A Study of the Plausibility of Attention between RNN Encoders in Natural Language Inference

TL;DR

This paper evaluates the plausibility of attention maps in RNN-based models for natural language inference, finding that heuristic-based annotations correlate better with human judgments than raw attention weights, raising questions about their interpretability.

Contribution

It provides a preliminary assessment of attention map plausibility in sentence comparison tasks, specifically in natural language inference, and compares heuristic and human annotations.

Findings

Heuristic annotations reasonably correlate with human judgments.

Raw attention weights are only loosely related to plausible explanations.

Attention maps' interpretability in NLI remains uncertain.

Abstract

Attention maps in neural models for NLP are appealing to explain the decision made by a model, hopefully emphasizing words that justify the decision. While many empirical studies hint that attention maps can provide such justification from the analysis of sound examples, only a few assess the plausibility of explanations based on attention maps, i.e., the usefulness of attention maps for humans to understand the decision. These studies furthermore focus on text classification. In this paper, we report on a preliminary assessment of attention maps in a sentence comparison task, namely natural language inference. We compare the cross-attention weights between two RNN encoders with human-based and heuristic-based annotations on the eSNLI corpus. We show that the heuristic reasonably correlates with human annotations and can thus facilitate evaluation of plausible explanations in sentence comparison tasks. Raw attention weights however remain only loosely related to a plausible explanation.