Developing New Linguistic Resources and Tools for the Galician Language

Developing New Linguistic Resources and Tools for the Galician Language Rodrigo Agerri *, Xavier G´omez Guinovart **, German Rigau *, Miguel Anxo Solla Portela ** * IXA NLP Group, University of the Basque Country UPV/EHU | ** TALG Research Group, University of Vigo rodrigo.agerri@ehu.eus, xgg@uvigo.es, german.rigau@ehu.eus, miguelsolla@uvigo.es Abstract In this paper we describe the work towards developing new resources and Natural Language Processing (NLP) tools for the Galician language. First, a new corpus, manually revised, for POS tagging and lemmatization is described. Second, we present a new manually annotated corpus for Named Entity tagging for Galician. Third, we train and develop new NLP tools for Galician, including the first publicly available Galician statistical modules for lemmatization and Named Entity Recognition, and new modules for POS tagging, Wikification and Named Entity Disambiguation. Finally, we also present two new Web demo applications to easily test the new set of tools online. Keywords: Galician, Less-resourced languages, Language Resources, Linguistic Tools 1. Introduction We present new developments on linguistic resources and Natural Language Processing tools for the Galician lan- guage. There are previous works addressing the cre- ation of resources to allow the automatic processing of the Galician language, including the Galician WordNet (G´omez Guinovart and Solla Portela, 2017), the Gali- cian SemCor (Solla Portela and G´omez Guinovart, 2017), works on terminology (Solla Portela and G´omez Guino- vart, 2015) or annotation of large corpora (G´omez Guino- vart and L´opez Fern´andez, 2009). Furthermore, there are also publicly available Natural Language Processing tools for the Galician language, notably Freeling (Padr´o and Stanilovsky, 2012) and Linguakit (Gamallo and Garcia, 2017). However, it is remarkable the lack of linguistic processors that are available for other less-resourced lan- guages, such as statistical tools for lemmatization or Named Entity Recognition. In this work we aim to contribute to the development of NLP resources for Galician by providing a new manually revised corpus for POS tagging and lemmatization, and a new manually annotated corpus for Named Entity Recog- nition. This would allow us to train previous unavailable statistical tools for the processing of Galician texts. As a result of our effort, seven new linguistic processors for Galician are presented: a rule-based tokenizer and statis- tical tools for POS tagging, lemmatization, Named Entity Recognition, Named Entity Disambiguation, Wikification and graph-based Word Sense Disambiguation. In particu- lar, this paper presents the first open source statistical lem- matizer and Named Entity tagger for the Galician language. 2. Corpora and Lexical Resources In this paper the main work on corpora and lexical re- sources was undertaken in order to create new resources to train a statistical POS tagger and lemmatizer, and a new Named Entity Recognition and Classification tagger. The basis for the corpora required is the CTG Galician Tech- nical Corpus1 (TALG Research Group, 2016). The CTG 1http://sli.uvigo.gal/CTG/ corpus contains around 18 million words from various do- mains: • GALEX Galician legal texts (∼3M words). • XIGA Galician texts on computing and telecommuni- cations (∼3M words). • AUGA Galician texts on ecology and environmental …

Content truncated to 500 words