Multilingual Irony Detection with Dependency Syntax and Neural Models

TL;DR

This work investigates whether dependency-syntax information, encoded via Universal Dependencies, improves irony detection across four languages (English, Spanish, French, Italian). It compares three experimental settings: (i) classical ML with syntactic features, (ii) UD-based word embeddings within neural models, and (iii) syntactically augmented Multilingual BERT. Across languages, syntactic features—especially Sidorov-inspired dependencies—contribute to better performance, with mixed results for UD-based embeddings depending on data quality. The findings support incorporating morpho-syntax into multilingual irony detection and guide future integration of syntax-pragmatics in deep models, aiming for more robust cross-lingual performance.

Abstract

This paper presents an in-depth investigation of the effectiveness of dependency-based syntactic features on the irony detection task in a multilingual perspective (English, Spanish, French and Italian). It focuses on the contribution from syntactic knowledge, exploiting linguistic resources where syntax is annotated according to the Universal Dependencies scheme. Three distinct experimental settings are provided. In the first, a variety of syntactic dependency-based features combined with classical machine learning classifiers are explored. In the second scenario, two well-known types of word embeddings are trained on parsed data and tested against gold standard datasets. In the third setting, dependency-based syntactic features are combined into the Multilingual BERT architecture. The results suggest that fine-grained dependency-based syntactic information is informative for the detection of irony.

Figures (2)

• Figure 1: The dependency tree in UD format.
• Figure 2: Dependency-based syntactic tree of an English tweet.