of words. Check out our Cortex 2021 paper for a recent review and localization with cytoarchitectonic maps.
English
Happy to announce our new #JNeurosci @SfNJournals paper! We used fMRI RSA searchlight mapping to determine which cortical areas represent information about the meanings of individual words, and the results were beautiful. 1/9 jneurosci.org/content/42/37/…
English
@alex_ander Trying to imagine how your “semantic reconstruction” would work once we systematically sprinkle your text with negations and other logical operators. My guess: it would collapse very quickly, as such elements seem to be ignored by the decoder.
English
@LanguageMIT Hmmm. Didn’t know it gets this low here. Straight up Chomsky-bashing, no evidence no nothing.
English
In the 1990s, these were Noam Chomsky's go-to responses for questions he didn't like
Christopher Hammerly@chrishammerly
I'll join others complaining about LSA reviews. R1 said I "need to take a syntax class" and that I "clearly had never taken introduction to linguistics". Unbelievably unprofessional, and sadly it wasn't even paired with a coherent critique. Other reviewers were positive at least.
English
@pictureofitself @EvelinaLeivada Some comparative insights wouldn’t hurt before such sweeping conclusions about refereeing are reached: Neither cogneuro nor cogsci fare any better. Problem goes way beyond linguistics.
English
hot take: this isn't an LSA problem, the LSA just provides one of the purest distillations of it. peer review in linguistics is irrevocably broken (and I say this as someone who fared okay-to-moderately-well in my time). review of full-length papers is just as broken as this.
English