📣 NEW PAPER 📣 
Excited to share my first PhD paper accepted to NAACL 2024 Findings!
Can smaller #LLMs use fine-grained feedback to refine their translations? 🤔 🔎

arxiv.org/abs/2404.07851

Our EMNLP 2023 paper (that started as a discussion with Alexander Hoyle at a lab social :P) explores how a human utterance can be viewed as a collection of propositions representing the communicative intent. Turns out, that greatly helps interpret socially-laden language!

We develop a framework for measuring factual errors—facts about entities that are nonfactual, but seem contextually plausible, and contextual errors—facts about entities that are factual, but contextually irrelevant, in referring expression generation setting.

🤩 Super excited and honored to give an invited talk at *SEM *SEM 2024 #starsem co-located with #ACL2023NLP about our work on discourse and QUD! sites.google.com/view/starsem20…

The Outstanding Study Design Award for ICWSM 2023 goes to: 'Mainstream News Articles Co-Shared with Fake News Buttress Misinformation Narratives' by Pranav Goel 🇺🇦 [email protected] Philip Resnik

Sentences in a document are 🖊️answers to implicit questions from context🖊️ -- and our #ACL2023NLP Findings paper presents the first ever system to parse this! arxiv.org/abs/2210.05905
This is the third of our sequence of work co-led w/ Greg Durrett on Questions Under Discussion 🧵

Could not be more excited and proud!! Will always be deeply grateful to have worked with such a wonderful mentor :)

While current bias-detection algorithms require a list of pre-designed stereotypes, we propose an open-ended framework to uncover social biases in models. Our work will be presented at #EACL2023 (arxiv.org/abs/2210.07269).

Many thanks to Rachel Rudinger Jieyu Zhao and Zongxia!

🦷 Another preprint 🦷
Information-seeking Qs often contain questionable assumptions that models should be robust to. 'When did Marie Curie discover Uranium?' is an example. We propose (QA)^2, a test set evaluating the capacity to handle such Qs. (1/n)

arxiv.org/abs/2212.10003

When evaluating a method, it makes sense to ask: what's the method used for?

In our Findings of EMNLP paper 'Are Neural Topic Models Broken?', we draw from content analysis to inform topic model evaluation metrics

Pranav Goel Rupak Philip Resnik
arxiv.org/abs/2210.16162
1/11

Two new NL deduction papers with Zayne Sprague, Swarat Chaudhuri & Greg Durrett @ EMNLP'22!

In arxiv.org/abs/2201.06028, we split search & generation to boost consistency.
In arxiv.org/abs/2211.00614, led by Zayne, we infer unstated premises (+bonus: new dataset!) [1/7]

If you are looking to do a computational Ph.D. in a linguistics department or a linguistic-y Ph.D. in CS, consider applying to join me, Jessy Li Jessy Li, and Katrin Erk sites.utexas.edu/compling/ at UT Austin.

New paper alert!

GPT-3 is getting really popular and tons of applications are getting built with it. But before we deploy it in real-life, let’s first answer the important question: How reliable is GPT-3?

(Hint: it can be more reliable than you think!)

🧵(1/n)

If anyone needs a hard task that current transformers can't beat, Martha Palmer's #NLProc keynote COLING 2022 suggests we should test models on implicit arguments more. She reports 10-30% of English sentences have an implicit argument (I think that's likely an underestimate).

When NLI models can ignore context, do they? Our #NAACL2022 paper explores this + revisits the role of partial-input baselines in NLI! Joint work w/ Rachel Rudinger

🔗 arxiv.org/abs/2205.12181
I'll be presenting this today:
📆 In-Person PS 3 (10:45 PT)
📆 Virtual PS 2 (14:15 PT)

Hi #NAACL2022 folks! I'll be poster-ing about mine and Rachel Rudinger's paper 'Recognition of They/Them as Singular Personal Pronouns in Coreference Resolution' Monday at 10:45. Come check it out! 💛🤍💜🖤

Paper: aclanthology.org/2022.naacl-mai…
Dataset: github.com/ctbaumler/wino…

[1/5]

NAACL HLT 2024 University of Maryland researchers presenting their work at #NAACL2022 include professors Philip Resnik, Marine Carpuat, Hal Daumé III, Jordan Boyd-Graber, linda zou and Rachel Rudinger; and PhD students Neha Srikanth, Eleftheria Briakou, and Trista Cao.

Learn more: go.umd.edu/cqP

Our work w/ Ashwin Devaraj, William Sheffield, and byron wallace on factuality & simplification has been selected as an Outstanding Paper at ACL 2022! #acl2022nlp

Our brief summary video at #ACL2021nlp is now live!
underline.io/events/167/ses…

We'll be presenting this work at Poster Session #3 at the GEM workshop this Friday (1pm PT/ 4pm ET), come by and chat about our work!