In recent years, researchers investigating changes in word meaning have employed computational methods to identify instances of change and to trace patterns of semantic shifts over time (Hamilton et al. 2016, Rodda et al.2017; see Tang 2018, Tahmasebi et al. 2018 for overviews). This vein of research relies on principles of distributional semantics. Distributional methods, based on the seminal work of Harris (1954), rely on the assumption that semantic similarity among words is measured by similarity in contexts of use.
In this talk I will present findings from some of my recent work on semantic change in Spanish and Portuguese using distributional methods,specifically word embeddings. I will present results from a study on the diachrony of algo in Spanish (Amaral et al.Forthcoming) as well as on the grammaticalization of adversative connectives in Portuguese (Amaral et al. 2022). I will also discuss how these methods compare to other more established methodologies in the field (e.g.variationist studies of language change) and their potential contributions to historical linguistics.
Given the novelty of word embedding models in historical research, assessing the validity of these methods, especially given the reduced corpora available for older stages of the languages, has been one issue of concern. I will mention some recent work addressing the evaluation of word embedding models and the results obtained so far for Portuguese and Spanish data.
Register in adavance for this meeting via the "Register now" link in the right hand corner, after registering, you will receive a confirmation email containing information about joining the meeting.