Speech acts classification of Farsi texts

The objective of this paper is to design a system for classification of Farsi speech acts. The driving vision for this work is to provide intelligent text to Farsi speech (TTS) systems that are sensitive to the speech acts of the input texts and can pronounce the corresponding intonation correctly. Seven speech acts were considered and 3 classification methods including of (1) Naive Bayes, (2) K-Nearest Neighbors (KNN), and (3) Tree learner were used. The performance of speech act classification was evaluated using these methods including 10-Fold Cross-Validation, 70-30 Random Sampling and Area under ROC. KNN with an accuracy of 72% was shown to be the best classifier for classification of Farsi speech acts. It was observed that the amount of labeled training data has an important role in the classification performance.