Title: Two types of grounding in dialogue
Abstract: In 1991, Brennan asked: Why is it that natural language has yet to become a widely used modality of human/computer interaction? Almost 30 years later de Vries (2020) and colleagues ask the same question again. In this talk I will argue that a fundamental limitation of dialogue systems is that they cannot ground language into the world in a human-like way. I will distinguish two types of different but related kinds of grounding: perceptual and collaborative.
First, I will describe our work on perceptual grounding [1]. I will show recent advances on referential dialogue adapting multimodal architectures for visual question answering. I will discuss the limitations of treating dialogue as question answering. Second, I will present a recent opinion piece and state of the art survey on collaborative grounding [2]. Our central claim is that current dialogue systems try to avoid mistakes at all costs and this approach misses the key point that errors are a crucial mechanism in dialogue; for it is the ability to recover from them that makes dialogue such a robust process.
[1] Answering Different Visual Questions Requires Different Grounding Strategies. Alberto Testoni, Claudio Greco, Tobias Bianchi, Mauricio Mazuecos, Agata Marcante, Raffaella Bernardi. Proceedings of the International Workshop on Spatial Language Understanding. EMNLP 2020. https://www.aclweb.org/anthology/2020.splu-1.4/ [2] Grounding as a Collaborative Process. Luciana Benotti & Patrick Blackburn. Long paper accepted at the 16th Conference of the European Chapter of the Association for Computational Linguistics, EACL 2021. 19 to 23 April. Online.
Bio: Luciana Benotti is an Associate Professor at the Department of Computer Science in the Universidad Nacional de Cordoba, from Argentina. Her research interests include different aspects of situated dialogue systems, including the study of misunderstandings, clarification requests and grounding. She has an Erasmus Mundus MSc, and a PhD in Computer Science completed at INRIA Nancy Grand Est in France. She received an IBM SUR award for her work on robust conversational interfaces, and a Google RISE award for her outreach efforts in developing AI-based technology for education. She has been an invited scholar at the University of Trento (2019), Stanford University (2018), Roskilde University (2014), University of Lorraine (2013), Universidad de Costa Rica (2012), and University of Southern California (2010). She regularly serves under different roles in the Association for Computational Linguistics (ACL) community. She has been a volunteer during conferences, a reviewer since 2010, an area chair for dialogue and interactive systems several times, and a member of the executive board of SIGDIAL and SIGSEM. She is currently an elected member of the executive board of the North American Chapter of the ACL.
To obtain the Zoom link for this event, please see the event announcements from MLFL on the college email lists or contact Kalpesh Krishna.