in: Journal of Artificial Intelligence Research 6, pages. 35-85, 1997

POSTSCRIPT

SCREEN: Learning a Flat Syntactic and Semantic Spoken Language Analysis Using Artificial Neural Networks

Stefan Wermter wermter@informatik.uni-hamburg.de
Volker Weber weber@informatik.uni-hamburg.de
Department of Computer Science
University of Hamburg
22527 Hamburg, Germany

Abstract:

Previous approaches of analyzing spontaneously spoken language often have been based on encoding syntactic and semantic knowledge manually and symbolically. While there has been some progress using statistical or connectionist language models, many current spoken-language systems still use a relatively brittle, hand-coded symbolic grammar or symbolic semantic component.

In contrast, we describe a so-called screening approach for learning robust processing of spontaneously spoken language. A screening approach is a flat analysis which uses shallow sequences of category representations for analyzing an utterance at various syntactic, semantic and dialog levels. Rather than using a deeply structured symbolic analysis, we use a flat connectionist analysis. This screening approach aims at supporting speech and language processing by using (1) data-driven learning and (2) robustness of connectionist networks. In order to test this approach, we have developed the SCREEN system which is based on this new robust, learned and flat analysis.

In this paper, we focus on a detailed description of SCREEN's architecture, the flat syntactic and semantic analysis, the interaction with a speech recognizer, and a detailed evaluation analysis of the robustness under the influence of noisy or incomplete input. The main result of this paper is that flat representations allow more robust processing of spontaneous spoken language than deeply structured representations. In particular, we show how the fault-tolerance and learning capability of connectionist networks can support a flat analysis for providing more robust spoken-language processing within an overall hybrid symbolic/connectionist framework.

• Introduction
  • Screening Approach: Flat Representations Support Robustness
  • Flat Representations Learned in a Hybrid Connectionist Framework
  • Organization and Claim of the Paper
• Processing Spoken Language
  • ``Noise'' in Spoken Language
  • ``Noise'' from a Speech Recognizer
• Flat Category Representation: An Intermediate Connecting Representation
  • Categories for Flat Syntactic Analysis
  • Categories for Flat Semantic Analysis
• The Architecture of the SCREEN System
  • General Motivation for the Architecture
  • An Overview of the Architecture
  • A More Detailed View
    • Speech Sequence Construction Part
    • Speech Evaluation Part
    • Category Part
    • Correction Part
• Detailed Analysis with Examples
  • The Overall Environment
  • Analyzing the Final Snapshot in Short Sentence Hypotheses
  • Phrase Starts and Phrase Groups in Longer Sentence Hypotheses
  • Dealing with Noise as Repairs
• Design Analysis of SCREEN
  • Why Did We Use Connectionist Networks in SCREEN?
  • Improvement in the Hypothesis Space
  • SCREEN's Network Performance and Why the Networks Yield High Accuracy with Little Training
  • SCREEN's Overall Output Performance
  • SCREEN's Overall Performance for an Incomplete Lexicon
  • Comparison with the Results in a New Different Domain
  • An Illustrative Comparison Argument Based on a Symbolic Parser
  • Comparisons with Related Hybrid Systems
• Discussion
• Conclusions
• Appendix - Animation of parses
  • Appendix A
  • Appendix B
  • Appendix C
• References
• About this document ...