there is this guy on my flight to shanghai sitting next to me checking out the emnlp program and a bunch of papers and his bsky feed is 90% emnlp stuff and on the one hand it would be nice to chat but on the other hand it‘s a 12-hour flight so maybe i it’a better if i focus on my netfix downloads..
I will be attending EMNLP in China to present our paper with @bylinina.bsky.social (who will be in China, too) and Jakub Dotlacil in the BabyLM workshop! Looking forward to meeting people there! ✨ 😊 #EMNLP2025 @emnlpmeeting.bsky.social
lnkd.in/e-Bzz6De
oh super-interesting
who'll be at emnlp?
got a tiny (approx 50k) grant from NWO to do something about whether (instruction-tuned) lms are an 'agent', a superposition of agents, what's going on there epistemically and also how people interact with these 'personae' -- we'll seeeeee www.nwo.nl/en/researchp...
🚨Report your NSF grant terminations! 🚨
We are starting to collect information on NSF grant terminations to create a shared resource as we have for NIH. The more information we collect, the more we can organize, advocate, and fight back! Please share widely!
airtable.com/appGKlSVeXni...
Four large Dutch universities, including Leiden University where I work, have decided to throw international psychology bachelor programs under the bus in an effort to appease the rightwing government.
Here's my blog why this is a terrible idea.
eiko-fried.com/cutting-inte...
i just need students to see the difference between base and instruction-tuned models trying out different types of prefixes, without them needing to write any code or send their info anywhere
do we know a pair of base vs. instruct models that are both deployed by an inference provider on hf (or maybe a hf space but less preferable..) AND that don't require students sending their info for the license agreement?
i mean i'd be really surprised if what lms generate as 'reasoning' text faithfully reflected the ways they come up with the answer. like, what would guarantee that
nice!!
![from minicons import scorer
from nltk.tokenize import TweetTokenizer
lm = scorer.IncrementalLMScorer("gpt2")
# your own tokenizer function that returns a list of words
# given some sentence input
word_tokenizer = TweetTokenizer().tokenize
# word scoring
lm.word_score_tokenized(
["I was a matron in France", "I was a mat in France"],
bos_token=True, # needed for GPT-2/Pythia and NOT needed for others
tokenize_function=word_tokenizer,
bow_correction=True, # Oh and Schuler correction
surprisal=True,
base_two=True
)
'''
First word = -log_2 P(word | <beginning of text>)
[[('I', 6.1522440910339355),
('was', 4.033324718475342),
('a', 4.879510402679443),
('matron', 17.611848831176758),
('in', 2.5804288387298584),
('France', 9.036953926086426)],
[('I', 6.1522440910339355),
('was', 4.033324718475342),
('a', 4.879510402679443),
('mat', 19.385351181030273),
('in', 6.76780366897583),
('France', 10.574726104736328)]]
'''](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:njnapclhkbrhe3hsq44q2e4w/bafkreibw7fjs4zeocjmfvpr4fo7wikiqenbdz7b3clwq2rolckpoqwkssu)
another day another minicons update (potentially a significant one for psycholinguists?)
"Word" scoring is now a thing! You just have to supply your own splitting function!
pip install -U minicons for merriment
ah that's great, makes a lot of things much faster to try out!
or what happened!
you have to tell me which starter pack i apparently suddenly ended up in
Sounds familiar
the tiny books have arrived
Forthcoming titles in Elements in Semantics: 1. Abzianidze, @bylinina.bsky.social, Paperno Deep Learning and Semantics. 2.K. Davidson Semantics of Depiction, 3. Chatzikyriakidis Cooper, Gregoromichelaki, Sutton Types and the structure of meaning: Issues in compositional and lexical semantics
(2/3)
waluigi!!
all invitations i find in my inbox are actually invitations to work a bit more
yeah it's super-interesting to me somehow suddenly which i didn't expect and i don't know what to do with it but i'll just be curious about it i guess
this is so cool - it's the 2nd time i see this thread and again i think how cool it is. you know why? well for obv reasons but also bc i've been thinking recently about how the linguistic will of one person or group of people (prescriptive organizations but not necessarily) can do things to language
... buying out research time with grant budgets -- most likely gone. maybe that's just the reality of an assistant prof position (and up), maybe also amplified by budget cuts -- but is that it? am i just going to be talking most of the time rather than doing anything? depressing really
in order to actually do smth in research directions i'm interested in i need some bandwidth: research time, phd students to work with, experiment budgets. in nl it's getting more and more complicated (for obv reasons): some ways to get phd students are frozen, some grants not announced anymore..
meeee!
nah i wasn’t serious
thx!! now i’m annoyed i’m not in it
linguists? computational linguists? nlp people? semantics people? anybody?
