Regularizing Dialogue Generation by Imitating Implicit Scenarios

Abstract

Human dialogues are scenario-based and appropriate responses generally relate to the latent context knowledge entailed by the specific scenario. To enable responses that are more meaningful and context-specific, we propose to improve generative dialogue systems from the scenario perspective, where both dialogue history and future conversation are taken into account to implicitly reconstruct the scenario knowledge. More importantly, the conversation scenarios are further internalized using imitation learning framework, where the conventional dialogue model that has no access to future conversations is effectively regularized by transferring the scenario knowledge contained in hierarchical supervising signals from the scenario-based dialogue model, so that the future conversation is not required in actual inference. Extensive evaluations show that our approach significantly outperforms state-of-theart baselines on diversity and relevance, and expresses scenario-specific knowledge. But the ad said they will be sold for a week. I'm glad these cheap and cheerful batteries are on sale. Unfortunately, all types of batteries are costing more these days. Could you cut the price a little, please? Scenario 1 Scenario 2But the ad said they will be sold for a week.

Towards Standard Criteria for human evaluation of Chatbots: A Survey

“…Likert rating Shum et al [35] Chatbot Paper Whether responses are grammatically correct and sound natural Likert rating Ji et al [40] Chatbot Paper Generate utterance is readablity and grammatical correctness Likert rating Deriu and Cieliebak [92] Chatbot Paper Question: Which entities' language is more fluent and grammatically correct? Pairwise comparison Feng et al [45] Chatbot Paper Question: how likely the generated response is from human? Pairwise comparison Gao et al [16] Chatbot Paper the grammatical correctness of responses Pairwise comparison Yang et al [48] Chatbot Paper If the response is fluent without any grammatical errors?…”

Stop Filtering: Multi-View Attribute-Enhanced Dialogue Learning

“…The second further proposes the objectives that aligns with the goals of the conversation more effectively, such as MMI (Li et al, 2016a), CVAE (Serban et al, 2017b;Gu et al, 2019;Sun et al, 2021), RL (Li et al, 2016b;Zhang et al, 2018a;, and GAN Feng et al, 2020a). The third tries to endow the responses with topic (Xing et al, 2017;Feng et al, 2020b), emotion (Zhou et al, 2018;Rashkin et al, 2019), and persona (Qian et al, 2017;Zhang et al, 2018b;Song et al, 2020). Recently, another line of work Roller et al, 2020;Adiwardana et al, 2020;Bao et al, 2020), called the pre-trained dialogue model, relies on an efficient neural network and large-scale datasets to further improve the response quality.…”

ProphetChat: Enhancing Dialogue Generation with Simulation of Future Conversation

“…In response generation, Feng et al (2020a) proposed to use gold futures as the conditions of two discriminators and adopted adversarial training to encourage diversity. Feng et al (2020b) employed gold dialogue futures to learn a future-aware teacher model and transferred the knowledge to a history-to-response student model via imitation learning. These works only use the future information in the training phase, while we utilize the simulated dialogue future in the inference phase to provide the history-to-response generation model with direct help.…”