Feature structures, unification and finite-state transducers

  • Authors:
  • Rémi Zajac

  • Affiliations:
  • New Mexico State University

  • Venue:
  • FSMNLP '09 Proceedings of the International Workshop on Finite State Methods in Natural Language Processing
  • Year:
  • 1998

Quantified Score

Hi-index 0.01

Visualization

Abstract

We present a new framework for describing morphological models which combines several types of descriptions in a single unified declarative language: simple morphological rules relate strings to feature structures; rules can be grouped into inheritance hierarchies of paradigms; rules can be composed for describing simple agglutinative morphology; they can also be combined to describe more complex morphotactic structures. Rules are compiled as extended finite-state transducers where left projections are characters and right projections are feature structures: unification is used instead of concatenation to compute the output of a transduction.