Machine learning techniques in dialogue act recognition

Mark Fišel


This report addresses dialogue acts, their existing applications and techniques of automatically recognizing them, in Estonia as well as elsewhere. Three main applications are described: in dialogue systems to determine the intention of the speaker, in dialogue systems with machine translation to resolve ambiguities in the possible translation variants and in speech recognition to reduce word recognition error rate.

Several recognition techniques are described on the surface level: how they work and how they are trained. A summary of the corresponding representation methods is provided for each technique. The paper also includes examples of applying the techniques to dialogue act recognition.

The author comes to the conclusion that using the current evaluation metric it is impossible to compare dialogue act recognition techniques when these are applied to different dialogue act tag sets. Dialogue acts remain an open research area, with space and need for developing new recognition techniques and methods of evaluation.



conversation analysis; computational linguistics; dialogue act; machine learning; Bayes classifier; hidden Markov model; neural network; decision tree

Full Text:




  • There are currently no refbacks.

Copyright (c) 2012 Mark Fišel

Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

ISSN 1736-2563 (print)
ISSN 2228-0677 (online)
DOI 10.5128/ERYa.1736-2563