Natural language generation using transformer network in an open-domain setting

Abstract

Prior works on dialog generation focus on task-oriented setting and utilize multi-turn conversational utterance-response pairs. However, natural language generation (NLG) in the open-domain environment is more challenging. The conversations in an open-domain chit-chat model are mostly single-turn in nature. Current methods used for modeling single-turn conversations often fail to generate contextually relevant responses for a large dataset. In our work, we develop a transformer-based method for natural language generation (NLG) in an open-domain setting. Experiments on the utterance-response pairs show improvement over the baselines, both in terms of quantitative measures like BLEU and ROUGE and human evaluation metrics like fluency and adequacy.