Dr. Phil Cohen, Openstream.ai® - joint work with Dr. Lucian Galescu

Title: A Planning-Based Explanatory Collaborative Dialogue System

Abstract: Eva is a multimodal conversational system that helps users to accomplish their domain goals through collaborative dialogue. The system does this by inferring users’ intentions and plans to achieve those goals, detects whether obstacles are present to their achievement, finds plans to overcome those obstacles or to achieve higher-level goals, and plans its actions, including speech acts,  to help users accomplish them. In that sense, Eva is a  collaborative “planning-based” dialogue system – one whose dialogue engine is a planner. In doing so, the system maintains a detailed "theory of mind" that enables it to reason with its own beliefs, goals and intentions, and to reason about those of its user. Belief reasoning is accomplished with a modal Horn-clause meta-interpreter. The planning and reasoning subsystems obey the principles of persistent goals and intentions as described in our prior work, including the formation, relativization, and decomposition of intentions to perform complex actions, as well as the conditions under which they can be given up. Among the system’s repertoire of actions are speech acts, some of whose definitions have been given in various papers of ours. In virtue of its multiagent planning process, the system treats its speech acts just like its other actions — physical acts affect physical states, digital acts affect digital states, and speech acts affect mental and social states. This general approach enables Eva to plan a variety of speech acts in dialogue, including requests, informs, questions, confirmations, recommendations, offers, acceptances, greetings, and emotive expressions. Each of these has a formally specified semantics which is used during the planning and reasoning processes. Because it can keep track of different users’ mental states, it can engage in multi-party dialogues. Importantly, Eva can explain its utterances because it has created a plan that has led to each of them. Finally, Eva employs multimodal input and output, driving an avatar that can perceive and employ facial and head movements along with emotive speech acts.

Bios: Dr. Phil Cohen, Chief Scientist at Openstream, has long engaged in research in the fields of human-computer dialogue, multimodal interaction, and multiagent systems. He is a Fellow of the Association for Computing Machinery, Fellow of the Association for the Advancement of Artificial Intelligence, Fellow and Past President of the Association for Computational Linguistics, and the recipient of the 2017 Sustained Accomplishment Award from the International Conference on Multimodal Interaction. He is a co-inventor of the Open Agent Architecture at SRI International, which led to Apple’s SiriTM digital assistant. Cohen is one of the two winners of the Inaugural Influential Paper Award  from the International Foundation for Autonomous Agents and Multi-Agent Systems for his paper with Hector Levesque “Intention is Choice with Commitment”, Artificial Intelligence 42(2-3), 1990. He is also an Adjunct Professor of Data Science and Artificial Intelligence, in the Faculty of Information Technology, Monash University, Melbourne, Australia.

Dr. Lucian Galescu is the Director of Dialogue Research at Openstream, where he focuses on plan-based conversational AI. Earlier, he was a research scientist with the Florida Institute for Human and Machine Cognition. Dr. Galescu’s research has spanned a broad spectrum of topics in dialogue systems, natural language understanding, knowledge extraction, intention recognition, and statistical language modeling. He has a number of widely cited publications in conversational AI and holds two patents in related technologies.

Prof. Ryuichiro Higashinaka, Nagoya University

Title: What's missing in current dialogue systems?

Abstract: Dialogue systems have evolved considerably over the past few years. This is mainly due to large-scale language models. In my talk, I will describe two competitions that I'm involved with (dialogue system live competition and dialogue robot competition) and a character-oriented chatbot we developed, and discuss the effectiveness of large-scale language models and the errors that they cause. Then, to cope with such errors, I emphasize the importance of building common ground. Towards a dialogue system that can build common ground with users, I will present some of our recent work on the analysis of how common ground is realized among humans.

Bio: Prof. Higashinaka is a professor at the Graduate School of Informatics at Nagoya University, Japan. He is a visiting senior distinguished researcher at NTT Human Informatics Laboratories. His goal is to realize dialogue systems that can achieve mutual understanding and intellectual collaboration with humans. He was a program co-chair for SIGDIAL 2016. From 2017 to 2021, he served on the SIGdial board. He is currently serving as an associate editor for the Dialogue and Discourse journal.

Prof. Marilyn Walker, University of California Santa Cruz

Title: Controlling Dialogue Acts with Zero-Shot Response Generation and Ranking 

Abstract: Dialogue systems need to produce responses that realize multiple types of dialogue acts (DAs) with high semantic fidelity. In the past, response generation has been typically handled by training a data-to-text natural language generator (NLG) from a large paired corpus that maps from a domain-specific meaning representation (MR) that specifies the desired DA and associated attributes, to one or more reference utterances. However recent advances in pretrained language models offer new possibilities for semantically controlled NLG. Here we apply an overgenerate and rank method to compare seven zero-shot prompt styles that include a novel method of generating from textual pseudo-references using a textual style transfer approach, a second novel approach that provides definitions of DAs in the prompts, inspired by previous work on schema-guided NLG, and a baseline of simply linearizing the MR. We then develop eight ranking functions to automatically identify outputs with both the correct DA and high semantic accuracy. To our knowledge, this is the first work on NLG for dialogue that automatically evaluates and ranks outputs using DA accuracy. For completeness, we also fine-tune few-shot models with up to 100 instances per DA. Our results show that several prompting styles achieve perfect DA accuracy, but with significantly worse semantic accuracy (94.7%) compared to few-shot fine-tuning (97.7%). However, few-shot fine-tuning performs much worse on DA accuracy (80.6%). Our results also show that formulating the data-to-text task as textual style transfer using pseudo-references performs the best. 

Bio: Marilyn Walker, is a Professor of Computer Science and Engineering at UC Santa Cruz, and a fellow of the Association for Computational Linguistics (ACL), in recognition of her for fundamental contributions to statistical methods for dialog optimization, to centering theory, and to expressive generation for dialogue. Her current research includes work on computational models of dialogue interaction and conversational agents, evaluation methods for dialogue systems, analysis of affect, sarcasm and other social phenomena in social media dialogue, and statistical methods for training the dialogue manager and the language generation engine for dialogue systems. Before coming to Santa Cruz, Marilyn was a Professor of Computer Science at University of Sheffield. From 1996 to 2003, she was a Principal Member of Research Staff in the Speech and Information Processing Lab at AT&T Bell Labs and AT&T Research. While at AT&T, Marilyn worked on the AT&T Communicator project, where she developed a new architecture for spoken dialogue systems and statistical methods for dialogue management and generation. Marilyn has published more than 200 papers, and has 10 granted/pending U.S. patents. She received her Ph.D. in Computer Science from University of Pennsylvania (1993).