Annotating Verbal Multiword Expressions in Arabic: Assessing the Validity of a Multilingual Annotation Procedure

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 1839–1848 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 1839 Annotating Verbal Multiword Expressions in Arabic: Assessing the Validity of a Multilingual Annotation Procedure Najet Hadj Mohamed1,2, Cherifa Ben Khelil1, Agata Savary3, Iskandar Keskes2, Jean-Yves Antoine1, Lamia Belguith Hadrich2 1 University of Tours, LIFAT, ICVL; 2 University of Sfax, MIRACL; 3 University of Paris-Saclay, LISN Abstract This paper describes our efforts to extend the PARSEME framework to Modern Standard Arabic. The applicability of the PARSEME guidelines was tested by measuring the inter-annotator agreement in the early annotation stage. A subset of 1,062 sentences from the Prague Arabic Dependency Treebank PADT was selected and annotated by two Arabic native speakers independently. Following their annotations, a new Arabic corpus with over 1,250 annotated VMWEs has been built. This corpus already exceeds the smallest corpora of the PARSEME suite, and enables first observations. We discuss our annotation guide- line schema that shows full MWE annotation is realizable in Arabic where we get good inter-annotator agreement. Keywords: Arabic multiword expressions, PARSEME, annotation guidelines 1. Introduction The importance of multi word expressions (MWEs), such as by and large ‘generally’, a cheap shot ‘a cruel verbal attack’, or to eat dirt ‘to have to accept bad treatment’, has long been recognized in Natural Language Process- ing (NLP) (Baldwin and Kim, 2010; Constant et al., 2017). Their idiosyncratic (i.e. special and peculiar) behavior calls for language resources in which they are explicitly identified and described. Moreover, in order to enable cross-language stud- ies of idiosyncrasy, MWE modelling should ide- ally be performed in a unified framework. The PARSEME community has undertaken such an effort of setting up unified annotation guidelines for verbal MWEs (VMWEs) in many languages (Savary et al., 2018; Ramisch et al., 2018; Ramisch et al., 2020). The principle of this framework is to represent in a unified way only those phe- nomena which are truly similar, thus emphasizing those which are specific to particular languages. So far, twenty-five national teams have prepared cor- pora in their languages, annotated manually for VMWEs according to the unified guidelines, and released under open licenses. This boosted the de- velopment of MWE-aware NLP tools, most promi- nently VMWE identifiers. Each time a new language joins PARSEME, the guidelines are tested for their applicability to this language, and modified or extended if needed. This paper describes such an ongoing effort of extending PARSEME framework to Modern Standard Arabic (MSA), henceforth called Arabic for short. The paper is organized as follows: Section 2 describes previous works dedicated to Arabic MWEs, Section 3 is an introduction to the Ara- bic language from the NLP perspective, Section 4 describes linguistic properties of Arabic MWEs, Section 5 explains the construction of the Ara- bic VMWE-annotated corpus, including the val- idation and adaptation of the annotation guide- lines, as well as some specific features of Arabic relevant to the annotation process. In Section 6 we report …

Content truncated to 500 words