Croatian Error-Annotated Corpus of Non-Professional Written Language

Croatian Error-Annotated Corpus of Non-Professional Written Language Vanja ˇStefanec∗, Nikola Ljubeˇsi´c†, Jelena Kuvaˇc Kraljevi´c‡ ∗ Center for Postgraduate Studies, University of Zagreb Zvonimirova 8, HR-10000 Zagreb, Croatia vanja.stefanec@gmail.com † Dept. of Information and Communication Sciences, University of Zagreb Ivana Luˇci´ca 3, HR-10000 Zagreb, Croatia nljubesi@ffzg.hr ‡ Laboratory for Psycholinguistic Research, Department of Speech and Language Pathology, University of Zagreb Zvonimirova 8, HR-10000 Zagreb, Croatia jkuvac@erf.hr Abstract In the paper authors present the Croatian corpus of non-professional written language. Consisting of two subcorpora, i.e. the clinical subcorpus, consisting of written texts produced by speakers with various types of language disorders, and the healthy speakers subcorpus, as well as by the levels of its annotation, it offers an opportunity for different lines of research. The authors present the corpus structure, describe the sampling methodology, explain the levels of annotation, and give some very basic statistics. On the basis of data from the corpus, existing language technologies for Croatian are adapted in order to be implemented in a platform facilitating text production to speakers with language disorders. In this respect, several analyses of the corpus data and a basic evaluation of the developed technologies are presented. Keywords: error corpus, language disorders, Croatian language 1. Introduction In this paper we describe the Croatian corpus of non- professional written language (Kuvaˇc Kraljevi´c et al., in press) that is based on written language production of two groups of participants: healthy speakers and speakers with various types of language disorders (LDs). This resource, created in an interdisciplinary environment, offers valuable data for research in different fields of science. It provides data about language production which can be of interest to speech and language pathologists working in clinical set- tings, but also to linguists and neuroscientists studying lan- guage processing in general. Besides that, it is a unique lan- guage resource that can be used for improving existing or building new and specialized language technologies (LTs). Roughly containing 0.5M tokens, it is, as to our best knowl- edge, the biggest corpus of this type in general. It is needless to say that resources of this type are scarce. Moreover, this is not the case only in linguistically under- resourced languages like Croatian. Specialized corpora which contain samples of written language produced by speakers with developmental or acquired language disor- ders, i.e. error corpora, are generally rare (with the ex- ception of the learner corpora, which, we believe, capture different types of errors than those targeted here), and es- pecially those entirely composed of samples produced by the speakers with language disorders. But, nevertheless, besides for scientific research of the underlying phenom- ena, such corpora are irreplaceable for developing various language tools, such as contextually-aware spelling correc- tors. In the absence of a better resource, needed corpora are sometimes artificially created by introducing errors into correct text (Pedler, 2007, p. 43). Other commonly used resources (and often wrongly referred to as ‘corpora’) are simply lists of commonly misspelled words, extracted from every context, paired with their correctly written equiva- …

Content truncated to 500 words