CxGBERT: BERT meets Construction Grammar

TL;DR

This paper investigates whether BERT captures constructional information, finding that it does encode significant constructional knowledge, which has implications for understanding what deep learning models learn from language.

Contribution

The study introduces probes and experiments demonstrating BERT's access to constructional information, bridging linguistic theory and deep learning.

Findings

BERT encodes constructional information extensively.

Constructional information is redundantly represented in lexico-semantics.

BERT's knowledge includes aspects traditionally associated with constructions.

Abstract

While lexico-semantic elements no doubt capture a large amount of linguistic information, it has been argued that they do not capture all information contained in text. This assumption is central to constructionist approaches to language which argue that language consists of constructions, learned pairings of a form and a function or meaning that are either frequent or have a meaning that cannot be predicted from its component parts. BERT's training objectives give it access to a tremendous amount of lexico-semantic information, and while BERTology has shown that BERT captures certain important linguistic dimensions, there have been no studies exploring the extent to which BERT might have access to constructional information. In this work we design several probes and conduct extensive experiments to answer this question. Our results allow us to conclude that BERT does indeed have access to a significant amount of information, much of which linguists typically call constructional information. The impact of this observation is potentially far-reaching as it provides insights into what deep learning methods learn from text, while also showing that information contained in constructions is redundantly encoded in lexico-semantics.