Cory Shain ([email protected])
@coryshain
Followers
1,876
Following
758
Media
104
Statuses
1,608
Language in minds, brains, and machines. Postdoc, BCS @MIT . Faculty, Linguistics @Stanford (fall 2024). He/him.
Providence, RI
Joined May 2016
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Mbappe
• 997108 Tweets
Fauci
• 888119 Tweets
Real Madrid
• 722890 Tweets
Farage
• 250096 Tweets
Venom
• 155637 Tweets
WNBA
• 138539 Tweets
Aziz
• 135388 Tweets
Reform
• 133992 Tweets
Jadue
• 126681 Tweets
Nigerians
• 102703 Tweets
Marjorie Taylor Greene
• 101861 Tweets
Acun
• 97808 Tweets
مدريد
• 96545 Tweets
#loveIsland
• 84699 Tweets
Mimi
• 76310 Tweets
Caitlin Clark
• 75798 Tweets
sabrina
• 70241 Tweets
Djokovic
• 62646 Tweets
Clacton
• 59554 Tweets
Palmer
• 40654 Tweets
#HappyBirthdayKeremBürsin
• 40456 Tweets
Stephen A
• 38507 Tweets
Chloe
• 36776 Tweets
Larry Allen
• 32839 Tweets
marti
• 24669 Tweets
AIDS
• 22764 Tweets
Trent
• 19907 Tweets
Cerundolo
• 19891 Tweets
Marge
• 19446 Tweets
Bosnia
• 16906 Tweets
Jaime Caravaca
• 12494 Tweets
Pat McAfee
• 10709 Tweets
Epoch Times
• 10403 Tweets
Pinned Tweet
Now out in
@PNASNews
! Large-scale reading evidence that next-word predictability effects in humans are driven by *inference* (logarithmic in predictability) rather than preactivation (linear in predictability).
1/10
2
47
149
My toddler has learned that the formula for English question formation is "what the heck X?" like "what the heck blanket?" (where's the blanket) or "what the heck mac n cheese?" (can I have lunch). What the heck?
7
48
659
I can't believe I'm saying this, but here goes: I'm going to be a professor! At Stanford Linguistics (starting fall 2024)!
58
19
592
One thing that has struck me as a linguist+parent is the fact that my kids sometimes learn rules that are wrong, idiosyncratic, and productive. Two examples:
1/10
3
40
319
👋I boost a lot of job opportunities on here and now it's time to boost my own! I'll be arriving at
#Stanford
in fall of 2024 and I'm looking for awesome people to help me figure out language. 🧵👇
4
80
221
Example 1: When my twins were toddlers, they learned the following rule for forming reciprocals, which is more complex than the actual rule ("each other"):
our -> ourchothers
your -> yourchothers
their -> theirchothers
2/10
4
1
116
Want to understand recursion? Try reading papers the way I do it! WARNING: I have never finished a paper.
read <- func(a) {
for b in cited_by(a) {
read(b)
}
}
8
5
105
Since joining
@ev_fedorenko
's lab from A Linguistics Department, I've been surprised at the combination of misinterpretation and rancor directed toward her functional localization paradigm for language.
1/
This is a great example of how psychologists continue to use folk-scientific ideas about language even in their cutting-edge research, and why that matters: 🧵 1/
5
49
271
5
14
100
One of my academic heroes is William Sealy Gosset, aka "Student". My admiration for him comes less from his contributions (which were huge) than from the model of his life. Many haven't heard of him, so here's a 🧵:
1/20
1
19
95
This is why language development is such a fun puzzle: it's hard to explain BOTH the striking uniformity of developmental trajectories and adult language competence AND the flexible data-driven learning processes that yield idiosyncratic and contingent incorrect inferences.
8/10
2
2
89
The Gricean Toddler
Me: I love you.
T: I love Mom.
Me: I love Mom, too, but don't you love me?
T (pause): I *just* love Mom.
5
4
90
*New preprint!* "Word Frequency and Predictability Dissociate in Naturalistic Reading". TLDR; Multiple datasets show that word frequency has a big effect on reading times that is not reducible to predictability, even during naturalistic reading.
1/22
2
15
88
Our toddler used to say "waba" for "water bottle" but eventually figured out that this isn't enough syllables. So now it's been corrected to "wabawaba".
2
4
86
New preprint! Large-Scale Evidence for Logarithmic Effects of Word Predictability on Reading Time . With
@clara__meister
,
@tipimentelms
,
@ryandcotterell
, &
@roger_p_levy
.
TLDR; supports a key role of probabilistic inference in language processing.
1/
3
28
82
New preprint!
"'Constituent length' effects in fMRI do not provide evidence for abstract syntactic processing"
With
@HopeKean
,
@ben_lipkin
,
@AffourtitJosef
,
@MattSiegelman
,
@FrancisMollica
, &
@ev_fedorenko
1/
3
22
81
And yes :), if you did the math, I have 4 kids. They do a lot of talking and I try to do a lot of listening.
11/10
3
0
74
Example 2: My youngest (still a toddler) has learned to always add the contracted copula in addition to the actual copula:
I'm am
we're are
you're are
He otherwise uses subject pronouns correctly (I want, you like, etc.)
3/10
1
2
74
New preprint! "Robust effects of working memory demand during naturalistic language comprehension in language-selective cortex" with
@IbanDlank
@ev_fedorenko
@LanguageMIT
and William Schuler
1/
1
16
76
New preprint! "A Deep Learning Approach to Analyzing Continuous-Time Systems" (with William Schuler)
We propose a way to use deep learning to *analyze* (not just imitate) complex continuous-time processes.
1/
2
19
71
It also strikes me how easy it is to miss these observations precisely because they're idiosyncratic. Non-dorky non-linguist parents may not notice or report them, and most experimental and corpus methods are designed to find population-level patterns.
9/10
1
2
65
New paper alert! "Similarity of Computations Across Domains Does Not Imply Shared Implementation: The Case of Language Comprehension" with
@ev_fedorenko
in Current Directions in Psychological Science: .
1/
2
11
67
These errors are very cute so we naturally repeat them back to them all the time and even use them jokingly among each other, reinforcing the wrong pattern possibly even more frequently than the right one.
4/10
2
1
62
I followed the advice to delete "very" in my writing, so I did that ewhere and now ething is so much punchier. Works e time!
1
1
61
These errors can also be contagious for other learners. The twins' older brother started (earnestly) using their incorrect reciprocal form based on their input, even though he had never done this before.
5/10
1
1
57
What kinds of clues about our amazing language learning abilities are lurking in the diversity of errorful learning trajectories that we explore (and eventually find our way out of) as individual children?
10/10
3
0
56
I'm not aware of previous reports of these specific error patterns, either from the research literature or from other parents. Despite highly similar genes and experiences, my kids borked the English language in unique and revealing ways, all on their own.
7/10
2
1
53
New preprint! "No evidence of theory of mind reasoning in the human language network" with
@ampaunov
,
@Jessica_ChenXY
,
@ben_lipkin
, &
@ev_fedorenko
.
*Alex, Jessica and I contributed equally
1/
3
13
58
Nothing can stop Praat from playing that sound you clicked on. Nothing.
3
4
53
Whenever I get too jazzed about the power of human statistical learning I remember that I'm still at chance accuracy on which switch turns on the light vs the ceiling fan
3
0
50
Now to appear in Cognition! Continuous-time deconvolutional regression (CDR, ) estimates cognitive dynamics from online experimental measures. Tested on synthetic, self-paced reading, eye tracking, and fMRI data (also applies to EEG, MEG, and others).
1/
2
14
49
Despite these adverse learning conditions, they correct these patterns relatively quickly (a few months).
6/10
1
0
47
This morning I swooped my 3yo up in a big hug and said, "I love you".
His eyes were kind, but full of pity.
He replied, "I don't love you. I love Mom."
3
0
45
Doing the same thing over and over expecting a different result. You may call it insanity. I call it sampling.
0
3
45
My lab will focus broadly on language in minds, brains, and machines (learning, representation, and processing), using computational, behavioral, and neuroimaging approaches. If you're a prospective student interested in this stuff, let's talk!
4
3
44
To be safe, I'm correcting my p value by the total number of hypothesis tests ever conducted. Anyone have that number off hand?
3
1
35
Excited to announce a new paper with Micha Elsner
@conll_conf
: "Acquiring language from speech by learning to remember and predict".
1/n
2
8
32
So excited to be in Pittsburgh for
#HSP2023
, my first in-person conference in almost 4 years 🤯!
Here are some thumbnails of the work I'll be presenting this week. Find me in person or online to chat about any of them (or anything else!):
1
6
31
Our
#Cognition
paper on continuous-time deconvolutional regression (
#CDR
) is finally out!
I've already talked on here about the general idea 👇, so instead, here's a practical thread on how to actually use it to analyze your data.
1/
Now to appear in Cognition! Continuous-time deconvolutional regression (CDR, ) estimates cognitive dynamics from online experimental measures. Tested on synthetic, self-paced reading, eye tracking, and fMRI data (also applies to EEG, MEG, and others).
1/
2
14
49
1
5
32
My approach to the cocktail party problem of being an interdisciplinary researcher
Them: So what do you do?
Me: I'm a postdoc at blabla
Them: Oh really? What field?
Me: I don't know
0
2
28
Incidental finding 1: GPT-2(-small) is a much better estimator of human surprisal than GPT-3, even though GPT-3 is 1000x bigger and trained on much more data. Large transformer LMs may be "too good" at next-word prediction.
18/
1
6
27
Kid got a homework problem asking them to show that there are no positive integers a, b, c such that a^n + b^n = c^n for integer n > 2. I worked on it all evening and I'm stumped. These new math standards are tough!
4
0
25
Now in press at JNeurosci!
New preprint! "Robust effects of working memory demand during naturalistic language comprehension in language-selective cortex" with
@IbanDlank
@ev_fedorenko
@LanguageMIT
and William Schuler
1/
1
16
76
0
4
26
Pro tip: I use the
@zotero
browser widget to quickly save papers I see on social media to my reading list. Not having time to read papers that are on my list feels much better than not having time to read papers that are not on my list.
2
0
24
PSA in hopes of making everyone feel better about themselves: I just finished the poster for the conference abstract *that was accepted as an oral presentation*.🤦
0
0
24
Toddlers reveal our hidden assumptions by messing with them. Today I let mine put his bread in the toaster and he put it in UPSIDE DOWN. This makes no practical difference but it bothers me to no end.
1
1
24
1. He embodied the disinterested pursuit of knowledge. His pay and reputation were totally disconnected from his academic work. He was not after citation counts, society honors, or "legacy".
10/20
1
1
24
Hi all! A 🧵about my
#ACL2021NLP
paper on discovering complex dynamics in human cognition, which I'll be taking questions about live on Tuesday Aug 3 @ 12:30 EST in ACL session 11B.
1/
1
2
23
My laptop just briefly attained sentience, typing random numbers and letters on its own. After a panicked restart and some web shopping for new computers, realized the toddler had found my bluetooth keyboard in the other room 🥴.
0
2
22
@LanguageMIT
The fact that that's not what it is makes me suspect this comes from siblings instead of me
0
0
20
lol
2 of the key examples of America's "education dominance" (MIT & Caltech) don't even have legacy admissions
4
2
22
My 3yo says "hostable" instead of "hospital". I like this because it swaps places of articulation but keeps the voicing features where they should be.
3
2
21
A quick bump for those in the back: PhD applications for
#StanfordLinguistics
are due *Nov 30*: . Please RT!
1
12
21
Sometimes I look at my own sentences and feel profound relief that they are not in the treebank I'm annotating.
1
0
19
8 yo spontaneously made me regalia and a diploma for my zoom commencent 😍
0
0
20
My toddler has started asking what time it is.
He has no concept of time.
I tell him it's 26.
1
0
17
I owe this to a huge number of people, but esp
@RSmithShain
and my (official and unofficial) mentors William Schuler, Micha Elsner,
@ev_fedorenko
,
@marty_with_an_e
,
@LanguageMIT
, and
@roger_p_levy
.
1
0
18
Proposal: Give authors+reviewers anonymous email addresses so they can ask each other clarifying q's before submitting official reviews/responses.
1/10
2
0
18
If you're a psychologist interested in generating good nonce words for experiments, my toddler is serviceable. A piece of playground equipment was "spoodgered" and he keeps telling me to "landle down". But they're catchy so act now before they catch on.
0
1
19
You know a field is moving (too?) fast when people keep referring to papers from last year as "classical"
1
1
18
Can I make a humble request to get a decision *immediately* whenever I apply for something? This would improve my mental health greatly. Thx.
1
0
18
An EvLab study led by
@tamaregev
has in fact shown exactly this! The morphophonological plausibility of nonwords modulates the degree to which they engage the language network.
8/
Excited to share the first preprint from my postdoc!
We show that brain regions that process “high-level” meaning in language also process meaningless phoneme clusters, according to how much they respect linguistic phoneme-combinatorial constraints.
1/n
3
13
82
3
0
17
8yo: How do you say "I agree" in French?
me: Je suis d'accord.
8yo: I am... agree?
me: *long explanation about how translation isn't word by word, with several examples*
8yo (smirking): I am agree!
0
0
18
This Christmas my toddler gave us a new word: forbrought. Used when you forgot to put a toy you were holding away and brought it along instead.
0
2
16
As he made discoveries, he wanted to make them public, but was contractually prevented from doing so by his employer. The workaround they converged on was for him to publish under a pseudonym. He chose "Student". That's the only author attribution in any of his >20 papers.
6/20
1
0
17
So exciting to see a computational psycholinguistics paper in Nature! Congrats
@LakeBrenden
& Marco Baroni!
Today in Nature, we show how a standard neural net, optimized for compositional skills, can mimic human systematic generalization (SG) in a head-to-head comparison. This is the capstone of a 5 year effort with Marco Baroni to make progress on SG. (1/8)
24
396
2K
0
1
17
And on the very last day that this can be called Shain, Paunov, Chen et al *2022*, our article on language and theory of mind is out in Cerebral Cortex!
@ampaunov
@Jessica_ChenXY
@ben_lipkin
@ev_fedorenko
Happy New Year everybody!
New preprint! "No evidence of theory of mind reasoning in the human language network" with
@ampaunov
,
@Jessica_ChenXY
,
@ben_lipkin
, &
@ev_fedorenko
.
*Alex, Jessica and I contributed equally
1/
3
13
58
0
7
17
I had a great time talking at virtual
#CUNY2020
today, thanks all for your great questions! Stay tuned for a link to the recording, which should be up soon...
#CUNY2020
great presentation by
@coryshain
#CUNY2020
who finds that Surprisal, both on the basis of word strings (5 grams) and at the level of grammatical constructions (e.g, NP VP; 'syntax') increase activation in the language-network
1
3
17
0
0
16
(Let's pause and note how few structural incentives he had to do public-facing research. He did not hold a research job. Publication was actively discouraged. And publications were pseudonymous. As a result, he is to this day much less well known than his contemporaries.)
7/20
1
0
16
2. "Large-Scale Evidence for Logarithmic Effects of Word Predictability on Reading Time" (Spoken Session 5, submission
#32
), with
@clara__meister
,
@tpimentelms
,
@ryandcotterell
, and
@roger_p_levy
Plenary talk!
For more details, see this thread:
New preprint! Large-Scale Evidence for Logarithmic Effects of Word Predictability on Reading Time . With
@clara__meister
,
@tipimentelms
,
@ryandcotterell
, &
@roger_p_levy
.
TLDR; supports a key role of probabilistic inference in language processing.
1/
3
28
82
2
4
16
Only partly joking: I think a good indicator of the reliability of a scientific idea/result is the number of times the authors have publicly reversed themselves.
1/4
1
2
16
Arguably the best rhetorical trick is to put "arguably" in front of claims that you can't defend but still want people to believe.
1
2
15
I had a great time talking with Stephen in depth about our latest preprint on language, working memory, and the brain: ! If you're interested, I'll be taking questions at
@SNLmtg
slide slam session B, Tues Oct 5 3:30pm ET (or wherever questions are asked).
Language Neuroscience Podcast
#16
, I talk with
@CoryShain
about his fascinating new preprint on working memory demand in naturalistic language comprehension. Get yourself in the mood for
@SNLmtg
this week!
@ev_fedorenko
@IbanDlank
@LanguageMIT
1
22
82
0
3
14
We're trying to build a plausibly "complete" set of word features that are thought to affect lexical access. Punting to the community: anyone have thoughts?
#AcademicTwitter
@ev_fedorenko
@GretaTuckute
@IbanDlank
8
6
13
8yo: Wacha doing?
Me: Writing a paper
8yo: Boring
Me: YOU'RE boring
8yo: You're MORE boring
Me: You're INFINITY boring
8yo: 😑
Did I win that? Yes I did.
1
0
14
Got a new under-review paper with William Schuler on deconvolutional regression for psycholinguistics!
5
2
14
He seems to have been motivated by (i) enjoyment of intellectual life, as revealed by hundreds of letters to colleagues, and (ii) belief that his results would be useful public knowledge and therefore worth fighting to publish, even if detached from his own name.
11/20
1
1
14
3. He chose the pseudonym "Student", which suggests a posture of intellectual humility in which the primary role of an expert is to learn rather than pontificate.
13/20
1
0
13
Just like my kids, ChatGPT is only funny when not trying to be
1
0
13
Saw a sign at the zoo saying beavers are 2nd only to humans in their impact on their environment... which... I mean... good for you, beavers... but, let's be honest, it's a distant 2nd. Come back when you can bring about your own extinction.
1
1
12
My 2yo says "on" to mean both "on" and "off", so today started flipping light switches going: "on, on, on, on, ..."
0
0
12
I helped implement a biologically plausible neural network and now it wakes me up crying at all hours of the night.
AI researchers: Be careful what you wish for!
1
0
12
Parenthood is listening politely while your kid reads aloud the entire copyright page of the book he's reading.
0
0
11
3yo pulls something down and then catches it before it hits the ground, says:
I almost caught it, but I did.
0
0
12
(Ironically, it was actually this acoustic degradation contrast that was used to obtain the key results in the epic new study by
@saima_mm
and colleagues and that
@pictureofitself
is critiquing, *not* the S>N task being critiqued!)
16/
1
0
12
But my hope here is only to draw attention to Gosset's apparent model of ego-free science (much in contrast to his contemporaries), which I'm inspired by (and try to emulate).
19/20
2
0
12
@hakeemjefferson
I think the simplest solution to this problem is for you to run for Congress
0
0
12
Anyway, publish he did for three decades until his death, and by going above and beyond in his efforts to solve practical problems at a specific brewery, he laid methodological and theoretical foundations used by scientists in all fields for reasoning under uncertainty.
8/20
1
1
12
Excited to announce that I (a computational linguist) have accepted an invitation to be a Speaker at International Webinar on Diamond, Carbon & Advanced Energy Materials for the theme track "Explore New Trends for Next Generations"!
0
0
12
5. He seems to have been very community oriented, collaborating at points with his wife Marjory Gosset (herself a superstar hockey player whose professional career lasted into her 50s!) and fondly remembered by his children and a wide circle of friends.
15/20
1
1
12
4. He took methods seriously. He recognized that his goals (food production) were blocked by general technical limitations and so dedicated himself to working on general solutions, even though this was likely not the fastest path to near-term success at the brewery.
14/20
1
1
12
I've submitted one paper a day for the past two days. Assuming I continue at this rate and everything gets accepted, I'll catch up with Paul Erdős in... FOUR YEARS
0
0
10
As a scientist, I think it's antiscientific for us to be so influenced by a few top journals that chase sexy results and bury everything important in supplemental.
But as a lazybones, it's such a relief to only have to read 4-6 pages...
0
0
11
It's ok, we don't need to bother bringing a water bottle, they have a drinking fountain there
1
0
11
2. He appears to have been genuinely humble. It's a trait mentioned in the recollections of many who knew him, and his letters reveal his belief that (i) his research was only relevant to a narrow few (!) and (ii) anything he found would have been found by others.
12/20
1
0
11
For some reason by toddler has concluded that "hate" means "like", so when we're doing something fun he'll give a big affectionate smile and say "I hate this".
0
0
10
This is just... really awesome
Can you make a jigsaw puzzle with two different solutions? Or an image that changes appearance when flipped?
We can do that, and a lot more, by using diffusion models to generate optical illusions!
Continue reading for more illusions and method details 🧵
15
119
602
0
0
11
One of my 3yo's favorite new words is "restrupt". I'll leave the meaning to your imagination, but no matter what you definitely do *not* want to get restrupted.
2
0
11