Local Structure Matters Most: Perturbation Study in NLU

TL;DR

This paper investigates how neural language understanding models primarily rely on local text structure rather than global word order, revealing their insensitivity to word-order perturbations across various tokenization schemes.

Contribution

It introduces order-altering perturbations at multiple linguistic levels and demonstrates that models depend mainly on local structure, challenging assumptions about their sensitivity to global word order.

Findings

Models are insensitive to global word order changes.

Local structure is crucial for model understanding.

Perturbations mainly affect global, not local, text structure.

Abstract

Recent research analyzing the sensitivity of natural language understanding models to word-order perturbations has shown that neural models are surprisingly insensitive to the order of words. In this paper, we investigate this phenomenon by developing order-altering perturbations on the order of words, subwords, and characters to analyze their effect on neural models' performance on language understanding tasks. We experiment with measuring the impact of perturbations to the local neighborhood of characters and global position of characters in the perturbed texts and observe that perturbation functions found in prior literature only affect the global ordering while the local ordering remains relatively unperturbed. We empirically show that neural models, invariant of their inductive biases, pretraining scheme, or the choice of tokenization, mostly rely on the local structure of text to build understanding and make limited use of the global structure.