Cross-Lingual Dependency Parsing with Late Decoding for Truly Low-Resource Languages

TL;DR

This paper introduces a graph-based neural network parser for cross-lingual dependency parsing that improves transfer accuracy by avoiding early decoding, leading to better performance across multiple languages.

Contribution

It presents a novel end-to-end parser that directly projects edge score matrices, simplifying the process and enhancing accuracy over previous methods.

Findings

Achieved 2.25% absolute improvement across 10 languages.

Simplified the cross-lingual dependency parsing process.

Demonstrated effectiveness of late decoding in transfer learning.

Abstract

In cross-lingual dependency annotation projection, information is often lost during transfer because of early decoding. We present an end-to-end graph-based neural network dependency parser that can be trained to reproduce matrices of edge scores, which can be directly projected across word alignments. We show that our approach to cross-lingual dependency parsing is not only simpler, but also achieves an absolute improvement of 2.25% averaged across 10 languages compared to the previous state of the art.