A Hierarchical and Attentional Analysis of Argument Structure Constructions in BERT Using Naturalistic Corpora

TL;DR

This paper explores how BERT processes argument structure constructions using a multi-dimensional analysis framework, revealing a hierarchical structure where construction-specific information emerges and is maintained across layers.

Contribution

It introduces a comprehensive analytical approach combining multiple techniques to understand BERT's internal representations of argument structures.

Findings

Hierarchical organization of argument structure information in BERT layers

Construction-specific information appears early and remains through layers

Clusters of argument structures are maximally separable in middle layers

Abstract

This study investigates how the Bidirectional Encoder Representations from Transformers model processes four fundamental Argument Structure Constructions. We employ a multi-dimensional analytical framework, which integrates MDS, t-SNE as dimensionality reduction, Generalized Discrimination Value (GDV) as cluster separation metrics, Fisher Discriminant Ratio (FDR) as linear diagnostic probing, and attention mechanism analysis. Our results reveal a hierarchical representational structure. Construction-specific information emerges in early layers, forms maximally separable clusters in middle layers, and is maintained through later processing stages.