Towards Making a Dependency Parser See

TL;DR

This paper investigates using eye-tracking data during training to improve dependency parsing, demonstrating a multitask learning approach that leverages gaze features as auxiliary tasks without requiring them at inference.

Contribution

It introduces a novel multitask learning framework that incorporates eye-tracking data during training to enhance dependency parsing performance.

Findings

Accuracy gains are modest but positive.

The approach can utilize disjoint datasets for training.

It shows feasibility of leveraging eye-tracking data in parsing models.

Abstract

We explore whether it is possible to leverage eye-tracking data in an RNN dependency parser (for English) when such information is only available during training, i.e., no aggregated or token-level gaze features are used at inference time. To do so, we train a multitask learning model that parses sentences as sequence labeling and leverages gaze features as auxiliary tasks. Our method also learns to train from disjoint datasets, i.e. it can be used to test whether already collected gaze features are useful to improve the performance on new non-gazed annotated treebanks. Accuracy gains are modest but positive, showing the feasibility of the approach. It can serve as a first step towards architectures that can better leverage eye-tracking data or other complementary information available only for training sentences, possibly leading to improvements in syntactic parsing.