# The (Non-)Utility of Structural Features in BiLSTM-based Dependency Parsers

## TL;DR

This paper investigates whether BiLSTM-based dependency parsers implicitly capture structural features traditionally used in non-neural models, finding that explicit structural features become redundant but still influence performance.

## Contribution

It provides a detailed analysis of the implicit structural information in BiLSTM representations and demonstrates the significance of structural context through ablation studies.

## Key findings

- Structural features are largely redundant in BiLSTM parsers.
- Implicit structural information significantly influences parser performance.
- BiLSTM representations encode structural context implicitly.

## Abstract

Classical non-neural dependency parsers put considerable effort on the design of feature functions. Especially, they benefit from information coming from structural features, such as features drawn from neighboring tokens in the dependency tree. In contrast, their BiLSTM-based successors achieve state-of-the-art performance without explicit information about the structural context. In this paper we aim to answer the question: How much structural context are the BiLSTM representations able to capture implicitly? We show that features drawn from partial subtrees become redundant when the BiLSTMs are used. We provide a deep insight into information flow in transition- and graph-based neural architectures to demonstrate where the implicit information comes from when the parsers make their decisions. Finally, with model ablations we demonstrate that the structural context is not only present in the models, but it significantly influences their performance.