Seminarium i datorlingvistik

  • Datum: –14.30
  • Plats: Engelska parken 9-3042
  • Föreläsare: Joakim Nivre
  • Kontaktperson: Marie Dubremetz
  • Telefon: 0046184711414
  • Seminarium

Universal Dependencies -- Dubious Linguistics and Crappy Parsing?

Universal Dependencies is a framework for cross-linguistically consistent treebank annotation that has so far been applied to over 50 languages. It was developed primarily to support multilingual parsing research, but the resources have proven useful for a wide range of studies that were not foreseen originally, including research on language typology. A basic design principle in Universal Dependencies is to give priority to grammatical relations between content words, which are more likely to be parallel across languages, and to treat function words essentially as features of content words. This principle has been criticized both for being incompatible with theoretical linguistics, which tend to treat function words as syntactic heads, and for being suboptimal as a representation for dependency parsing, where higher accuracy is often observed with function words as heads. I will argue that both of these criticisms rest on a misinterpretation of the syntactic representations, and I will show that an alternative interpretation is compatible with both sound linguistics and improved parsing technology.