A new research paper explores how large language models (LLMs) handle anaphor resolution, a linguistic task where a word or phrase refers back to another. The study tested five open-weight LLMs—GPT-2 XL, Llama-3.1:8b, Pythia-12B, Mistral:7b, and Mistral 24B—using methods that mimic human cognitive processes, such as measuring model "surprisal" to approximate reading times and comparing model accuracy to human comprehension. The findings indicate that some LLMs demonstrate human-like sensitivity to factors like discourse prominence and distance in resolving anaphors, while showing less consistent performance with semantic interference effects. AI
IMPACT This research provides insights into the linguistic capabilities of LLMs, potentially guiding future model development for more human-like language understanding.
RANK_REASON Research paper published on arXiv detailing LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
