Automatic Detection of Gender and Number Agreement Errors in Spanish Texts Written by Japanese Learners

Copyright 2012 by Maria del Pilar Valverde Ibanez and Akira Ohtani 26th Pacific Asia Conference on Language,Information and Computation pages 299–307 Automatic Detection of Gender and Number Agreement Errors in Spanish Texts Written by Japanese Learners Maria del Pilar Valverde Iba˜nez Faculty of Foreign Languages Aichi Prefectural University 1522-3 Ibaragasama, Nagakute-shi Aichi, 480-1198, Japan valverde@for.aichi-pu.ac.jp Akira Ohtani Faculty of Informatics Osaka Gakuin University 2-36-1 Kishibe-minami, Suita-shi Osaka, 564-8511, Japan ohtani@ogu.ac.jp Abstract This paper describes the creation of a gram- mar to automatically detect agreement errors (gender and number) in Spanish texts written by Japanese learners. The grammar has been written using the Constraint Grammar formal- ism (Karlsson et al., 1995), and uses as input the morphosyntactic analysis provided by the Spanish parser HISPAL (Bick, 2006). For de- veloping and testing the grammar, a learner corpus of 25,000 words has been manually an- notated with agreement error tags. Both the grammar and the data from the corpus serve us to draw some conclusions about the character- istics of agreement errors in Japanese learners’ Spanish. 1 Introduction In this paper we describe the creation of a grammar to automatically detect agreement errors -in gender and number- in Spanish texts written by Japanese learners. Automatic detection of grammatical learner er- rors can be used for the automatic annotation of learner corpora and for the creation of intelligent computer-assisted language learning systems (Heift and Schulze, 2007). Such tools can benefit both teachers -who will be able to study learner er- rors and the language acquisition process more systematically- and learners -who can foster their language learning with the help of automatic tools and improved traditional language materials-. There are two reasons why we focus on agree- ment errors. First, for Japanese students, agreement is a problematic aspect for learning Spanish and in- deed agreement errors are significantly more fre- quent among Japanese learners than among speak- ers of other languages (Fern´andez, 1997). Second, agreement errors in texts can be identified and cor- rected straightforwardly by a native speaker, unlike other type of errors like article and preposition us- age, for example, where annotator agreement may be problematic. While there is a substantial research on detect- ing grammatical errors in Learners’ English, Span- ish has received little attention, probably because of the lack of freely available large learner cor- pora (Lozano, 2009). For the construction of the grammar, we have manually annotated with agree- ment error tags a fragment of 25,000 words from the CORANE learner corpus (Mancera et al., 2001) and to control false positives of the grammar, we have also used native corpora: 22,000 words for develop- ment amd 12,000 words for test. The paper is organized as follows. Section 2 deals with the characteristics of gender and number agree- ment in Spanish and the coverage of the grammar, section 3 deals with the development phase (the cor- pus, grammar formalism and design principles), sec- tion 4 gives the results and analysis of the evaluation, section 5 studies the data in the learner …

Content truncated to 500 words