Detecting Structural Metadata with Decision Trees and Transformation-Based Learning

Abstract

The regular occurrence of disfluencies is a distinguishing characteristic of spontaneous speech. Detecting and removing such disfluencies can substantially improve the usefulness of spontaneous speech transcripts. This paper presents a system that detects various types of disfluences and other structural information with cues obtained from lexical and prosodic information sources. Specifically, combinations of decision trees and language models are used to predict sentence ends and interruption points and given these events transformation based learning is used to detect edit disfluencies and conversational fillers. Results are reported on human and automatic transcripts of conversational telephone speech.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 2004
Accession Number
ADA457891

Entities

People

  • Joungbum Kim
  • Mari Ostendorf
  • Sarah E. Schwarm

Organizations

  • University of Washington

Tags

Communities of Interest

  • Human Systems

DTIC Thesaurus Topics

  • Computational Science
  • Computer Science
  • Electrical Engineering
  • Event Detection
  • Hidden Markov Models
  • Language
  • Learning
  • Linguistics
  • Machine Learning
  • Markov Models
  • Metadata
  • Natural Language Processing
  • Natural Languages
  • Probability
  • Recognition
  • Speech
  • Word Recognition

Readers

  • Computer Vision.
  • Speech Processing/Speech Recognition.