WordNet-QU: Development of a Lexical Database for Quechua Varieties

Proceedings of the 29th International Conference on Computational Linguistics, pages 4429–4433 October 12–17, 2022. 4429 WordNet-QU: Development of a Lexical Database for Quechua Varieties Nelsi Melgarejo Pontifical Catholic University of Peru nelsi.melgarejo@pucp.edu.pe Rodolfo Zevallos Pompeu Fabra University rodolfojoel.zevallos@upf.edu Héctor Gómez Pontifical Catholic University of Peru hector.gomez@pucp.edu.pe John E. Ortega Northeastern University j.ortega@northeastern.edu Abstract In the effort to minimize the risk of extinction of a language, linguistic resources are funda- mental. Quechua, a low-resource language from South America, is a language spoken by millions but, despite several efforts in the past, still lacks the resources necessary to build high-performance computational systems. In this article, we present WordNet-QU which signifies the inclusion of Quechua in a well- known lexical database called wordnet. We propose WordNet-QU to be included as an extension to wordnet after demonstrating a manually-curated collection of multiple digi- tal resources for lexical use in Quechua. Our work uses the synset alignment algorithm to compare Quechua to its geographically nearest high-resource language, Spanish. Altogether, we propose a total of 28,582 unique synset IDs divided according to region like so: 20510 for Southern Quechua, 5993 for Central Quechua, 1121 for Northern Quechua, and 958 for Ama- zonian Quechua. 1 Introduction and related work Lexical databases and resources have been used in the past for various natural language process- ing (NLP) tasks ranging from information retrieval (IR) to machine translation (MT). While many re- cent NLP approaches rely on deep learning tech- niques like transformers, namely BERT (Devlin et al., 2018), where attention (Vaswani et al., 2017) is used to create a semantic representation of text, more traditional approaches relied on purely lin- guistic and syntactic features. More often than not, recent deep-learning approaches require a large amount of data to perform better than traditional ones (e.g. on the order of millions of words for ma- chine translation (Koehn and Knowles, 2017; Bah- danau et al., 2014)). This makes NLP approaches with low-resource languages, languages that are measured in the thousands typically, much more difficult to solve with recent approaches thus forc- ing the use of traditional approaches to solve prob- lems. One low-resource language from South Amer- ica, called Quechua, is spoken by nearly 8 million people1 yet still does not have enough resources to effectively compete with other high-resource languages as has been shown in previous research (Ebrahimi et al., 2021; Ortega et al., 2021, 2020). Oftentimes, due to insufficient resources, scores such as BLEU (Papineni et al., 2002) and accu- racy are more than three times lower. This lack of resources thus drives the need for traditional techniques such as the use of lexical databases, grammars, and other linguistic cues such as tree banks and more. One such resource that has been commonly used for traditional approaches is called wordnet (Fellbaum, 1998) which was originally cre- ated in the 1990s yet is still used today, especially for low-resource languages like Quechua. The need to build digital resources is greater for endangered languages like Quechua and others since there is …

Content truncated to 500 words