Myanmar Dialogue Act Recognition Using Bi-LSTM RNN

Abstract

Spoken language understanding (SLU) is an essential element of any dialogue system to understand the language where dialogue act (DA) recognition is also critical aspects of pre-processing step for speech understanding and dialogue system. This paper proposes a deep learning-based DA model which use a deep recurrent neural network (RNN) with bi-directional long short-term memory (Bi-LSTM). The model mainly consists of a word-encode layer, a Bi-LSTM layer, and a softmax layer. For corpus preparation, we collected and annotated a large dialog act annotation corpus, which is called MmTravel (Myanmar Travel) corpus, on travel domain human-human conversations dataset (consists of 80k utterances). This paper reports analysis and comparison of proposed model Bi-LSTM with RNN, LSTM, and baseline SVM model. Experiments on the dataset is shown that our proposed DA model performs better than our previous work, support vector machine (SVM) models, which achieve an improvement of more than 2% accuracy increase in classification on the dataset.