Cross-Lingual Transfer of Semantic Roles: From Raw Text to Semantic Roles

TL;DR

This paper presents a transfer method for semantic role labeling across languages using only raw text and parallel data, leveraging deep character-based models to outperform supervised methods in most tested languages.

Contribution

Introduces a novel annotation projection approach for semantic role labeling that relies solely on word and character features, eliminating the need for supervised syntactic annotations.

Findings

Outperforms state-of-the-art supervised methods on 6 of 7 languages

Uses character-based and unsupervised stem embeddings effectively

Achieves cross-lingual transfer without supervised linguistic features

Abstract

We describe a transfer method based on annotation projection to develop a dependency-based semantic role labeling system for languages for which no supervised linguistic information other than parallel data is available. Unlike previous work that presumes the availability of supervised features such as lemmas, part-of-speech tags, and dependency parse trees, we only make use of word and character features. Our deep model considers using character-based representations as well as unsupervised stem embeddings to alleviate the need for supervised features. Our experiments outperform a state-of-the-art method that uses supervised lexico-syntactic features on 6 out of 7 languages in the Universal Proposition Bank.