Acquiring training data to improve the robustness of dialog systems can be a painstakingly long process. In this work, we propose a method to reduce the cost and effort of creating new conversational agents by artificially generating more data from existing examples, using paraphrase generation. Our proposed approach can kick-start a dialog system with little human effort, and brings its performance to a level satisfactory enough for allowing actual interactions with real end-users. We experimented with two neural paraphrasing approaches, namely Neural Machine Translation and a Transformerbased seq2seq model. We present the results obtained with two datasets in English and in French: a crowd-sourced public intent classification dataset and our own corporate dialog system dataset. We show that our proposed approach increased the generalization capabilities of the intent classification model on both datasets, reducing the effort required to initialize a new dialog system and helping to deploy this technology at scale within an organization.

As dialog systems are increasingly used, a major challenge for building new ones is the lack of annotated training data. The necessary data collection and annotation efforts are laborious and time-consuming. A potential solution is to augment initial seed data by automatically paraphrasing existing samples. In this paper, we propose a novel dataefficient approach towards this goal. Our method can kick-start a dialog system with minimum human effort while delivering a performance strong enough to allow real-world usage. We ran experiments using Neural Machine Translation on two open corpora. On both of them, the proposed approach improved the generalization capabilities of the model. Our results suggest that paraphrase generation techniques could be used as-is to provide a boost in performance to dialog systems in an early phase.

No abstract