SyntaxGym is a unified platform for targeted syntactic evaluation of language models. The Gym supports all steps of the evaluation process, from designing test suites to visualizing final results. Our goal is to make psycholinguistic assessment of language models more standardized, reproducible, and accessible to a wide variety of researchers.
Create new psycholinguistic test suites, or browse existing ones in our database.
Evaluate a set of neural language models ranging in architecture and size.
Visualize results across models and test suites through interactive charts.