Lexical Function Detection in Spanish Collocations using Transformer Architecture
Abstract
In this article we study the abilities of transformer models to detect verb-noun lexical functions in Spanish collocations from context. The concept of lexical functions is a formalism to represent recurrent relations among words. A lexical function (LF) takes a word as input and outputs a set of words related to the input in a paradigmatic or syntagmatic way. For example, the syntagmatic LF Oper1 takes the noun decision as input and outputs the verb make with the semantics of ‘Agent realizes the action denoted by the noun’. Oper1 captures the relation between the noun and the verb in many collocations such as make a decision, take a walk, give a lecture, pay a compliment, keep a promise, etc. The numeric part of the Oper1 notation represents that (1) the action of the verb is performed by the agent which is the first argument in the verb’s subcategorization frame, (2) the syntactic function of the noun is subject. In general, lexical functions represent common semantic and syntactic patterns typical for certain word classes and can aid in many natural language processing tasks, especially in word sense disambiguation. In this article we report the results of our experiments with transformer models on the task of detecting verb-noun lexical functions.
Keywords
Collocation, lexical function, syntagmatic relations, transformer models, deep learning