Sara Librenjak

Recognizing Verb-Based Croatian Idiomatic MWUs

This paper tackles the computational problems of Croatian verbal idioms. Croatian language has very rich phraseme structure, as described in Matesic (1982), Menac (2007) and Menac-Mihalic (2007), as well as many others. This work is one of the few attempts of computational analyis of idioms in Croatian language as multi-word units. We used rule-based approach and NooJ syntactic grammars in order to recognize any verb based idiom (of the ~1500 analyzed) in any syntactic position. The Croatian Dictionary of Idioms (Menac et al. 2003) was used for the initial list, which was implemented with new additions during training phase. Grammars were tested within the corpora constructed specifically for this work, and used to calculate statistical measures of recall, precision and f-measure for our grammars. With the final results of recall < 98 %, precision < 96 % and f-measure < 97 %, we consider this a successful attempt in the recognition of verb based idioms in Croatian language.

Recognizing Diminutive and Augmentative Croatian Nouns

In this paper, the authors present NooJ morphological grammars for recognizing Croatian diminutive and augmentative nouns for those common nouns that already exist in the Croatian NooJ dictionary. The purpose of this project is twofold. The first one is to recognize both diminutive and augmentative forms of each noun existing in our dictionary (over 20 000 common nouns) if such a form occurs in a text. The second purpose is to determine types of texts in which these words appear the most (or if they even appear) which is the reason why we divided our corpus in two thematic categories (children literature, novels). The results of our algorithm are high on both types of text [overall P = 0.82; R = 0.80; f-measure = 0.81]. Although NooJ dictionary allows direct entrance of such derivations as an attribute-value description of a main noun, we have opted for the second option, i.e. writing a morphological grammar that will recognize the needed form. In this way, we are saving the space and time needed to add all the existing forms to the noun’s dictionary.