Samhällsvetenskapliga fakulteten

Seminarium i datorlingvistik

  • Datum: –15.00
  • Plats: Engelska parken 9-3042
  • Föreläsare: Jussi Karlgren
  • Arrangör: Joakim Nivre
  • Kontaktperson: Joakim Nivre
  • Seminarium

Utterance Spaces --- how to represent lexical items, constructions, and
contextual data in a unified vector space

High-dimensional semantic spaces have proven useful and effective for aggregating and processing lexical information for many language processing tasks which are increasingly data-oriented. Mostly, what is represented in such a space are lexical items and their occurrence contexts in distributional vectors or word embeddings. This talk shows how other linguistic items, such as constructions or contextual variables can be represented concurrently with lexical information in an integrated vector space. This makes possible more informed hypothesis testing using symbolic and more sophisticated feature sets used as an input to processing models such as neural models which expect continuous representations as an input.