Learning Music Helps You Read: Using Transfer to Study Linguistic Structure in Language Models

Abstract

We propose transfer learning as a method for analyzing the encoding of grammatical structure in neural language models. We train LSTMs on non-linguistic data and evaluate their performance on natural language to assess which kinds of data induce generalizable structural features that LSTMs can use for natural language. We find that training on non-linguistic data with latent structure (MIDI music or Java code) improves test performance on natural language, despite no overlap in surface form or vocabulary. Training on artificial languages containing recursion (hierarchical structure) also improves performance on natural language, again with no vocabulary overlap. Surprisingly, training on artificial languages consisting of sets of separated pairs of words, but with no recursion, improves performance on natural language as well as recursive languages do. Experiments on transfer between natural languages show that zero-shot performance on a test language is highly correlated with typological syntactic similarity to the training language, suggesting that representations induced from natural languages correspond to the cross-linguistic syntactic properties studied in linguistic typology. Our results provide insights into the ways that neural models represent abstract syntactic structure, and also about the kind of structural inductive biases which a learner needs to model language.