arXiv:2606. 16407v1 Announce Type: cross Abstract: Faithful and robust pronoun use is important for fair and coherent generations, yet large language models largely fail when multiple referents use different pronouns.
By Katharina Trinley, Jesujoba O. Alabi, Dietrich Klakow, Vagrant Gautam
The study uses sparse autoencoders (SAEs) to probe how the Pythia-70M language model encodes grammatical number. By identifying and manipulating SAE‑derived features such as "plural noun" and "generate a plural verb," the authors show that removing these features forces the model to produce singular verbs, and vice versa. Similar asymmetries were observed in larger models (Gemma‑2‑2b and Llama‑3.1‑8b), indicating a privative representation of number in English.
By Jacob Solinsky (University of Minnesota), Serguei Pakhomov (University of Minnesota)
MechaTerp-TRACE is a new framework that systematically ablates individual components of language models to measure their causal contribution to producing a named entity. By applying TRACE to thirteen instruction‑tuned dense decoder models, the study finds that a small set of positionally fixed components consistently carry the most influence across models and prompts, while the remaining support is evenly distributed. This suggests that entity knowledge is largely embedded in generic generation machinery rather than in isolated, findable components.
By Brandon Colelough, Davis Bartels, Madeline Bittner, Dina Demner-Fushman
The study examines how vision‑language models handle multi‑turn pragmatic interpretation in iterated reference games, where participants repeatedly identify novel referents using language. Researchers compared human performance with that of several models, manipulating context by varying its amount, order, and relevance. While humans consistently performed well, the models could use prior context but struggled to build relevant context for effective interpretation, indicating missing core skills for efficient linguistic collaboration.
By Alvin Wei Ming Tan, Ben Prystawski, Veronica Boyce
arXiv:2608. 13484v1 Announce Type: cross Abstract: When asked about entities outside their knowledge boundary, LLMs routinely fabricate plausible-sounding details rather than backing off to safer, more general claims.
By Dananjay Srinivas, Saksham Khatwani, Maria Pacheco
arXiv:2609.14207v1 Announce Type: new
Abstract: We propose to finetune vision-language models to generate more pragmatically optimal referring expressions by transforming observations of incremental...
By T\'ea Wright, Alane Suhr