Hindi Sentence Classification for Expressive Storytelling Systems

This paper proposes a method to classify sentences taken from short stories written in Hindi language into their discourse modes - narrative and dialogue. The automated classification improves the user experience with expressive speech. The classification has been attempted using the Bidirectional LSTM (Long Short Term Memory) units RNN (Recurrent Neural Network) model and a hybrid of CNN (Convolutional Neural Network) and LSTM. CNN-SVM (Support Vector Machine) which is the current best model has been taken as baseline. The proposed CNN-LSTM hybrid model with contextual word embeddings achieves better accuracy in Hindi language story sentence classification.