Towards a Graded Dictionary of Spanish Collocations

Abstract

[Abstract] Several recent studies have observed that texts of different quality and written by learners at different proficiency levels also vary in the lexical combinations they contain. Such variation can be operationalized by quantitatively measuring the association between the components of these lexical combinations. In particular, pointwise mutual information (MI) has proved to be a good predictor of proficiency development, as several studies on English learners’ writing have shown. This paper examines whether association measures are also a good predictor for the proficiency level of texts written by learners of Spanish, with a view to using such information for grading lexical combinations in order to include them in a collocation dictionary of Spanish. The study also investigates whether the association measures that correlate with learners’ proficiency level can discriminate between phraseological collocations and noncollocations. Our results show that, whereas the MI of learner texts’ lexical combinations is a better predictor of author proficiency than frequency, the latter performs better in identifying phraseological collocations among the whole set of lexical combinations.