Božo Bekavac

Implementation of Croatian NERC System

In this paper a system for Named Entity Recognition and Classification in Croatian language is described. The system is composed of the module for sentence segmentation, inflectional lexicon of common words, inflectional lexicon of names and regular local grammars for automatic recognition of numerical and temporal expressions. After the first step (sentence segmentation), the system attaches to each token its full morphosyntactic description and appropriate lemma and additional tags for potential categories for names without disambiguation. The third step (the core of the system) is the application of a set of rules for recognition and classification of named entities in already annotated texts. Rules based on described strategies (like internal and external evidence) are applied in cascade of transducers in defined order. Although there are other classification systems for NEs, the results of our system are annotated NEs which are following MUC-7 specification. System is applied on informative and noninformative texts and results are compared. F-measure of the system applied on informative texts yields over 90%.

Inflectionally sensitive web search in Croatian using Croatian lemmatization server

Web search engines are becoming everyday commodity but still they are mainly oriented towards English. Other languages, with their different word structures, are not represented well with available Web search engines. We present one solution for a morphologically sensitive search engine which enables the Croatian-speaking users to use query words more naturally and get the results that cover all and only all word-forms of searched words

Identification of Croatian Light Verb Constructions with NooJ

This paper deals with the identification of verbal multiword expressions in Croatian using NooJ. We focus on light verb constructions, i.e. constructions consisting of a light verb and a nominal part (e.g. donijeti odluku ‘to make a decision’ or ‘to reach a decision’). The nominal part in light verb constructions (LVCs) can be either a noun phrase (e.g. dati udarac ‘to give a blow’) or a prepositional phrase (staviti na raspolaganje ‘to put at disposal’). Light verbs are entirely or partially deprived of their lexical meaning and the meaning of the whole construction is conveyed by their complimentary NPs or PPs. All elements in these constructions act as a single unit at some level of linguistic analysis, particularly at the level of semantics and, more recently, of syntax.

Detection of Verb Frames with NooJ

This paper deals with semi-automatic extension of CroDeriV with verb valency frames. CroDeriV is a morphological database of Croatian verbs. In its present shape the database comprises 14 500 verbs in infinitive forms. Each verb in CroDeriV is segmented into lexical and derivational morphemes and verbs of the same root are mutually linked. In order to further enrich the CroDeriV with semantic and syntactic information, we have used the NooJ platform to recognize derivationally related verbs, find the verb frames and to speed up the sentence processing.