🍾New preprint just out, following a 3-year-long brewing! Almost all theories of bilingual cognition recognize the role of language distance in modulating cognitive control or in affecting language processing. Yet, nobody really knows how to measure distance in global way! 1/7
@PlakiasNikos
@mkaristianou
Eσείς κε. Πλακιά πως μπορείτε; Διαφωνείτε με τις δράσεις του συλλόγου; Δεκτό. Δεν επροσωπείστε εκεί και δεν σας εκπροσωπεί. Εσείς όμως θέλετε να έχετε δικαίωμα λόγου σε έναν σύλλογο που δεν σας εκπροσωπεί. Ε, δεν γίνεται. Λέγεται δημοκρατία.
Is it possible that I'm seriously asked to choose between caring for my preterm baby and keeping my grant? Really, is it? Here's a story of what
#WomenInScience
often have to endure in silence. Not this time. Please share. Please help dispel the silence.
They have been fed mountains of (copyrighted) data, sure, but still LLMs can't reliably track entities in simple discourse. A super simple passive sentence is enough. That much for "knowing more"!
@GaryMarcus
He didn't say it was better at scientific reasoning; he said it knew more than you, which it objectively does.
Call me when you've read everything crawled by CC, 20 times.
🔔Can
#ArtificialInteligence
apps understand the world behind words? New experiment with
@GaryMarcus
,
@ElliotMurphy91
& Fritz Günther. We tested humans & LLMs in a new domain: the ability to read leet code. Results: HUM4NS C4N UNDERST4ND THIS BUT LLMs 4PP4RENTLY C4N'T! Why? 1/2
🔔Call for 3 postdoc positions associated with my project TURING (The Language Understanding of Artificial Intelligence Applications) is now open! More info Apply here . Deadline 31 July. Location: Beautiful Barcelona, but flexible😎
Big, wonderful news! Super happy to announce that I got my DREAM job! In 2 weeks I join the
@icreacommunity
as a full Research Professor at
@CLT_uab
! To those who helped/taught/worked with me over the years, THANK YOU!🙏And now your girl is off for bubbles! Cheers!🥳
#Lingtwitter
Our #ΟΑ article is out
@CambridgeCore
!
#Bilingualism
with minority languages. We discuss monolingual norms, native-speakerism, the quest for testing specific profiles & the use of powerful narratives that evoke problematic labels/ableist terminology.
Today we heard the baby's heartbeat for the first time and the tears keep falling! It's my 3rd pregnancy, but I have only one child. I cross fingers that we make it this time!🤞🥰
Since
#linguistics
twitter is so hyped about grammaticality judgments, some shameless self-advertizing is due. It's not grammaticality judgments, it's acceptability judgments, and these two scales don't always coincide.
SUPER excited about our new preprint!
@vittoria_vd
,
@GaryMarcus
,
@ElliotMurphy91
& I examine what happens with language when Large Language Models are pushed beyond their training data 💣. Short thread👇1/8
Cuando publiqué mi historia, no sabía que quería lograr. Supongo que quería ser escuchada y desahogarme. Pero mira, hemos conseguido un cambio importante! Y hoy canto victoria! Gracias
@DianaMorantR
@ZulemaAltam
#womeninSTEM
It´s not an amazing time to work in the cogn science of language. You read that your work "HAS TOTALLY FALLEN TO PIECES". Does it matter that this is not what
@_jennhu
et al find re our PNAS paper? No. And labeling people as generative or non-generative. Truly necessary? 1/2
It is an amazing time to work in the cognitive science of language. Here are a few remarkable recent results, many of which highlight ways in which the critiques of LLMs (especially from generative linguistics!) have totally fallen to pieces.
My latest article appeared
@inferencereview
. It's an article I've been told to not publish until I get tenure. But I did bc proving ourselves wrong as quickly as possible is the only way to progress. (I didn't say that. Richard Feynman did.) Happy reading!
I´ve been quite silent on twitter recently. For a reason. I just updated my CV, adding the section Career Breaks to explain my productivity drop(s) in 20-21. Only a few lines, but they took me days to write. 1/4 👇
📢Article out in
@SpringerNature
Humanities & Soc Sci Communications!🥳 I identify & explain the cognitive biases behind the most viral
#linguistic
universal: the order of adjectives. Key notions: Zipf's law, Ambiguity Intolerance, Novel Information Bias.
As editor I always feel uncomfortable when a reviewer says that it shows that Eng is not the author's native language. Especially when the article has been proofread, but the reviewer thinks that the writing doesn't flow. What a privilege it must be to write without this worry.
People, stop tweeting that Noam Chomsky has passed away unless you have credible internal information. Bill said that Mary said that a colleague in MA said is not good. No news site confirms the news. Let's be respectful.
@GaryMarcus
@ElliotMurphy91
Short explanation: LLMs lack grounded cognition; fixed associations between words & word vectors don't entail grasping the concepts behind words. Long explanation: Preprint updated with an experimental component that tests LLMs in a novel way! 2/2
Often
#WomenInScience
don't talk about these experiences. Sometimes we´re advised to not talk about them. Today, as I look into the eyes of my precious daughter Alba, I feel her strength, and I'm breaking my silence. 4/4
Today is one of the days that I lose faith in
#science
and
#academia
. A thread. A promising new journal on
#language
development rejected our work. On grounds of scope. After months of review. After asking for and getting major revisions. After both the 1st action editor and 1/5
@Soupiera
Τι λέτε τώρα. Σε πιλοτικά προγραμμάτα του εξωτερικού, στις αίθουσες τοκετών υπάρχουν ως και ναργιλέδες. Το να παίζει και το ε ντε λα μαγκέτ ντε Βοτανίκ είναι ακόμα προαιρετικό, αλλά φαίνεται να επιδρά επίσης θετικά στην υγεία των νεογνών.
Does
#bilingualism
confer robust positive & negative effects? Studies & meta-analyses are mixed. Why? Our answer was just published, and in it, we tackle the critical question of why the results seem to tell a different story depending on who's narrating.👇
Hey
#linguistics
twitter. I´ll probably repent this thread, but here it goes.
@spiantado
and Yuan Yang wrote an excellent paper. I agree with its results and I partly agree with its conclusions. What worries me, however, is the framing. In the 1st tweet of the thread below 1/5
I am sooooooo excited for this paper. We've spent years developing a super fast program induction library. We use it to learn key pieces of language structure.
So much of what Chomskyan linguists say about learnability is totally wrong.
🧵
🥳Fantaaastic news!👇The
#Consolidaci
ónInvestigadora results are out and
@AgEInves
funded my project TURING (The Language Understanding of Artificial Intelligence Applications) with ca. 340K!! Elated for this new adventure! Stay tuned, I'll be soon hiring 2 or 3 post-docs.
Again we see terms like 'Chomskyan linguistics', 'Chomsky-esque science', 'Chomsky army'. As a female scholar whose work has been *mis*characterized this way, I don't particularly enjoy when my work is linked to someone's else (always a senior male) name.
I spent most of the sleepless nights of my maternity leave preparing the project ATOM (Adaptations to Multilingualism)! Happy to see it was selected for funding! 😍
‼️ La
@AgEInves
publica la propuesta de resolución provisional de la convocatoria Proyectos de Generación de Conocimiento 2021.
3.079 ayudas a proyectos propuestos para financiación.
🔗
GREAT NEWS! As a
#RYC
I couldn't extend my grant to take care of a minor with health conditions. Now this changes, as the new
#LeyDeLaCiencia
makes provisions for cases like mine. An important positive step! Gracias
@ZulemaAltam
@DianaMorantR
🤩👏
@EvelinaLeivada
Dear Evelina, I am glad to inform you that yesterday the new Science Act was approved and cases as yours are explicitly protected:
➡️Art.21/22 include prorrogue/suspension of contract for any caring situation
A step ahead for work-life balance in research, thanks to your case
💜
New paper with
@ElliotMurphy91
and
@GaryMarcus
online! DALLE 2 has been widely celebrated as showing sophisticated
#linguistic
abilities. Yet, an analysis of its outputs reveals that it fails to perform above chance in a range of syntactic conditions.👉
And that's me being mansplained and called "honey". In public. By a fellow academic I don't know. Work hard, he says. Read my CV, you'll know what hard work looks like!
💥New paper out in
@SciReports
with
@vittoria_vd
and
@masullo_camilla
!💥Contrary to widespread belief, we find that bilingual advantages and bilingual disadvantages go hand in hand. And this happens for a reason (i.e. cognitive trade-offs).
To sum up, if I disagree with a claim X in your work, I don't dismiss the value of your work, I don´t question you as a scholar, I don´t reject your theory/framework in its entirety, and I don´t forfeit the right to agree with another claim Z in your work. Paz🤠
⚠️Verdict returned: Not all critiques of LLMs have fallen into pieces (yet?)! Summarizing thread, fresh results, a short response (), a long response () & a plot twist. 👀 1/20
Today it´s two years since my first daughter Agape was born sleeping. I haven´t seen her, the state doesn´t even recognize her existence because she was born at 20 weeks of gestation. She never existed for this world. Except she did. Today we raise our glass to Agape.
Paula Badosa, I hope you're better in tennis than in linguistics, because Catalan is a language. Dunno, I'd feel ashamed to casually reveal such a profound ignorance of the linguistic reality of the place I was born and brought up.
Terrible que una persona que ha viscut la majoria de la seva vida a Barcelona i València digui sense cap mena de pudor que "el català no és una llengua". És un exemple d'ignorància terrible. Vergonya,
@paulabadosa
@_jennhu
@spiantado
@vittoria_vd
as a tenured professor to talk like this publicly about the work of a young researcher that hasn´t even defended her dissertation. I´d like to think I´d be more sensitive, more careful, more gentle. Less labeling and more caramaderie would be great for all of us. 3/3
Our latest article with
@masullo_camilla
&
@vittoria_vd
on the sociolinguistic vs. cognitive origin of bilingual adaptations is just out in Bilingualism: Language & Cognition
@CambridgeCore
! And it's Open Access and Open Materials. 🥳
A
@Nature
editor invited me & 2 scholars to submit a short commentary on LLMs in Nov. We submitted in June. They've been sitting on it for 2 months! Today they rejected it "bc it's no longer Nov 22". Sorry we took time to work on this folks! Also I was on pregnancy leave! My bad.
What about LLMs?🧐We tested their ability to process grammatical illusions, semantic anomalies & complex nested hierarchies. Short answer: Even the much praised next-word prediction abilities of LLMs face issues of robustness, when pushed beyond training data. 6/8
In one of my very first talks, when I was still an undergrad, a senior & famous ling professor fell asleep. Fine by me. He then woke up from the clapping & asked questions that referred to specific points of the talk! That's the closest I've ever been to paranormal activity!👽
@Duncan_Wilson78
This is wonderful. I once chaired a lecture (which is perhaps more embarrassing?) where a senior lady fell asleep, and woke up shouting 'rice pudding' in the middle of the lecture - only for the speaker to say, 'please leave your questions til the end'... Mortifying.
📌What kind of theories does
#Minimalism
involve?
📌Are data from a single person good enough?
📌Does generative
#linguistics
fare better than other branches of psychology by virtue of being less interested in stats?
@KKGrohmann
& I have some answers 👇
*Always*. Because, you know, Chomsky has been a groundbreaking figure in the field of LLM research for decades, but alas his experiments in this field never quite fooled us: deep down, we always knew. This framing does injustice to the v. good paper by Kallini et al. 1/2
You know what's missing from the Piantadosi list of remarkable results? This by
@celestekidd
&
@Abebab
. Perhaps that's a good candidate for critiques of LLMs that haven't fallen into pieces? Extra points if the critique comes from non-generative linguists.
Oh no!! The burden of proof lies on the one making the claim. If you say LLMs understand language, you need to defend this position. It's not on critics to prove otherwise. That's the fallacy of appealing to ignorance.
No, ChatGPT does not have theory of mind. (It doesn't have a mind either, but that's for another day.) I just asked it what it would expect to find in a box that has the label 'paper' but that Ann opened and saw that is full of chocolates, and it said paper.
@sodalitatum
AI is confused because it knows how ασπάζομαι behaves in Ancient Greek? Surprising, given how bad it is in modern Greek. Cypriot and Standard.
My last article, just published OA in Zeitschrift für Sprachwissenschaft, discusses the mid-level generalizations of generative
#linguistics
. What are they? What´s their cognitive basis? How robust are they from an experimental point of view? Some answers👇
If you "ask reviewers to submit their review report within 7 days" in the middle of August, without even a personalized email, no surprise you don´t find reviewers.
The effect of
#bilingualism
on cognition is a heated topic. Some studies find robust effects, others don´t. Why? Jason Rothman
@lab_polar
,
@MaritWestergaa1
,
@jdunabeitia
& I discuss this in an article just published in Bilingualism: Language and Cognition!
@ToLizaki21
@Chandlers_Duck
Σε οργανωμένες κοινωνίες με πρόνοιες για τους νέους γονείς, δεν χρειάζεσαι "βοήθεια από γονείς" (aka πως να φορτώσεις κουρασμένους ανθρώπους με ένα νήπιο που τρέχει ασταμάτητα, σακατεύοντας την μέση τους). Η απουσία κράτους πρόνοιας είναι έκδηλη νομίζω.
I recommend Kallini et al, so I agree on this one. But I wonder: Is it good practice to treat scientific work as a student essay? Criticism? Sure. Write a proper reply, a thread, a blog. This school-type marking that transpires a "I fixed for you" air is not promoting dialogue.
I know I am late in the project cycle for this, but I do have suggested edits for the team behind My overall comment is that the central claims in the original are lacking in empirical support.
Being on high risk pregnancy leave means I must spend a good part of the day lying down. Good books keep good company, and I can't recommend
@JWGrieve
& Helena Woodfield's The Language of Fake News highly enough!
Partly inspired by a Data Colada post, I published a paper that casts some doubt on the results of a very well-known study in
#bilingualism
. Now the journal wants to remove it from the Special Issue it appears in for well no reason. Here's a funny story!
of any. I know people that agree with some of Chomsky´s claims & disagree with others. Using a binary Chomskyan/non Chomskyan label is not useful & not capturing the voices of many of us. Let´s find a way to integrate these voices into a dialogue that does not use binary labels.
Excellent piece! This thinking is pervasive in smaller fields like
#linguistics
. An editor once wrote me "We don't *usually* publish critical replies". Meaning: We won't take this from you, female junior scholar, but if Famous Linguist sent sth similar, we'd happily consider it.
“While critique is certainly an important part of improving scientific fields, we do not want to set a precedent or encourage submission of articles that only critique the methods used by others.”
Attention all units! Before you put Chomsky's name on a paper that claims that Universal Grammar is dead *because of LLMs*, let's ask an LLM what it thinks, especially if you believe it has really cracked the language threshold, hence it can opine on the matter. 😺
I read tweets about
#WomenInScience
and
#GenderEquality
with great interest, as I'm signing the documents to reduce my working hours to 30% *without having the possibility to extend my grant*, effectively being punished for wanting to be both a
#scientist
and a
#mother
. Cheers.
Interested in a PhD in linguistics? Come work with us
@rollingURV
!
@isabel_oltra
& I have 2 Martí i Franquès COFUND positions available, one on language processing & the bilingual mind and one on grammar, grammaticality and acceptability. Pls RT!
Submitted to a journal with big IF. They: Out of scope. Me: Why? They: Silence. Me: Btw I did an extensive check and you´ve published such papers. Here are the links. They: Silence. Me: But a senior male is corresponding author in ALL of them. They: In scope! Pls resubmit asap🧐
🔥New preprint w
@ElliotMurphy91
&
@vittoria_vd
! Are Large Language Models adequate models of language once decomposed into their essential components? How do they deviate from the real-world system they model? What risks and challenges do they run into?
@Trashisinthehe1
@madam_shou
Αυτά τα κάνουμε εμείς οι χρηματοδοτούμενοι από σκοτεινά κέντρα και συμφέροντα. Ο #πετρακος είναι υπεράνω, διότι σπούδασε στο πανεπιστήμιο Fame Story 2. Σε άλλη χώρα θα είχε παρέμβει κάποια αρμόδια αρχή σίγουρα.
My baby died inside me at 20 weeks. I was close to death too: acute septic shock due to chorioamnionitis. Scared to live in a world that doesn't allow abortions, therapeutic or not. My uterus, my life, my choice.
New
#OA
paper in
@PNASNews
with
@vittoria_vd
& Fritz Gunther! Systematic testing of 3
#LLMs
reveals low language accuracy, absence of response stability, and a yes-response bias! Short thread 1/10 👇
@sofoulinii
βρωμάει σεξισμό ο μεσιέ Σπανούλης. Λες και τα παιδιά τα έκανε μόνη της η Ολυμπία, λες και ο πατέρας είναι κάτι σαν background prop, αρκεί να φέρνει λεφτά στο σπίτι. Τι να κάνεις, το να είσαι καλός στο να κυνηγάς το τόπι δεν σε κάνει και σωστό άνθρωπο.
My mom called to say she read the last article I tweeted about. She said 'isn't that great? You were cited many times!' I said 'mama, I wrote it, it's normal to refer to prev work.' Her enthusiasm dropped. 'The others don't cite you?' 'They do.' Well, she'll be watching y'all😂👁️
Very interesting work by
@ev_fedorenko
and colleagues! PS. People who don't clutch tightly to previous findings and are happy to revise their beliefs based on new evidence deserve special applause IMO.
@CarinaKauf
’s new work overturns our earlier claim () that local semantic composition drives the language system. New data show that linguistic inputs must be syntactically parsable to fully engage the lang system (and they need not be semant. plausible).
Contributing to the mentality that treats English monolingualism as the norm means perpetuating a racist ideology. "Small languages" don't need extra justification to be considered valid domains of research. The journal deserves to be named & shamed for letting this comment pass.
I cannot believe that some reviewers consider that the the relevance of our research was of limited interest because “The domain of argument mining is quite limited” and “Catalan is a relatively small language which may decrease the impact of this work”. Surrealistic.
@_jennhu
And I want to bring another dimension on the table. The PNAS paper that
@spiantado
so charitably described as having fallen into pieces, was led by my PhD student
@vittoria_vd
. A female young student in her earliest steps. Perhaps it's a gender thing, but I'd be reluctant 2/3
Just gave positive to COVID19. The hardest part is not the pain, not the fever, not the sore throat. It is being isolated and away from my 1yo immunocompromised baby, hoping and praying she didn´t get it. Stay safe y'all.
@Phmichaelides
@lola_cons
@CommentaryJho
Είναι αναξιοπρεπές να προσκομίζεις βεβαιώσεις όταν εγείρονται υποψίες για συγκεκριμένες παρατυπίες; Να το πω και του πρύτανή μου καλέ, να μην δίνει βεβαιώσεις αβέρτα το πανεπιστήμιο διότι η χρήση τους σημαίνει ότι δεχόμαστε να εξευτελιζόμαστε. Αλάνθαστη λογική.
I never ask for RTs. But this time seeing a. how unfairly a tenured male professor publicly slams the work of a female PhD student & b. how many people liked his tweet, I'll do it to pass a message. Please RT if as a junior scholar you'd find this unduly harsh too
#WomenInScience
@_jennhu
And I want to bring another dimension on the table. The PNAS paper that
@spiantado
so charitably described as having fallen into pieces, was led by my PhD student
@vittoria_vd
. A female young student in her earliest steps. Perhaps it's a gender thing, but I'd be reluctant 2/3
📌My project Adaptations to Multilingualism offers a funded, flexible/hybrid, 6month research assistant position, 50% dedication, running 01/09/24-28/02/25. If you know Spanish & R, apply here! If you don't like hybrid, Barcelona is a treat.😉 Please RT!
Our article with
@MaritWestergaa1
and Natalia Mitrofanova
@AcqVA_Aurora
on lexical vs structural similarity & crosslinguistic influence in L3 acquisition was just published in Linguistic Approaches to Bilingualism!
Does ´More people have been to Russia than I have´ sound meaningful? If yes, try again! That´s a comparative illusion; a sentence that looks ok, but is not. Why do we fall for illusions? How even? My last experiment, hot off the press, has some answers!
If you read one article over Aug holidays, this should be the one: On the Need for Theoretically Guided Approaches to Possible Bilingual Advantages. Excellent work by
@BlancoElorrieta
& Alfonso Caramazza, that puts this heated topic in its right dimension.
by reducing its potential into sth that looks dangerously like a beef with a nonagenarian who is unable to defend his ideas. Make no mistake: his ideas ab. LLMs may be wrong. But "we ALWAYS knew"? No, sorry, that's not what scientific debates should look like.
Straight to rejection, citing divergent assumptions that undermine relevance to the field. I´m sorry to say that it seems to me that these senior, tenured editors simply do not realize how detrimental this delay can be for early career researchers. 4/5
@happytinypill
Γιατί οπωσδήποτε το πρόβλημα ξεκινά από τους γονείς και όχι από την απουσία κράτους, που πρέπει να κάνεις 50 τεμενάδες για να σε αφήσουν να φύγεις από κάποιες δουλειές χωρίς early notice. Αφήστε, είχα το μωρό μου σε θερμοκοιτίδα 3 μήνες σχεδόν στην Ισπανία, ξέρω καλά το θέμα.
He had me eye-rolling at there's someone called
@GaryMarcus
. I'd also say one needs to be careful: attaching medical labels & pseudodiagnoses to people may get you in trouble. Confabulation is a neuropsychiatric disorder. I'm not saying this. The National Library of Medicine is.
New preprint out! Many people have argued that language distance modulates the degree of recruitment of cognitive control mechanisms, affecting the likelihood of bilingual adaptations.
@tagara76
@nataliagermanou
Δηλαδή πως η τοποθέτηση κάποιου στο τέλος της ζωντανής τοποθέτησης κάποιου άλλου συνιστά παγίδευση; Η μήπως θα βαράμε προσοχή στον κο Ψεκτράκο και δεν επιτρέπεται αντίλογος;
I understand some people do lots of things for a click, but it would never occur to me to LABEL people I may disagree with this way. Mary: Impossible. John: Sloppy. Kate: Unnatural. So cringe.
Today at Alba's evaluation. Doc: Preterm, right? Me: Born at 25 weeks. D: Consequences? Me: Anemia of prematurity, treated already, bronchopulmonary dysplasia, getting better, no perm damage. D: That's all? You're SO lucky! Trust me, doctor, I think so every day! 🤩
@AgEInves
resolved Consolidación Investigadora! My project TURING (The Language Understanding of Artificial Intelligence Applications) will be hiring 3 postdocs/res assistants. If you know linguistics,
#Basque
or
#Catalan
, &
#Rstats
, TURING is looking for people like you! Pls RT!
The latest addition was written in the Neonatal Intensive Care Unit. As was the last work-related email. I'm writing while sitting next to my newborn daughter who is at the NICU due to prematurity. I feel extremely lucky she's well and I hope she comes home soon. 2/4
My second story was just published. Don't let the title fool you. This is a story of love. It´s also the type of story that often stays in the margins. Let´s change that.
#WomenInScience
#SinCienciaNoHayFuturo
It does. Some scholars should stop being so self-righteous. A reviewer of my last paper wrote "I appreciate crosslinguistic investigation of illusory phenomena. However, it’s unclear what testing illusions in Greek will get us". Ugh whatever Eng got us! *You* are not the default.
the binarity. Am I a Chomskyan linguist? I´ve published results that disagree with Chomsky. And I´ve published results that agree with Chomsky. This binary framing makes me feel that I´m stuck in the middle of the road and thus hit by both generative and non-generative cars. 3/5
The new year starts in a promising way. An author submitted a revised version, responding to Reviewer 3 (me) that they believe that they read Leivada (2017) right. I´ll make a tea and re-read Leivada (2017). I hope I side with myself, but who knows these days! Happy 2021 y´all!
80% of accurate answers=LLMs not hallucinating!! This reminds me of a tweet I read once. When chocolate cookies are made out of 80% chocolate and 20% rat poison, the result is NOT delicious chocolate treats.
@GaryMarcus
@monamhamdy
@ESYudkowsky
happy to take this bet
@GaryMarcus
:-) how about we define "largely" as: 80% of the time the model produces a verifiably accurate response to a wide range of questions. and for "by 2025", let's take the middle of the year so June 2025. (if I'm right, I'll donate my proceeds)
Interesting discussion in this thread 👇'Form sentences' in AI does not equal the human ability to form sentences. If you ask me to give you a sentence you've never heard before that contains a new word, I can give you thousands. ChatGPT can't.
Does
#linguistics
twitter realize that the problem of well-formedness has been solved by technology based on next-word prediction? Imho this settles the question of the autonomy of syntax in favor of cognitive-functional and usage-based theories
My daughter's 1st earthly possession was not sth I bought. It couldn't have been, at 26 weeks nobody was ready. The nurses wrote her name on the wall in colorful letters, that was her 1st possession. I just put these letters on her door at home! Their gift to my little miracle!🙏