Filtered Corpus Training (FiCT) Shows that Language Models can Generalize from Indirect Evidence

TL;DR

This study introduces Filtered Corpus Training to evaluate if language models can generalize linguistically from indirect evidence, revealing that both LSTM and Transformer models perform well on such tasks despite differences in perplexity.

Contribution

The paper presents a novel training method, Filtered Corpus Training, to assess linguistic generalization capabilities of language models from indirect evidence.

Findings

Transformers outperform LSTMs in perplexity.

Both models perform equally well on linguistic generalization.

Language models can generalize from indirect evidence.

Abstract

This paper introduces Filtered Corpus Training, a method that trains language models (LMs) on corpora with certain linguistic constructions filtered out from the training data, and uses it to measure the ability of LMs to perform linguistic generalization on the basis of indirect evidence. We apply the method to both LSTM and Transformer LMs (of roughly comparable size), developing filtered corpora that target a wide range of linguistic phenomena. Our results show that while transformers are better qua LMs (as measured by perplexity), both models perform equally and surprisingly well on linguistic generalization measures, suggesting that they are capable of generalizing from indirect evidence.