AlphaMWE-Arabic: Arabic Edition of Multilingual Parallel Corpora with Multiword Expression Annotations

Proceedings of Recent Advances in Natural Language Processing, pages 448–457 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_050 448 AlphaMWE-Arabic: Arabic Edition of Multilingual Parallel Corpora with Multiword Expression Annotations Najet Hadj Mohamed 1,2∗ Malak Rassem 3∗ Lifeng Han 4 Goran Nenadic 4 1 University of Tours, France 2 Arabic Natural Language Processing Research Group, University of Sfax, Tunisia 3 Institute for Natural Language Processing (IMS), University of Stuttgart, Germany 4 Department of Computer Science, University of Manchester, United Kingdom * co-first authors Abstract Multiword Expressions (MWEs) have been a bottleneck for Natural Language Understanding (NLU) and Natural Lan- guage Generation (NLG) tasks due to their idiomaticity, ambiguity, and non- compositionality. Bilingual parallel cor- pora introducing MWE annotations are very scarce which set another challenge for current Natural Language Process- ing (NLP) systems, especially in a mul- tilingual setting. AlphaMWE-Arabic,This work presents an Arabic edition of the AlphaMWE parallel corpus with MWE annotations. We introduce how we created this corpus including machine translation (MT), post-editing, and anno- tations for both standard and dialectal va- rieties, i.e. Tunisian and Egyptian Ara- bic. We analyse the MT errors when they meet MWEs-related content, both quanti- tatively using the human-in-the-loop met- ric HOPE and qualitatively. We report the current state-of-the-art MT systems are far from reaching human parity performances. We expect our bilingual English-Arabic corpus will be an asset for multilingual research on MWEs such as translation and localisation, as well as for monolin- gual settings including the study of Arabic- specific lexicography and phrasal verbs on MWEs. Our corpus and experimen- tal data are available at https://github. com/aaronlifenghan/AlphaMWE 1 Introduction Multiword Expressions (MWEs), such as “a cheap shot” (a cruel verbal attack) or “take it with a grain of salt” (regard something as exaggerated), are combinations of words that function as a sin- gle unit and have a specific meaning (Baldwin and Kim, 2010), typically regarded as a ‘pain in the neck’ to Natural Language Processing (NLP) tasks, particularly in the field of machine transla- tion (MT) (Sag et al., 2002) and information ex- traction (Kovaˇcevi´c et al., 2013; Maldonado et al., 2017). Translating MWEs accurately poses a sig- nificant challenge for statistical and neural MT systems (Han, 2022b; Han et al., 2021, 2020b). The difficulty lies in the idiomatic, colloquial or culture-specific nature of MWEs, which requires a deep understanding of their meaning, context, and cultural references (Moreau et al., 2018). Ad- ditionally, MWEs can be interpreted into multi- ple possible meanings, further complicating their translation. Therefore, a parallel corpus that in- corporates MWE annotation is expected to be use- ful for improving the MT quality via system fine- tuning and error analysis. However, Arabic seems to lack a satisfactory corpus for such use. The lit- erature describes several English–Arabic parallel corpora. However, to the best of our knowledge, none of these corpora includes the MWEs annota- tion. In this paper, we describe our ongoing effort to extend AlphaMWE coordinated by Han et al. (2020a), a multilingual parallel corpus with anno- tation of MWEs, …

Content truncated to 500 words