gApp: a text preprocessing system to improve the neural machine translation of discontinuous multiword expressions

Abstract

In this paper we present research results with gApp, a text-preprocessing system designed for automati-cally detecting and converting discontinuous multiword expressions (MWEs) into their continuous forms so as to improve the performance of current neural machine translation systems (NMT) (see Hidalgo-Ternero, 2021 & 2022, Hidalgo-Ternero & Corpas Pastor, 2020, 2022a & 2022b, Hidalgo-Ternero, Lista, and Corpas Pastor, 2022, and Hidalgo-Ternero and Zhou-Lian, 2022a & 2022b). To test its effectiveness, eight experiments with several NMT systems such as DeepL, Google Translate, ModernMT and VIP have been carried out in different language directionalities (ES/FR/IT > ES/EN/DE/FR/IT/PT/ZH) for the trans-lation of somatisms, i.e., MWEs containing lexemes referring to human or animal body parts (Mellado Blanco, 2004). More specifically, we have analysed both flexible verb-noun idiomatic constructions (VNICs) and flexible verb + prepositional phrase (VPP) constructions. In this regard, the promising results obtained for these typologies of MWEs throughout experiments 1-8 will shed some light on new avenues for enhancing MWE-aware NMT systems.