The 2nd International Workshop on Search-Oriented Conversational AI (SCAI)
at EMNLP 2018, Brussels, Belgium 🇧🇪, October 31.
More and more information is found and consumed in a conversational form rather than using traditional search engines. Chatbots, personal assistants in our phones and eyes-free devices are being used increasingly more for different purposes, including information retrieval and exploration. On the other side, information retrieval empowers dialogue systems to answer questions and to get context for assisting the user in her tasks. With the recent success of deep learning in different areas of natural language processing, this appears to be the right foundation to power search conversationalization. Yet, we believe more can be done for theory and practice of conversation-based search and search-based dialogues.
This workshop aims to bring together researchers from the NLP, Machine Learning, and IR communities to lay the ground for search-oriented conversational AI and establish future directions and collaborations.
The 1st edition of the workshop was co-located with International Conference on the Theory of Information Retrieval (ICTIR 2017).
- Title: Towards natural conversation with machines using deep learning
- Abstract: Deep learning has made a revolution in machine learning, natural language processing and computer vision. In this talk, I will explain how deep learning can help solve some of the problems that dialogue modelling is facing. These include: scalable belief tracking, policy optimisation for large action spaces, and data-driven user modelling. I will also briefly advertise an initiative of the Cambridge Dialogue Systems Group to address the problem of evaluation of dialogue systems.
- Title: Understanding the User in Socialbot Conversations
- Abstract: Much past research on human-computer dialog has addressed task-oriented scenarios, but there is growing interest in building systems with social interaction capabilities, from companionship chitchat to information and opinion exchange. For systems that emphasize social interaction (e.g. a socialbot), user modeling can be especially important – people have different tastes in conversation topics as well as different interaction styles. This talk looks at the user in spoken interactions enabled by Sounding Board, a socialbot developed for the 2017 Amazon Alexa Prize competition, which enabled collection of millions of conversations with real users. We describe mechanisms for characterizing user variation and first steps towards predicting conversational preferences.
- Title: Wizard of Wikipedia: Knowledge-Powered Conversational Agents
- Abstract: In open-domain dialogue intelligent agents should exhibit the use of knowledge, however there are few convincing demonstrations of this to date. The most popular sequence to sequence models typically “generate and hope” generic utterances that can be memorized in the weights of the model when mapping from input utterance(s) to output, rather than employing recalled knowledge as context. Use of knowledge has so far proved difficult, in part because of the lack of a supervised learning benchmark task which exhibits knowledgeable open dialogue with clear grounding. To that end we collect and release a large dataset with conversations directly grounded with knowledge retrieved from Wikipedia. We then design architectures capable of retrieving knowledge, reading and conditioning on it, and finally generating natural responses. Our best performing dialogue models are able to conduct knowledgeable discussions on open-domain topics as evaluated by automatic metrics and human evaluations, while our new benchmark allows for measuring further improvements in this important research direction. This is joint work with Emily Dinan, Stephen Roller, Kurt Shuster, Angela Fan and Michael Auli (joint first authors).
- Title: Towards Open-Domain Conversational AI
- Abstract: Interacting with machines via natural language has been an emerging trend. The goal of developing open-domain dialogue systems that not only emulate human conversation but fulfill complex tasks, such as travel planning, seemed elusive. Recent advances in deep learning enabled new research frontiers for end-to-end conversational systems. This talk will review the research work about deep learning and reinforcement learning technologies that have been developed for two types of conversational agents. First is a task-oriented dialogue system that can help users accomplish tasks, ranging from meeting scheduling to vacation planning. Second is a social bot that can converse seamlessly and appropriately with humans. This talk will conclude with the advanced work that attempted to develop open-domain neural dialogue systems by combining the strengths of both types of agents.
- Title: Dialogues, Speech and Vision: Communication to make AI more human
- Abstract: Conversational AI such as Alexa, Siri, Google Home and Cortana are now strongly part of people lives. Numerous non-task oriented agents are also gaining importance such as Xiaoice and Ruuh. We will talk about the recent efforts going on to make these agents more human. We take a three pronged approach, dialogue, speech and vision to extend the humanness of these agents. In this talk we will touch upon some of the multi-faceted work going on in Microsoft IDC Hyderabad to attack some of these subfields which make AI more human.
- Neural Response Ranking for Social Conversation: A Data-Efficient Approach. Igor Shalyminov, Ondřej Dušek and Oliver Lemon (slides)
- Autonomous Sub-domain Modeling for Dialogue Policy with Hierarchical Deep Reinforcement Learning. Giovanni Yoko Kristianto, Huiwen Zhang, Bin Tong, Makoto Iwayama and Yoshiyuki Kobayashi (slides)
- Building Dialogue Structure from Discourse Tree of a Question. Boris Galitsky and Dmitry Ilvovsky (slides)
- A Methodology for Evaluating Interaction Strategies of Task-Oriented Conversational. Marco Guerini, Sara Falcone and Bernardo Magnini
- 12:45–14:00 Lunch break
- Milica Gašić, University of Cambridge
- Antoine Bordes, Facebook AI Research
- Jason Weston, Facebook
- Bill Dolan, Microsoft Research