Konbitzul: an MWE-specific database for Spanish-Basque

Konbitzul: an MWE-specific database for Spanish-Basque Uxoa I˜nurrieta, *Itziar Aduriz, Arantza D´ıaz de Ilarraza, Gorka Labaka, Kepa Sarasola IXA NLP group, University of the Basque Country *University of Barcelona usoa.inurrieta@ehu.eus, itziar.aduriz@ub.edu, {a.diazdeilarraza, gorka.labaka, kepa.sarasola}@ehu.eus Abstract This paper presents Konbitzul, an online database of verb+noun MWEs in Spanish and Basque. It collects a list of MWEs with their translations, as well as linguistic information which is NLP-applicable: it helps to identify occurrences of MWEs in multiple morphosyn- tactic variants, and it is also useful for improving translation quality in rule-based MT. In addition to this, its user-friendly interface makes it possible to simply search for MWEs along with translations, just as in any bilingual phraseological dictionary. 1. Introduction Multiword Expressions (MWEs), also called Phraseologi- cal Units (PUs), are combinations of words which together express a single meaning (Sag et al., 2002). They often have irregular lexical-semantic and/or morphosyntactic fea- tures, and they are not always translated word-for-word (Examples 1-3). This means they cause challenges in vari- ous disciplines, such as Lexicography, Translation and Nat- ural Language Processing (NLP). (1) EN: pull sb’s leg ES: tomar el pelo (a) lit. take sb’s hair EU: adarra jo lit. play the horn (to sb) (2) EN: take steps ES: dar pasos lit. give steps EU: urratsak egin lit. steps do (3) EN: take off ES: alzar el vuelo lit. raise the flight EU: aireratu lit. go-to-the-air Although interest in Phraseology has a longer history, stud- ies on MWEs have multiplied considerably over the last two decades (Baldwin and Kim, 2010; Savary et al., 2015). Most of the work undertaken within the field of NLP fo- cuses on MWE candidate extraction (Ramisch, 2015) – mainly for lexicographic purposes– or identification of MWE occurrences in corpora (Savary et al., 2017). How- ever, some research has also been conducted into improving Machine Translation (MT) quality by enhancing MWE pro- cessing (Kordoni and Simova, 2014; Seretan, 2014). Mean- while, a considerable amount of resources have been cre- ated for several languages, including MWE lists, lexicons and MWE-annotated treebanks (Losnegaard et al., 2016). Concerning Basque phraseology, research has been done both to describe some linguistic phenomena and to de- velop NLP tools (Alegria et al., 2004; Gurrutxaga and Ale- gria, 2012), but researchers have had an almost exclusively monolingual perspective. Thus, our aim is, on the one hand, to analyse how MWEs are translated, and, on the other hand, to propose a method to improve their computational treatment in bilingual tools. In this paper, we will present Konbitzul, a database of verb+noun MWEs in Spanish and Basque. As well as working as a bilingual phraseological dictionary, the database contains linguistic information which is useful for NLP-related tasks, notably for Parsing and MT. We will start by introducing the database, including: the verb+noun MWEs collected (Section 2.1.), how linguistic information is included in the database (Section 2.2.), and how the interface is structured (Section 2.3.). We will then go on to explain what the database can be used for: as …

Content truncated to 500 words