Collecting Visually-Grounded Dialogue with A Game Of Sorts

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 2257–2268 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 2257 Collecting Visually-Grounded Dialogue with A Game Of Sorts Bram Willemsen, Dmytro Kalpakchi, Gabriel Skantze Division of Speech, Music and Hearing, KTH Royal Institute of Technology Stockholm, Sweden {bramw, dmytroka, skantze}@kth.se Abstract An idealized, though simplistic, view of the referring expression production and grounding process in (situated) dialogue assumes that a speaker must merely appropriately specify their expression so that the target referent may be successfully identified by the addressee. However, referring in conversation is a collaborative process that cannot be aptly characterized as an exchange of minimally-specified referring expressions. Concerns have been raised regarding assumptions made by prior work on visually-grounded dialogue that reveal an oversimplified view of conversation and the referential process. We address these concerns by introducing a collaborative image ranking task, a grounded agreement game we call “A Game Of Sorts”. In our game, players are tasked with reaching agreement on how to rank a set of images given some sorting criterion through a largely unrestricted, role-symmetric dialogue. By putting emphasis on the argumentation in this mixed-initiative interaction, we collect discussions that involve the collaborative referential process. We describe results of a small-scale data collection experiment with the proposed task. All discussed materials, which includes the collected data, the codebase, and a containerized version of the application, are publicly available. Keywords: Visually-Grounded Dialogue, Data Collection, Referring Expressions 1. Introduction Visually-grounded dialogues are conversations in which entities in a co-observed visual context are refer- enced. In order for an addressee to successfully ground a referring expression, the description of the referent should be appropriately specified. If a speaker were to abide by the Gricean Maxim of Quantity (Grice, 1975), we would expect them to produce a referring expres- sion containing precisely the content necessary for the addressee to identify the target referent. Instead, in addition to such minimally-specified referring expres- sions, we commonly observe those that are over- and underspecified, meaning they contain more or less in- formation than is strictly necessary to correctly ground the description (Arts et al., 2011; Koolen et al., 2011; Gatt et al., 2013; Rubio-Fernandez, 2019). Moreover, as dialogue is inherently an interactive process, produc- ing a description for a referent is often a collaborative effort, rather than a unilateral transfer of well-formed, unambiguous referring expressions (Clark and Wilkes- Gibbs, 1986). Notably, participants in a conversation will iteratively refine and simplify their descriptions when repeatedly addressing the same referent. Over time, the conversational partners may form so-called conceptual pacts, as they come to an (implicit) agree- ment on a shared conceptualization of a referent (Clark and Wilkes-Gibbs, 1986). Established pacts are said to be part of the common ground (Stalnaker, 1978; Clark and Wilkes-Gibbs, 1986; Brennan and Clark, 1996; Clark, 1996), i.e., that which is believed by the con- versational partners to be mutual knowledge. How- ever, even seemingly stable pacts are not immutable and …

Content truncated to 500 words