Spotlight: Semantic processing

A module performing annotation at the tectogrammatical layer is run after the syntactic parse is available. Annotation of the meaning of a sentence at tectogrammatical layer is more explicit than its syntactic parse and lends itself for information extraction.

Semantic parsing resolves:

  1. Assignment of one of 69 semantic roles
  2. Coordinations
  3. Argument structure
  4. Partial ellipsis resolution
  5. Pronominal anaphora resolution.

The semantic tree structure is derived from the syntactic parse by set of rules, tree attributes are filled using fnTBL toolkit trained on PDT 2.0 corpora. Post-parsing detection and correction of ungrammatical edges caused by long user utterances is rule based.

The resulting semantic tree is matched against tree fragments using a tree querying engine PMLTQ developed at Charles University. Queries predicates cover topics from Wizard-of-Oz corpora that were implemented in form of DAFs. Extracted information is stored by Dialog Manager in Perl objects.

Contacts

Professor Marc Cavazza (Project Leader)
University of Teesside
School of Computing
Middlesbrough TS1 3BA, UK
Phone: +44 (0) 1642 342657
Email: m.o.cavazza@tees.ac.uk

Dr Debora Field (Project Manager)
University of Sheffield
Department of Computer Science
211 Portobello St
Regent Court
Sheffield S1 4DP, UK
Phone: +44 (0) 114 222 8359
Email: d.field@shef.ac.uk

European Commission • Sixth Framework Programme - Companions Project © 2010. All Rights Reserved | About this website