A Study on How Attention Scores in the BERT Model are Aware of Lexical Categories in Syntactic and Semantic Tasks on the GLUE Benchmark

TL;DR

This paper investigates how BERT's attention scores reflect lexical categories during fine-tuning on GLUE tasks, revealing task-dependent and task-agnostic lexical biases in attention patterns.

Contribution

It provides empirical evidence that BERT's attention scores are influenced by lexical categories and identifies layers with consistent lexical biases across tasks.

Findings

Attention scores vary with lexical categories during fine-tuning.

Content words are emphasized in semantic tasks, function words in syntactic tasks.

Certain BERT layers show task-independent lexical biases.

Abstract

This study examines whether the attention scores between tokens in the BERT model significantly vary based on lexical categories during the fine-tuning process for downstream tasks. Drawing inspiration from the notion that in human language processing, syntactic and semantic information is parsed differently, we categorize tokens in sentences according to their lexical categories and focus on changes in attention scores among these categories. Our hypothesis posits that in downstream tasks that prioritize semantic information, attention scores centered on content words are enhanced, while in cases emphasizing syntactic information, attention scores centered on function words are intensified. Through experimentation conducted on six tasks from the GLUE benchmark dataset, we substantiate our hypothesis regarding the fine-tuning process. Furthermore, our additional investigations reveal the presence of BERT layers that consistently assign more bias to specific lexical categories, irrespective of the task, highlighting the existence of task-agnostic lexical category preferences.