Modeling: A Novel Dataset For Testing Linguistic Reasoning In Language Models

Chi Nathan A., Malchev Teodor, Kong Riley, Chi Ryan A., Huang Lucas, Chi Ethan A., Mccoy R. Thomas, Radev Dragomir. Arxiv 2024

[Paper]    
Few Shot GPT Model Architecture Training Techniques

We introduce modeLing, a novel benchmark of Linguistics Olympiad-style puzzles which tests few-shot reasoning in AI systems. Solving these puzzles necessitates inferring aspects of a language’s grammatical structure from a small number of examples. Such puzzles provide a natural testbed for language models, as they require compositional generalization and few-shot inductive reasoning. Consisting solely of new puzzles written specifically for this work, modeLing has no risk of appearing in the training data of existing AI systems: this ameliorates the risk of data leakage, a potential confounder for many prior evaluations of reasoning. Evaluating several large open source language models and GPT on our benchmark, we observe non-negligible accuracy, demonstrating few-shot emergent reasoning ability which cannot merely be attributed to shallow memorization. However, imperfect model performance suggests that modeLing can be used to measure further progress in linguistic reasoning.

Similar Work