PoS Tagging, Lemmatization and Dependency Parsing of West Frisian

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 4789–4798 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 4789 PoS Tagging, Lemmatization and Dependency Parsing of West Frisian Wilbert Heeringa1, Gosse Bouma2, Martha Hofman1, Jelle Brouwer2, Eduard Drenth1, Jan Wijffels3, Hans Van de Velde1,4 1Fryske Akademy, 2University of Groningen, 3BNOSAC, 4Utrecht University 1Leeuwarden, 2Groningen, 3Brussels, 4Utrecht 1{wheeringa, mhofman, edrenth, hvandevelde}@fryske-akademy.nl, 2{g.bouma, jelle.brouwer}@rug.nl, 3jwijffels@bnosac.be Abstract We present a lemmatizer/PoS tagger/dependency parser for West Frisian using a corpus of 44,714 words in 3,126 sentences that were annotated according to the guidelines of Universal Dependencies version 2. PoS tags were assigned to words by using a Dutch PoS tagger that was applied to a Dutch word-by-word translation, or to sentences of a Dutch parallel text. Best results were obtained when using word-by-word translations that were created by using the previous version of the Frisian translation program Oersetter. Morphologic and syntactic annotations were generated on the basis of a Dutch word-by-word translation as well. The performance of the lemmatizer/tagger/annotator when it was trained using default parameters was compared to the performance that was obtained when using the parameter values that were used for training the LassySmall UD 2.5 corpus. We study the effects of different hyperparameter settings on the accuracy of the annotation pipeline. The Frisian lemmatizer/PoS tagger/dependency parser is released as a web app and as a web service. Keywords: Frisian, lemmatization, PoS tagging, dependency parsing, Universal Dependencies 1. Introduction We present a corpus of lemmatized, tagged and anno- tated text in the West Frisian language together with a web application and web service which can be used for lemmatization, part-of-speech (PoS) tagging and dependency parsing of Frisian text. West Frisian is an autochthonous low-resource language spoken in the Dutch province of Fryslˆan where it is recognized as an official language, in addition to Dutch. In 2018 about 400,000 native speakers were reported (Klinkenberg et al., 2018). 1.1. Choice of methodology A fast way getting a tool that can lemmatize, tag or parse West Frisian text may be the use of an approach that uses transformer-based context-sensitive language models such as BERT (Bidirectional Encoder Repre- sentations from Transformers). BERT was introduced by Google AI Research (Devlin et al., 2019) and is a language representation model pre-trained by jointly conditioning on both left and right context in all lay- ers. As a result, the pre-trained BERT model can be fine-tuned with just one additional output layer on sev- eral NLP tasks. While BERT was trained on English, multilingual BERT (mBERT) has been trained on the 104 languages with the largest Wikipedias, including West Frisian.1 Multilingual BERT was used in the multilingual de- pendency parsing model UDapter that was developed 1See https://github.com/ google-research/bert/blob/master/ multilingual.md. by ¨Ust¨un et al. (2020). With UDadater contextual- ized word representations are obtained. During train- ing adapters were used to capture both task-specific and language-specific information. Adapters are small modules added between layers of a pre-trained net- work. The adapters were enabled …

Content truncated to 500 words