Predicate-argument analysis to build a phraseology module and to increase conceptual relation expressiveness

Abstract

EcoLexicon, a multilingual and multimodal terminological knowledge base (TKB) on the environment, needs improvements: more expressive non-hierarchical relations and a phraseology module consistent with knowledge representation in the other modules of the TKB. Both issues must be addressed by analyzing predicate-argument structure in text. In this paper, we explain our methodology for predicate-argument analysis with the case study on the conceptual relation affects. We take a semi-automatic approach to extract term-verb-term collocates with Sketch Engine [1]. Then the verbs are classified according to the lexical domains proposed by Faber & Mairal [2] and the arguments in conceptual categories based on the knowledge contained in EcoLexicon. To validate the lexical domains and conceptual categories, an automatic clustering method based on word2vec [3] is applied. The analysis of verbs and arguments contributes to the refinement of our semantic relations and categories as well as to the population of the phraseological module.