A Taxonomy of Spanish Nouns, a Statistical Algorithm to Generate it and its Implementation in Open Source Code

A Taxonomy of Spanish Nouns, a Statistical Algorithm to Generate it and its Implementation in Open Source Code Rogelio Nazar, Irene Renau Instituto de Literatura y Ciencias del Lenguaje Pontificia Universidad Cat´olica de Valpara´ıso Av/El Bosque, 1290, Vi˜na del Mar (Chile) {rogelio.nazar,irene.renau}@pucv.cl Abstract In this paper we describe our work in progress in the automatic development of a taxonomy of Spanish nouns, we offer the Perl im- plementation we have so far, and we discuss the different problems that still need to be addressed. We designed a statistically-based taxonomy induction algorithm consisting of a combination of different strategies not involving explicit linguistic knowledge. Being all quantitative, the strategies we present are however of different nature. Some of them are based on the computation of distributional similarity coefficients which identify pairs of sibling words or co-hyponyms, while others are based on asymmetric co-occurrence and identify pairs of parent-child words or hypernym-hyponym relations. A decision making process is then applied to combine the results of the previous steps, and finally connect lexical units to a basic structure containing the most general categories of the language. We evaluate the quality of the taxonomy both manually and also using Spanish Wordnet as a gold-standard. We estimate an average of 89.07% precision and 25.49% recall considering only the results which the algorithm presents with high degree of certainty, or 77.86% precision and 33.72% recall considering all results. Keywords: corpus statistics, distributional semantics, Spanish, taxonomy induction 1. Introduction Influenced by classical logic (Aristotle, on the Categor- ies), modern linguistics since at least structuralist semantics defines a taxonomy of the nouns of a language as the spe- cification of the semantic hierarchy of words in a tree- shape structure, where each link between words establishes their hyponym-hypernym relationships (Lyons, 1977), of- ten called IS-A relations in NLP literature. A number of properties of such type of structures are well- known, such as the concept of inheritance, i.e., the fact that the “children” nodes inherit the properties of their “par- ent” nodes. From a lexical point of view, words in chil- dren node positions are hyponyms, and words playing the role of parent nodes are hypernyms. Other common prop- erties of taxonomies are asymmetry and transitivity. In some taxonomy X, asymmetry is simply defined as fol- lows: ∀a, b ∈ X, (a → b) ⇒ ¬(b → a), i.e. a can- not simultaneously be the parent and the child of b. And transitivity is analogous to the concept of inheritance, thus ∀a, b, c ∈ X, if (a → b) ∧ (b → c) ⇒ (a → c). This means that taxonomies are directed acyclic graphs, and that it would be an infringement for a taxonomy to contain cycles. This does not imply, however, that the same word cannot have different hypernymy chains, which is a differ- ent issue. It would be the case, for instance, of polysemous words, as will be shown later in the paper. Taxonomies can be a useful resource in a number of dif- …

Content truncated to 500 words