Investigating representations of verb bias in neural language models

TL;DR

This paper introduces DAIS, a large benchmark dataset for studying verb bias in English, and evaluates how well neural language models, especially transformers, capture human preferences in verb-argument constructions.

Contribution

The paper provides a new dataset, DAIS, and systematically compares neural models, revealing transformers' superior ability to model verb bias over recurrent models.

Findings

Larger models outperform smaller ones.

Transformers like GPT-2 outperform LSTMs.

Transformers better integrate lexical and grammatical info.

Abstract

Languages typically provide more than one grammatical construction to express certain types of messages. A speaker's choice of construction is known to depend on multiple factors, including the choice of main verb -- a phenomenon known as \emph{verb bias}. Here we introduce DAIS, a large benchmark dataset containing 50K human judgments for 5K distinct sentence pairs in the English dative alternation. This dataset includes 200 unique verbs and systematically varies the definiteness and length of arguments. We use this dataset, as well as an existing corpus of naturally occurring data, to evaluate how well recent neural language models capture human preferences. Results show that larger models perform better than smaller models, and transformer architectures (e.g. GPT-2) tend to out-perform recurrent architectures (e.g. LSTMs) even under comparable parameter and training settings. Additional analyses of internal feature representations suggest that transformers may better integrate specific lexical information with grammatical constructions.