Grammar Aware Language Models

This page cannot be shown here. You can still read it on the original site — the toolbar below keeps your place in the directory.

Large language models (LLMs) demonstrate emergent and strong grammatical fluency, but can still violate the constraints of a language’s grammar such as in code generation or formal reasoning. We investigate whether an explicit, lightweight grammar prior can improve syntactic correctness without modifying the weights of the model itself. We propose a grammar aware scoring framework that…

Read on /posts/grammar-aware-language-models/ ↗