Generation of a Spanish Artificial Collocation Error Corpus

Generation of a Spanish Artificial Collocation Error Corpus Sara Rodr´ıguez-Fern´andez1, Roberto Carlini1, Leo Wanner1,2 1NLP Group, Department of Information and Communication Technologies, Pompeu Fabra University C/ Roc Boronat, 138, 08018 Barcelona (Spain) 2Catalan Institute for Research and Advanced Studies (ICREA) sara.rodriguez.fernandez|roberto.carlini|leo.wanner@upf.edu Abstract Collocations such as heavy rain or make [a] decision are combinations of two elements where one (the base) is freely chosen, while the choice of the other (collocate) is restricted by the base. Research has consistently shown that collocations present difficulties even to the most advanced language learners, so that computational tools aimed at supporting them in the process of language learning can be of great value. However, in contrast to grammatical error detection and correction, collocation error marking and correction has not yet received the attention it deserves. This is unsurprising, considering the lack of existing collocation resources, in particular those that capture the different types of collocation errors, and the high cost of a manual creation of such resoures. In this paper, we present an algorithm for the automatic generation of an artificial collocation error corpus of American English learners of Spanish that includes 17 different types of collocation errors and that can be used for automatic detection and classification of collocation errors in the writings of Spanish language learners. Keywords: artificial corpus, collocations, collocation errors, second language learning, computer assisted language learning 1. Introduction Collocations, i.e., idiosyncratic word co-occurrences such as ask [a] question, commit [a] murder, surmount [an] ob- stacle, faint suspicion, high expectation, etc. are known to be one of the great challenges for language learners;1 see, among others, Granger (1998), Lewis and Conzett (2000), Nesselhauf (2005) and Lesniewska (2006). Ac- cording to Wible et al. (2003), “miscollocations” are the most frequent errors in the writings of students. Orol and Alonso Ramos (2013)’s study shows that the “collocation density” in learner corpora is nearly the same as in native corpora, while the collocation error rate in learner corpora is nearly 30% higher than in native corpora. Despite these palpable figures, collocation error identification and correc- tion has not yet received in Computer Assisted Language Learning (CALL) the attention it deserves. Current col- location checkers focus mainly on collocation validation or identification of miscollocations (usually using mutual information- or distribution-based metrics) in the writings of learners and a display of lists of possible corrections, or- dered in terms of the strength of their “collocationality” or similarity to the original miscollocation; see, e.g., (Chang et al., 2008; Liu et al., 2009; Wu et al., 2010; Ferraro et al., 2014). However, this is by far not sufficient. Ideally, learners should be given the same kind of feedback as given by language instructors when they mark students’ essays: they use, as a rule, error type-specific symbols or acronyms during marking (e.g., ‘SV’ for “subject verb agreement”, ‘WO’ for “wrong word order”, ‘WW’ for “wrong word”, etc.); see, e.g., (Nott, 2008). In other words, they classify the students’ mistakes. 1In a collocation, one of the elements (the base) …

Content truncated to 500 words