An Information-Theoretical Approach to Studying Phoneme Collocational Constraints.

Abstract

Phonologists have described the permissible combination of phonemes in the form of phonotactic rules and have shown that these rules often can be expressed in terms of phoneme equivalence classes. But the phoneme space can be partitioned in many other ways. It is conceivable that, by allowing phonemes to form various sets of equivalence classes and quantifying the constraining power for each set, we may discover phoneme classes that will provide the strongest constraints for lexical access. We investigated phoneme collocational constraints using a normalized measure of mutual information. A pair-wise, hierarchical clustering technique is used to combine phonemes into classes using this metric. Results of this clustering procedure can be displayed as a dendrogram, from which an arbitrary number of equivalence classes can be selected. We investigated collocation constraints of phoneme pairs and triplets and found that in many cases phonemes are organized into classes that share certain phonological features. Phonemes that have similar acoustic properties often exhibit similar collocation constraints. We also compared the constraining power of our phoneme classes with those chosen with a phonological criterion and found ours to be more than competitive. We conclude that our information theoretic metric is well suited to a description of lexical constraining power and discuss the implications for automatic speech recognition.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jul 01, 1991
Accession Number
ADA239141

Entities

People

  • Robert H. Kassel

Organizations

  • Massachusetts Institute of Technology

Tags

Communities of Interest

  • Autonomy
  • Energy and Power Technologies
  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Acoustic Detectors
  • Algorithms
  • Artificial Intelligence
  • Automated Speech Recognition
  • Computational Complexity
  • Computer Science
  • Computers
  • Electrical Engineering
  • Engineering
  • Information Theory
  • Language
  • Larynx
  • Pattern Recognition
  • Probability
  • Recognition
  • Syllables
  • Words (Language)

Readers

  • Computational Linguistics
  • Graph Algorithms and Convex Optimization.
  • Theoretical Analysis.

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms
  • Space