This paper describes a method to automatically identify stages of language change in diachronic corpus data, combining variability-based neighbour clustering, which offers objective and reproducible criteria for periodization, and distributional semantics as a flexible and objective representation of lexical meaning. This method partitions the history of a grammatical construction according to qualitative stages of productivity corresponding to different sets of semantic classes attested in one of its lexical slots. Two case studies are presented. The first case study on the “Verb the hell out of NP” construction shows that the semantic development of a construction does not always match that of its quantitative aspects, like token or type frequency. The second case study on the way-construction compares the results of the present method with those of collostructional analysis. While the results overlap to some degree, it is shown that the former measures semantic change with greater precision, both regarding the nature of changes and their chronology. In sum, this method offers a promising exploratory approach to capturing variation in the semantic range of lexical fillers of constructions and to modeling constructional change.
- distributional semantics