Kanishka Misra 🌊
@kanishka.bsky.social
Assistant Professor of Linguistics, and Harrington Fellow at UT Austin. Works on computational understanding of language, concepts, and generalization.
🕸️👁️: https://kanishka.website
🕸️👁️: https://kanishka.website
I’ll be in Boston attending BUCLD this week — I won’t be presenting but I’ll be cheering on @najoung.bsky.social who will present at the prestigious SLD symposium about the awesome work by her group, including our work on LMs as hypotheses generators for language acquisition!
🤠👻
🤠👻
November 5, 2025 at 1:40 AM
I’ll be in Boston attending BUCLD this week — I won’t be presenting but I’ll be cheering on @najoung.bsky.social who will present at the prestigious SLD symposium about the awesome work by her group, including our work on LMs as hypotheses generators for language acquisition!
🤠👻
🤠👻
These preferences are weakened in sentences that replace ARC with coordination structure -- e.g., "Sue is Max's girlfriend and was a tennis champ!" (both VPs are part of the main point), suggesting that these results are not due to recency effects!
October 21, 2025 at 2:02 PM
These preferences are weakened in sentences that replace ARC with coordination structure -- e.g., "Sue is Max's girlfriend and was a tennis champ!" (both VPs are part of the main point), suggesting that these results are not due to recency effects!
DGRC happens in 4 stages (Fig 1): 1) Divide a dialogue as a prompt, 2) Generate continuations for subparts using LMs, 3) Recombine the dialogue and continuations, and 4) Compare the likelihoods of the recombined sequences.
October 21, 2025 at 2:02 PM
DGRC happens in 4 stages (Fig 1): 1) Divide a dialogue as a prompt, 2) Generate continuations for subparts using LMs, 3) Recombine the dialogue and continuations, and 4) Compare the likelihoods of the recombined sequences.
To capture 1) systematically, we use a construction that neatly separates at-issue vs. not-at-issue (not the main point) divide at the sentence level. Namely, appositive relative clause (ARC).
![example of an appositive relative clause: Did you know that Sue, [who is Max's girlfriend], was a tennis champ. [brackets indicates ARC]](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:njnapclhkbrhe3hsq44q2e4w/bafkreihj24osummvoofiqvlfm3ljebhse4mpyjp2n4rmqvobnlhtzydndi@jpeg)
October 21, 2025 at 2:02 PM
To capture 1) systematically, we use a construction that neatly separates at-issue vs. not-at-issue (not the main point) divide at the sentence level. Namely, appositive relative clause (ARC).
If I spill the tea—“Did you know Sue, Max’s gf, was a tennis champ?”—but then if you reply “They’re dating?!” I’d be a bit puzzled, since that’s not the main point! Humans can track what’s ‘at issue’ in conversation. How sensitive are LMs to this distinction?
New paper w/ @sangheekim.bsky.social!
New paper w/ @sangheekim.bsky.social!
October 21, 2025 at 2:02 PM
If I spill the tea—“Did you know Sue, Max’s gf, was a tennis champ?”—but then if you reply “They’re dating?!” I’d be a bit puzzled, since that’s not the main point! Humans can track what’s ‘at issue’ in conversation. How sensitive are LMs to this distinction?
New paper w/ @sangheekim.bsky.social!
New paper w/ @sangheekim.bsky.social!
To test this, we introduce Wugnectives, a new dataset that uses nonce words to test for inferences about the world licensed by discourse connectives.
This is important because prior work has largely tested the opposite: predict connective, given its args (often grounded in the world!)
This is important because prior work has largely tested the opposite: predict connective, given its args (often grounded in the world!)
October 16, 2025 at 3:27 PM
To test this, we introduce Wugnectives, a new dataset that uses nonce words to test for inferences about the world licensed by discourse connectives.
This is important because prior work has largely tested the opposite: predict connective, given its args (often grounded in the world!)
This is important because prior work has largely tested the opposite: predict connective, given its args (often grounded in the world!)
"Although I hate leafy vegetables, I prefer daxes to blickets." Can you tell if daxes are leafy vegetables? LM's can't seem to! 📷
We investigate if LMs capture these inferences from connectives when they cannot rely on world knowledge.
New paper w/ Daniel, Will, @jessyjli.bsky.social
We investigate if LMs capture these inferences from connectives when they cannot rely on world knowledge.
New paper w/ Daniel, Will, @jessyjli.bsky.social
October 16, 2025 at 3:27 PM
"Although I hate leafy vegetables, I prefer daxes to blickets." Can you tell if daxes are leafy vegetables? LM's can't seem to! 📷
We investigate if LMs capture these inferences from connectives when they cannot rely on world knowledge.
New paper w/ Daniel, Will, @jessyjli.bsky.social
We investigate if LMs capture these inferences from connectives when they cannot rely on world knowledge.
New paper w/ Daniel, Will, @jessyjli.bsky.social
Happening now! Poster 42!
October 8, 2025 at 3:12 PM
Happening now! Poster 42!
The compling group at UT Austin (sites.utexas.edu/compling/) is looking for PhD students!
Come join me, @kmahowald.bsky.social, and @jessyjli.bsky.social as we tackle interesting research questions at the intersection of ling, cogsci, and ai!
Some topics I am particularly interested in:
Come join me, @kmahowald.bsky.social, and @jessyjli.bsky.social as we tackle interesting research questions at the intersection of ling, cogsci, and ai!
Some topics I am particularly interested in:
September 30, 2025 at 4:17 PM
The compling group at UT Austin (sites.utexas.edu/compling/) is looking for PhD students!
Come join me, @kmahowald.bsky.social, and @jessyjli.bsky.social as we tackle interesting research questions at the intersection of ling, cogsci, and ai!
Some topics I am particularly interested in:
Come join me, @kmahowald.bsky.social, and @jessyjli.bsky.social as we tackle interesting research questions at the intersection of ling, cogsci, and ai!
Some topics I am particularly interested in:
Looking forward to attending #cogsci2025 (Jul 29 - Aug 3)! I’m especially excited to meet students who will be applying to PhD programs in Computational Ling/CogSci in the coming cycle.
Please reach out if you want to meet up and chat! Email is the best way, but DM also works if you must!
quick🧵:
Please reach out if you want to meet up and chat! Email is the best way, but DM also works if you must!
quick🧵:
July 28, 2025 at 9:20 PM
Looking forward to attending #cogsci2025 (Jul 29 - Aug 3)! I’m especially excited to meet students who will be applying to PhD programs in Computational Ling/CogSci in the coming cycle.
Please reach out if you want to meet up and chat! Email is the best way, but DM also works if you must!
quick🧵:
Please reach out if you want to meet up and chat! Email is the best way, but DM also works if you must!
quick🧵:
I am especially excited about our cute little case study showcasing that within MLM embs, different dative-constructions with the same lexical items construe different levels of animacy vs. place-hood information onto the recipient (DO: animate; PO: place/location)
July 17, 2025 at 5:47 PM
I am especially excited about our cute little case study showcasing that within MLM embs, different dative-constructions with the same lexical items construe different levels of animacy vs. place-hood information onto the recipient (DO: animate; PO: place/location)
Our tool (semantic-features) uses a method developed by Gabriella Chronis, @kmahowald.bsky.social, Katrin Erk, and generalizes it to be applied to most Masked Language Models on huggingface. You can use it to project your fav. MLM’s representations to three different semantic-norms.
July 17, 2025 at 5:47 PM
Our tool (semantic-features) uses a method developed by Gabriella Chronis, @kmahowald.bsky.social, Katrin Erk, and generalizes it to be applied to most Masked Language Models on huggingface. You can use it to project your fav. MLM’s representations to three different semantic-norms.
I will unfortunately have to skip SCiL this year, but I am thrilled to share that Jwalanthi will be presenting this work by her, @rjha.bsky.social, me, and @kmahowald.bsky.social on a tool that allows you to project contextualized embeddings from LMs to interpretable semantic spaces!
July 17, 2025 at 5:47 PM
I will unfortunately have to skip SCiL this year, but I am thrilled to share that Jwalanthi will be presenting this work by her, @rjha.bsky.social, me, and @kmahowald.bsky.social on a tool that allows you to project contextualized embeddings from LMs to interpretable semantic spaces!
My grandmother passed away a few moments ago. She was suffering from a mental illness (a rare form of dementia). It’s likely that my love for language comes from her - she was a school teacher and taught Sanskrit, Odia, and English. I hope she is at peace now.
July 9, 2025 at 8:48 PM
My grandmother passed away a few moments ago. She was suffering from a mental illness (a rare form of dementia). It’s likely that my love for language comes from her - she was a school teacher and taught Sanskrit, Odia, and English. I hope she is at peace now.
News🗞️
I will return to UT Austin as an Assistant Professor of Linguistics this fall, and join its vibrant community of Computational Linguists, NLPers, and Cognitive Scientists!🤘
Excited to develop ideas about linguistic and conceptual generalization (recruitment details soon!)
I will return to UT Austin as an Assistant Professor of Linguistics this fall, and join its vibrant community of Computational Linguists, NLPers, and Cognitive Scientists!🤘
Excited to develop ideas about linguistic and conceptual generalization (recruitment details soon!)
June 2, 2025 at 1:18 PM
News🗞️
I will return to UT Austin as an Assistant Professor of Linguistics this fall, and join its vibrant community of Computational Linguists, NLPers, and Cognitive Scientists!🤘
Excited to develop ideas about linguistic and conceptual generalization (recruitment details soon!)
I will return to UT Austin as an Assistant Professor of Linguistics this fall, and join its vibrant community of Computational Linguists, NLPers, and Cognitive Scientists!🤘
Excited to develop ideas about linguistic and conceptual generalization (recruitment details soon!)
@kmahowald.bsky.social with a beautiful high-tech illustration 🎨 while describing @qyao.bsky.social's latest paper at the HSP online seminar series!
Paper: arxiv.org/abs/2503.20850
Paper: arxiv.org/abs/2503.20850
April 4, 2025 at 7:34 PM
@kmahowald.bsky.social with a beautiful high-tech illustration 🎨 while describing @qyao.bsky.social's latest paper at the HSP online seminar series!
Paper: arxiv.org/abs/2503.20850
Paper: arxiv.org/abs/2503.20850
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
"Word" scoring is now a thing! You just have to supply your own splitting function!
pip install -U minicons for merriment
![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@jpeg)
April 2, 2025 at 3:35 AM
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
"Word" scoring is now a thing! You just have to supply your own splitting function!
pip install -U minicons for merriment
eerily similar to dark2 in ggplot!
March 12, 2025 at 3:50 AM
eerily similar to dark2 in ggplot!
Quick minicons update to those who celebrate: Please pass `chat=True` to all scoring functions in case you are using an instruct tuned model with queries that are already chat formatted; and in case you are chat formatting your queries, this is a good practice to follow:
March 12, 2025 at 12:52 AM
Quick minicons update to those who celebrate: Please pass `chat=True` to all scoring functions in case you are using an instruct tuned model with queries that are already chat formatted; and in case you are chat formatting your queries, this is a good practice to follow:
COMPS (aclanthology.org/2023.eacl-ma...) now also available in 17 additional languages! A valiant effort by He and Nie et al. 🙏
Check it out here: arxiv.org/abs/2502.19737
Check it out here: arxiv.org/abs/2502.19737
March 3, 2025 at 4:42 AM
COMPS (aclanthology.org/2023.eacl-ma...) now also available in 17 additional languages! A valiant effort by He and Nie et al. 🙏
Check it out here: arxiv.org/abs/2502.19737
Check it out here: arxiv.org/abs/2502.19737
How do people manually add *ACL conference proceedings papers to their MyNCBI Bibliography?
January 9, 2025 at 9:30 PM
How do people manually add *ACL conference proceedings papers to their MyNCBI Bibliography?
