On the Branching Bias of Syntax Extracted from Pre-trained Language Models

TL;DR

This paper investigates the branching bias in syntax extraction from pre-trained language models, revealing how certain methods and factors can inflate performance on specific languages due to bias, and proposes a quantitative measurement approach.

Contribution

It introduces a method to measure branching bias by comparing language and reversed language performance, and analyzes factors influencing this bias in syntax extraction.

Findings

Existing methods exhibit branching biases.

Parsing algorithms, feature definitions, and language models can introduce bias.

Bias inflates performance on certain languages.

Abstract

Many efforts have been devoted to extracting constituency trees from pre-trained language models, often proceeding in two stages: feature definition and parsing. However, this kind of methods may suffer from the branching bias issue, which will inflate the performances on languages with the same branch it biases to. In this work, we propose quantitatively measuring the branching bias by comparing the performance gap on a language and its reversed language, which is agnostic to both language models and extracting methods. Furthermore, we analyze the impacts of three factors on the branching bias, namely parsing algorithms, feature definitions, and language models. Experiments show that several existing works exhibit branching biases, and some implementations of these three factors can introduce the branching bias.