Workshop has been cancelled due to lack of submissions of the interest form attached.

About this workshop

Motivation

In human-human communication, misunderstanding occurs in most conversations (Dingemanse et al., 2015) and it is overcome through conversational grounding (Brennan and Clark, 1996) where interlocutors provide clarification and repair when needed during conversation. Yet, most of the training data in visual tasks for spoken dialogue systems nowadays lack conversational grounding phenomena making model training difficult for systems aiming to learn to coordinate when miscommunication occurs (Lemon, 2022).

This workshop aims to address this issue by designing new data collection tasks with divergent agents. The design of new tasks will promote the development of models that learn rapidly to reach agreement on shared tasks when they may have different perspectives, perceptions, language, and plans that lead towards miscommunication and how to repair it. If you are interested in attending the workshop, please fill out our form here.

Plan

We plan to hold a half day hackthon/challange on September 12th, 2023.

The hackathon will consist of two steps:

1) Revise previous data sets that have already focused on collecting data for shared tasks but that have not used divergent agents.

2) Design data collection scenarios. Carefully design new data collection scenarios where differences in perspectives, perceptions, language, and plans are implemented.

Organisers