RNN Based Language Generation Models for a Hindi Dialogue System

Abstract

Natural Language Generation (NLG) is a crucial component of a Spoken Dialogue System. Its task is to generate utterances with intended attributes like fluency, variation, readability, scalability and adequacy. As the handcrafted models are rigid and tedious to build, people have proposed many statistical and deep-learning based models to bring about more suitable options for generating utterance on a given Dialogue-Act (DA). This paper presents some Recurrent Neural Network Language Generation (RNNLG) framework based models along with their analysis of how they extract intended meaning in terms of content planning (modelling semantic input) and surface realization (final sentence generation) on a proposed unaligned Hindi dataset. The models have shown consistent performance on our natively developed dataset where the Modified-Semantically-Controlled LSTM (MSC-LSTM) performs better than all in terms of total slot-error (T-Error).