Local Grammars and Compound Verb Lemmatization in Serbo - Croatian

The increasing production of electronic (digital) texts (either on the Web or in other electronically available forms, such as digital libraries or archives) demands appropriate computer tools that can help human users in text manipulation and, additionally, in performing automatic processing of language resources. In the first place, a natural language processing (NLP) system needs to implement models for recognition and isolation of various lexical constituents that occur in a digital text. The main purpose is to isolate and mark syntactical units for further analysis. In this paper we will analyze a method for modeling compound verb forms in Serbo-Croatian, and, consequently, an approach to automatic recognition of string sequences in a digital text which represent such forms. NLP processing of a text in a highly inflective language needs to include thorough lexical preprocessing and lemmatization, that must take into account various inflected forms of words. The goal of lemmatization is to determine a lemma for each textual word as well as the appropriate grammatical information that corresponds to it. For example, the process of lemmatization for nouns usually assigns to a textual word its nominative form and corresponding grammatical information (e.g. gender, number, case and other properties). The main objective of the paper is to analyze methods for lemmatization of compound verb forms that occur in a digital text. However, as parts of compound verb forms can be locally distributed within a sentence, the lemmatization has to “bring together” parts of the same verb form and to determine the lemma (i.e. the infinitive form of the verb) and corresponding grammatical information (e.g. gender, person, tense). Unlike the lemmatization of nouns and adjectives, which has been exhaustively studied (cf. [3] and [12]), the references on (compound) verb lemmatization are not numerous, even for the languages that are not highly inflective. As a starting point for our research we have used a framework exposed in [2], dealing with compound forms of verbs in English. The structure of the paper is as follows. First, we will briefly discuss problems of compound verb lemmatization and the resources and tools that we used. Afterwards, we will present a case study of lemmatization of the Preterit Tense in Serbo-Croatian and some other examples of verb phrase lemmatization.