EMPAC: an English–Spanish Corpus of Institutional Subtitles

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 392–396 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 392 Quality Focused Approach to a Learner Corpus Development Roberts Darģis1, Ilze Auziņa2, Kristīne Levāne-Petrova3, Inga Kaija4 Institute of Mathematics and Computer Science, University of Latvia1,2,3,4, Riga Stradiņš University4 Raina bulvaris 29, Riga, LV-1459, Latvia1,2,3,4, Dzirciema iela 16, Riga, LV-1007, Latvia4 {roberts.dargis, ilze.auzina, kristine.levane-petrova}@lumii.lv, inga.kaija@rsu.lv Abstract The paper presents quality focused approach to a learner corpus development. The methodology was developed with multiple design considerations put in place to make the annotation process easier and at the same time reduce the amount of mistakes that could be introduced due to inconsistent text correction or carelessness. The approach suggested in this paper consists of multiple parts: comparison of digitized texts by several annotators, text correction, automated morphological analysis, and manual review of annotations. The described approach is used to create Latvian Language Learner corpus (LaVA) which is part of a currently ongoing project Development of Learner corpus of Latvian: methods, tools and applications. Keywords: language acquisition, corpus development, validation 1. Introduction Learner corpora constitute a new resource for second language acquisition and foreign language teaching specialists. They are particularly useful if they are error- tagged. While error-tagging is problematic in many theoretical aspects, it is probably not controversial anymore that learner corpus with consistently annotated errors can be useful for a number of research questions. Unfortunately, there are multiple reasons why mistakes could be introduced in annotations. Different interpretations of the source text lead to different text corrections. Annotation guidelines do not contain all possible scenarios which also may result in correction differences. Besides, mistakes can be introduced due to carelessness, since error annotation is a highly repetitive task which leads to loss of focus. Multiple design principles are discussed in this article to reduce the risk of these mistakes in every corpus development step. The discussed principles are implemented in the corpus platform which is used to develop the error-tagged Learner corpus of Latvian (LaVA).! The LaVA corpus is developed as a part of an ongoing project Development of Learner corpus of Latvian: methods, tools and applications, started in September 2018. The project has several interrelated goals: (1) creation of infrastructure for corpus collection, annotation methodology (both error annotation and morphological annotation); (2) development of an error-tagged Learner Corpus of Latvian (LaVA); and (3) development of corpus- based learning materials and a self-assessment web platform for learners of Latvian. ! Latvian is a language with rich morphology and a relatively free word order. Latvian can be generally considered a phonetic language – a language with a relatively simple relationship between orthography and phonology. From the language acquisition perspective, Latvian has several specific properties: short and long vowels and diphthongs, a high degree of inflection, rather free word order. These properties have to be taken into account in error-annotation.! 2. Related Work A learner corpus is a computerized textual database of the language produced …

Content truncated to 500 words