Arabic Speech Act Recognition Techniques

摘要

This article presents rule-based and statistical-based techniques for Arabic speech act recognition. The proposed techniques classify an utterance into Arabic speech act categories based on three criteria: surface features, cue words, and contextual information. A rule-based expert system has been developed in a bootstrapping manner based on the fact that Arabic language syntax is inherently rule-based. Various machine-learning algorithms have been used to detect Arabic speech act categories: Decision Tree, Naive Bayes, Neural Network, and SVM. We compare the experimental results for both techniques (machine-learning and rule-based expert systems). Using a corpus of 1,500 sentences, the rule-based expert system achieved an accuracy rate of 98.92%, while the Decision Tree, Naive Bayes, Neural Network, and SVM achieved an accuracy rate of 97.09%, 96.48%, 93.50%, and 93.70%, respectively.