Interpreting Attention Models with Human Visual Attention in Machine Reading Comprehension

TL;DR

This study investigates how neural attention mechanisms in NLP models relate to human visual attention by using eye-tracking data, revealing that different architectures learn distinct attention strategies.

Contribution

The paper introduces a new eye-tracking dataset and compares neural attention with human attention across various models in machine reading comprehension.

Findings

LSTM and CNN models show higher similarity to human attention correlating with performance.

XLNet performs best overall but does not align closely with human attention.

Different neural architectures develop distinct attention strategies.

Abstract

While neural networks with attention mechanisms have achieved superior performance on many natural language processing tasks, it remains unclear to which extent learned attention resembles human visual attention. In this paper, we propose a new method that leverages eye-tracking data to investigate the relationship between human visual attention and neural attention in machine reading comprehension. To this end, we introduce a novel 23 participant eye tracking dataset - MQA-RC, in which participants read movie plots and answered pre-defined questions. We compare state of the art networks based on long short-term memory (LSTM), convolutional neural models (CNN) and XLNet Transformer architectures. We find that higher similarity to human attention and performance significantly correlates to the LSTM and CNN models. However, we show this relationship does not hold true for the XLNet models -- despite the fact that the XLNet performs best on this challenging task. Our results suggest that different architectures seem to learn rather different neural attention strategies and similarity of neural to human attention does not guarantee best performance.