Cross-lingual Metaphor Paraphrase Detection – Experimental Corpus and Baselines

Abstract

Correct understanding to metaphors is an integral part of natural language understanding. It requires, among other issues, the ability to decide whether a given pair of sentences – such that the first one contains a metaphor – form a paraphrase pair. Although this decision task is formally analogous to a “traditional paraphrase detection” task, it requires a different approach. Recently, a first monolingual corpus (in English) for metaphor paraphrasing was released – together with several baselines. In this work we are going to shift this task to a cross-lingual level: we state a task of cross-lingual metaphor paraphrase detection, introduce a corresponding experimental cross-lingual corpus (English-Czech) and present several approaches to this problem and set the baselines to this challenging problem. This cross-lingual approach may allow us to deal with tasks like multi-document summarization involving texts in different languages as well as enable us to improve information retrieval tools.