Recent Advances in Development of a Lexicon-Grammar of Polish: PolNet 3.0

Recent Advances in Development of a Lexicon-Grammar of Polish: PolNet 3.0 Zygmunt Vetulani, Grażyna Vetulani, Bartłomiej Kochanowski Adam Mickiewicz University in Poznań ul. Wieniawskiego 1, 61-712 Poznań, Poland E-mail: vetulani@amu.edu.pl, gravet,@amu.edu.pl, bkochanowski@amu.edu.pl Abstract In this paper we present recent works contributing to transformation of the initial PolNet, a Polish wordnet developed at the Adam Mickiewicz University, into a Lexicon Grammar of Polish. We focus on granularity issues that occurred at the stage of including verb- noun collocations as well as information related to language registers. Keywords: wordnet, synonimy, granularity, valency structure, lexicon grammar, collocations, registers 1. Introduction. In the mid-1980s, G. Miller started the development of a novel approach to systematize the semantic description of words (Miller, 1985). The leading idea was to organize a lexicon as a lexical database (Princeton WordNet / PWN) consisting of a hierarchical network of classes of synonyms. PWN appeared useful in AI applications involving NLP. Some PWN followers decided to go one step further and enrich the word descriptions with complex data structures to represent events, relations and situations. The forerunners of this idea (Gross (1994) and Polański (1992)) considered the elementary sentence as a “minimal unit of sense” and the sense of a word as determined by the minimal sentences containing this word. This led to the concept of lexicon-grammar1 where grammatical information sufficient to describe elementary sentences is contained in the lexical entries and where the elementary sentence is the basic unit of meaning. Their contributions preceded the future works within the FrameNet (Fillmore et al., 2002) and VerbNet (Palmer, 2009) projects. The first one referred to Frame Semantics developed by Fillmore. Frames describe lexical units (typically verbs) and their syntactic dependents characterized by frame elements. In the VerbNet project, verbs are grouped according to shared meaning and similar syntactic behavior. Palmer used thematic roles, selectional restrictions on the arguments, and frames containing syntactic description of the verb.2 2. The initial PolNet: inspiration and methodology We started PolNet in 2006 intending to build a lexical ontology as a wordnet similar to PWN. Initially PolNet was implemented for nouns. We decided to compile PolNet 1 First developed for French (since the early 1970s until late 1990s; Gross 1994). This idea was already implemented in our first implementations of the NL interfaces for Polish (Vetulani 1988). Several large scale projects have been recently launched in the area of valency dictionaries both for simple and compound verbs (Vetulani G. 2000, 2012), (Przepiórkowski et al. 2014). 2 http://verbs.colorado.edu/~mpalmer/projects/verbnet.html 3 For merge/expand models see (Vossen, 2002), p. 52, section 3.1. from scratch (merge development model3) in the way inspired by the PWN and EuroWordNet projects (Vossen 2002). This method guarantees (quasi)one-to-one correspondence between the structure of synsets and the conceptualization shared by a (quasi)totality of users of the language concerned4. This is an important quality factor often underestimated by the wordnet designers applying the less costly expand model. The first milestone (2009) of the PolNet project was reached on attaining over 10K synsets5 for some 10K words …

Content truncated to 500 words