An Enhanced Mapping Scheme of the Universal Part-Of-Speech for Korean

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 3826–3833 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 3826 An Enhanced Mapping Scheme of the Universal Part-Of-Speech for Korean Maria Myung-Hee Kim and Nathalie Colineau Defence Science and Technology (DST) DST Edinburgh, SA 5111 {maria.kim, nathalie.colineau}@dst.defence.gov.au Abstract When mapping a language specific Part-Of-Speech (POS) tag set to the Universal POS tag set (UPOS), it is critical to consider the individual language’s linguistic features and the UPOS definitions. In this paper, we present an enhanced Sejong POS mapping to the UPOS in accordance with the Korean linguistic typology and the substantive definitions of the UPOS categories. This work updated one third of the Sejong POS mapping to the UPOS. We also introduced a new mapping for the KAIST POS tag set, another widely used Korean POS tag set, to the UPOS. Keywords: Universal Part-of-Speech, Sejong POS, KAIST POS, Korean NER 1. Introduction The Universal Part-Of-Speech tag set (UPOS) (Petrov et al., 2012) has been widely used as a standard for mapping various POS tag sets across multiple languages to facilitate building and evaluating multi-lingual taggers and parsers (McDonald et al., 2013; Rosa et al., 2014; Park, 2017). In 2014, the extended UPOS (UD annotation guidelines v1, 2014) was released as part of the Universal Dependencies (UD) project, which aims to develop cross- language treebank annotation. The recent UD version 2.5 has released 157 treebanks in 90 languages in Nov 20191 . When mapping a language specific POS tag set to the UPOS, it is important to map it correctly based on: (1) the definitions of the UPOS categories; and (2) the individual language’s linguistic topology. It should not be just matching to the equivalent categories’ name2. When the original UPOS (UPOS12) was proposed by Petrov, each category was defined operationally (Petrov et al., 2012). However, there were some ambiguities to the interpretation of these categories. This problem was addressed in the extended UPOS (UPOS14) in 2014 (UD annotation guidelines v1, 2014) 3. Clear definitions and explanatory examples were provided for the 17 categories of the UPOS including five new categories4, which were identified as essential categories to many languages. Following the detailed guidelines, one can build an appropriate mapping scheme and justify their mapping results. The first morpheme-based mapping for Korean POS tag set to the UPOS12 was introduced by Park and colleagues (2016). They proposed a mapping for the Sejong POS tag set, one of the most widely used Korean POS tag set. 1 https://universaldependencies.org/ 2 Note that words called as numerals in some languages (e.g. Czech) should be tagged with the UPOS ADJ (adjective) as recommended by UD annotation guidelines v1, 2014. 3 UD annotation guidelines v2 was released in 2016 (UD annotation guidelines v2). The UPOS from v1 was kept the same in v2 except for two minor revisions (https://universaldependencies.org/v2/summary.html); it includes updates on definitions of AUX and PART tags. 4 PROPN (proper noun), AUX (auxiliary), SCONJ (subordinate conjunction), …

Content truncated to 500 words