The Grammar of Transformers: A Systematic Review of Interpretability Research on Syntactic Knowledge in Language Models

TL;DR

This systematic review analyzes 337 studies on Transformer models' syntactic abilities, highlighting strengths in form-oriented phenomena and weaknesses in syntax-semantics interface tasks, with recommendations for broader, more comprehensive future research.

Contribution

It provides a comprehensive overview of interpretability research on syntactic knowledge in Transformers, identifying current limitations and proposing directions for future work.

Findings

Transformers perform well on form-oriented syntactic phenomena.

Performance varies and is weaker on syntax-semantics interface phenomena.

Research is heavily focused on English and BERT, limiting generalizability.

Abstract

We present a systematic review of 337 articles evaluating the syntactic abilities of Transformer-based language models, reporting on 1,015 model results from a range of syntactic phenomena and interpretability methods. Our analysis shows that the state of the art presents a healthy variety of methods and data, but an over-focus on a single language (English), a single model (BERT), and phenomena that are easy to get at (like part of speech and agreement). Results also suggest that TLMs capture these form-oriented phenomena well, but show more variable and weaker performance on phenomena at the syntax-semantics interface, like binding or filler-gap dependencies. We provide recommendations for future work, in particular reporting complete data, better aligning theoretical constructs and methods across studies, increasing the use of mechanistic methods, and broadening the empirical scope regarding languages and linguistic phenomena.