Semantic relations between collocations: A Spanish case study

Abstract

Linguistics as a scientific study of human language intends to describe and explain it. However, validity of a linguistic theory is difficult to prove due to volatile nature of language as a human convention and impossibility to cover all real-life linguistic data. In spite of these problems, computational techniques and modeling can provide evidence to verify or falsify linguistic theories. As a case study, we conducted a series of computer experiments on a corpus of Spanish verb-noun collocations using machine learning methods, in order to test a linguistic point that collocations in the language do not form an unstructured collection but are language items related via what we call collocational isomorphism, represented by lexical functions of the Meaning-Text Theory. Our experiments allowed us to verify this linguistic statement. Moreover, they suggested that semantic considerations are more important in the definition of the notion of collocation than statistical ones. 45Resumen: La Lingüística, siendo el estudio científico del lenguaje humano, intenta describirlo y explicarlo. Sin embargo, es difícil demostrar la certeza de cualquier teoría lingüística por la naturaleza versátil del lenguaje como una convención humana y también por la imposibilidad de investigar todo lo que se habla y se escribe en la vida real. A pesar de estos problemas, las técnicas y los modelos computacionales pueden proporcionar la evidencia para que las teorías lingüísticas sean comprobadas o refutadas. A través de un estudio de caso, realizamos una serie de experimentos por computador en un corpus de colocaciones de verbo-sustantivo en español usando métodos de aprendizaje de máquina, con el fin de comprobar el hecho lingüístico de que las colocaciones del idioma no conforman un grupo sin estructura sino que son unidades lingüísticas relacionadas por medio de lo que denominamos isomorfismo de colocaciones, representado por las funciones léxicas de la Teoría Significado-Texto. Nuestros experimentos nos permiten verificar esa declaración lingüística. Asimismo, los experimentos sugieren que las consideraciones semánticas son más importantes en la definición de colocaciones que las estadísticas.Palabras Clave: Colocaciones, funciones léxicas, aprendizaje de máquina.

Towards Distributional Semantics-Based Classification of Collocations for Collocation Dictionaries

“…Given that these are figures achieved on emotion vocabulary, as in our experiments, they are directly comparable to the figures in our experiments, and we can thus conclude that it pays off to draw upon the contexts of collocations. Kolesnikova and Gelbukh (2012) achieve with their rule-based models an average F-score of 0.75 when classifying over the LF typology. Furthermore, in contrast to our experiments, this average does not reflect the classification of collocations with one single model, but, rather, the average of the best scores achieved with one of the twelve models from a series of models they use for each individual LF.…”

“…However, automation of semantic grouping (or classification) of the relations between elements of the extracted collocations has hardly been paid attention to so far in computational lexicography. Only a few proposals address the problem (Moreno et al 2013, Wanner 2004, Wanner et al 2006, Chung-Chi et al 2009, Kolesnikova and Gelbukh 2012. All of them, except Moreno et al (2013), who also use contextual features, draw exclusively upon WordNet (Fellbaum 1999) or EuroWordNet (Vossen 1998) features of the collocation elements, which limits their application to languages with large coverage WordNets.…”

“…Furthermore, they classify either with respect to the rather finegrained semantic typology of lexical functions (LFs) as defined in the Explanatory Combinatorial Lexicology (Mel'cˇuk 1996) or with respect to generic semantic categories like 'goodness', 'heaviness', 'measures', etc. See, for instance, (Wanner 2004, Wanner et al 2006, Kolesnikova and Gelbukh 2012 for the first and (Chung-Chi et al 2009) for the second. Moreno et al (2013) classify with respect to both LFs and generic semantic categories.…”

“…This observation led Mel'cˇuk define lexical functions (LFs); see (Mel'cˇuk 1996): each LF captures one type of collocation relation. As mentioned in Section 1, LFs have been used as underlying typology for automatic collocation classification by Wanner (2004) and Kolesnikova and Gelbukh (2012), among others.…”

Natutal language processing: Perspective of CIC-IPN

“…Namely, using the concept of lexical function, relationships between collocations can be established, a kind of proportion violent : storm = deep : silence [35]. Such proportions can be learnt automatically from examples, and used to predict the meaning of word combinations not present in the dictionary [36], for example, deep stillness.…”