Cross-lingual Semantic Role Labeling with Model Transfer

TL;DR

This paper introduces an end-to-end cross-lingual semantic role labeling model that leverages universal features, transfer methods, and pre-trained representations, demonstrating significant performance variations based on feature quality and type.

Contribution

It proposes a comprehensive end-to-end model for cross-lingual SRL incorporating diverse universal features and transfer techniques, filling a gap in prior research.

Findings

Gold syntax features significantly improve SRL performance.

Universal dependency features yield the best transfer results.

Pre-trained high-order features and contextualized embeddings enhance accuracy.

Abstract

Prior studies show that cross-lingual semantic role labeling (SRL) can be achieved by model transfer under the help of universal features. In this paper, we fill the gap of cross-lingual SRL by proposing an end-to-end SRL model that incorporates a variety of universal features and transfer methods. We study both the bilingual transfer and multi-source transfer, under gold or machine-generated syntactic inputs, pre-trained high-order abstract features, and contextualized multilingual word representations. Experimental results on the Universal Proposition Bank corpus indicate that performances of the cross-lingual SRL can vary by leveraging different cross-lingual features. In addition, whether the features are gold-standard also has an impact on performances. Precisely, we find that gold syntax features are much more crucial for cross-lingual SRL, compared with the automatically-generated ones. Moreover, universal dependency structure features are able to give the best help, and both pre-trained high-order features and contextualized word representations can further bring significant improvements.