DigitaL - Linguistic analysis with word embeddings - methods and examples

This webinar is part of the Linguistics reimagined Talk Series.

This event will be taking place in room CHBO 316 at University of Malta Msida Campus and online via (Zoom). The meeting will be recorded. For in-person participation, please note that this talk will be streamed from the room, with the speaker presenting online.

Speaker: Prof. Harald Baayen (Eberhard Karls Universität Tübingen)

Abstract

Word embeddings are high-dimensional numeric representations of word meaning, derived from large corpora. Prof. Baayen will first introduce the core concepts underlying embeddings. Prof. Baayen will then show, by means of a series of examples, how one can ‘data mine’ embeddings for linguistic analysis using some basic but surprisingly powerful statistical tools.

One set of examples will address the semantics of number in inflectional morphology. Prof. Baayen will show that the change from singular to plural in semantic space can depend on semantic class (English), case (Russian, Finnish), or on whether a plural is broken or sound (Maltese). A second set of examples will discuss recent results indicating that embeddings can be surprisingly predictive for the fine phonetic detail with which words are articulated.