Target-Guided Dialogue Response Generation Using Commonsense and Data Augmentation

Anonymous

Paper Link: https://openreview.net/forum?id=XVrgLklgZN

Paper Type: Long paper (up to eight pages of content + unlimited references and appendices)

Abstract:

Target-guided response generation enables dialogue systems to smoothly transition a conversation from a dialogue context toward a target sentence. Such control is useful for designing dialogue systems that direct a conversation toward specific goals, such as creating non-obtrusive recommendations or introducing new topics in the conversation. In this paper, we introduce a new technique for target-guided response generation, which first finds a bridging path of commonsense knowledge concepts between the source and the target, and then uses the identified bridging path to generate transition responses. Additionally, we propose techniques to re-purpose existing dialogue datasets for target-guided generation. Experiments reveal that the proposed techniques outperform various baselines on this task.
Finally, we observe that the existing automated metrics for this task correlate poorly with human judgement ratings. We propose a novel evaluation metric that we demonstrate is more reliable for target-guided response evaluation. Our work generally enables dialogue system designers to exercise more control over the conversations that their systems produce.

Response To Ethics Reviews (for Conditionally Accepted Papers Only):

> Concern about Mechanical Turk
Thanks for pointing this out. We have included mturk arrangements and worker pay in the last paragraph of the section 7.3 in detail and also in the Ethics section.

> Broader impact and potential uses of the proposed system
Thanks for the suggestion. We have included discussion about the broader impact of our work in the first paragraph of the ‘Ethics and Broader Impact’ section. We discuss potential use-cases in sectors such as e-commerce. We have also incorporated the suggestions from the ethics review in the first paragraph of the introduction.

> Risks with deployment of the dialog system
Thanks for the pointers. In the second para of the ‘Ethics and Broader Impact’ section, we point out potential misuses of the proposed research, especially in the context of high-risk applications such as education. We suggest some mitigation protocols to alleviate unethical uses of deployed goal-driven bots, including compliance with regulations such as the European Union’s regulatory framework proposal on artificial intelligence.

> Limitations and potential biases (such as related to concepts graphs)
In the third paragraph of the ‘Ethics and Broader Impact’ section, we point out various limitations of the system. We discuss potential biases in our system arising from the use of existing knowledge bases and limitations of ml techniques. For example, existing conversational systems struggle with empathy, morality, discretion, and factual correctness, and there is a risk that a target-driven system would ignore these factors to achieve the target.

Presentation Mode: This paper will be presented in person in Seattle

Virtual Presentation Timezone: UTC-7

Copyright Consent Signature (type Name Or NA If Not Transferrable): prakhar gupta

Copyright Consent Name And Address: CARNEGIE MELLON UNIVERSITY, 5000 FORBES AVENUE, Pittsburgh PA