Grounding as a Collaborative Process

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 515–531 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 515 Grounding as a Collaborative Process Luciana Benotti Universidad Nacional de C´ordoba CONICET, Argentina luciana.benotti@unc.edu.ar Patrick Blackburn Philosophy and Science Studies, IKH Roskilde University, Denmark patrickb@ruc.dk Abstract Collaborative grounding is a fundamental as- pect of human-human dialog which allows people to negotiate meaning. In this paper we argue that it is missing from current deep learning approaches to dialog and interactive systems. Our central point is that making mis- takes and being able to recover from them col- laboratively is a key ingredient in grounding meaning. We illustrate the pitfalls of being unable to ground collaboratively, discuss what can be learned from the language acquisition and dialog systems literature, and reflect on how to move forward. 1 Introduction Collaborative grounding is shaped by constraints that are not explicit in successful dialog turns. These constraints combine information from world causal relations, the task under discussion, the com- municative intents of the dialog partners, and much else besides. They are used to negotiate meaning, to decide which beliefs to add to the shared com- mon ground, which is constructed using the joint attention of the dialog partners to things either real or imagined. Once beliefs are grounded, they can- not be magically ungrounded without further ne- gotiation, for the dialog partners are committed to them. But it’s tricky: Human-analogous natural language understanding (NLU) is a grand challenge of artificial intelligence, which involves mastery of the structure and use of language and the ability to ground it in the world. (Bender and Koller, 2020) What does “the ability to ground it in the world” involve? Implicit constraints that shape ground- ing become explicit when communication starts to break. We claim that making mistakes and being able to recover from them collaboratively is a key ingredient of the ability to ground. We proceed as follows. We first discuss obsta- cles to current research on dialog and interactive systems (Section 2) and then present working defi- nitions of collaborative grounding and related con- cepts (Section 3). Section 4 plays with scenarios illustrating the pitfalls of being unable to ground collaboratively; we then turn to the language acqui- sition literature for insight into how humans do it (Section 5) and to the dialog systems literature to discuss what is known (Section 6). In Section 7, we reflect on the progress made and make recom- mendations, while in Section 8 we note possible objections to our account and conclude. 2 Motivation Dialog and interactive systems is one of the most popular research areas in computational linguistics nowadays. But — unlike machine translation and information retrieval — deep learning approaches to it have had little impact on products that people use daily.1 In 1991, cognitive scientist Brennan asked: Why is it that natural language has yet to become a widely used modality of human/computer interaction? (Brennan, 1991), and in 2020 the …

Content truncated to 500 words