CaSiNo: A Corpus of Campsite Negotiation Dialogues for Automatic Negotiation Systems

Abstract

Automated systems that negotiate with humans have broad applications in pedagogy and conversational AI. To advance the development of practical negotiation systems, we present CaSiNo: a novel corpus of over a thousand negotiation dialogues in English. Participants take the role of campsite neighbors and negotiate for food, water, and firewood packages for their upcoming trip. Our design results in diverse and linguistically rich negotiations while maintaining a tractable, closeddomain environment. Inspired by the literature in human-human negotiations, we annotate persuasion strategies and perform correlation analysis to understand how the dialogue behaviors are associated with the negotiation performance. We further propose and evaluate a multi-task framework to recognize these strategies in a given utterance. We find that multi-task learning substantially improves the performance for all strategy labels, especially for the ones that are the most skewed. We release the dataset, annotations, and the code to propel future work in human-machine negotiations: https:// github.com/kushalchawla/CaSiNo.

Opponent Modeling in Negotiation Dialogues by Related Data Adaptation

“…We achieve this by adapting two additional data sources for this task, allowing the data to be directly added to the primary training dataset and enabling end-to-end parameter sharing between these related tasks. Datasets: We leverage two datasets in this work: CaSiNo (Chawla et al, 2021) and DealOrN-oDeal (Lewis et al, 2017). As discussed before, CaSiNo is grounded in a camping scenario, containing negotiations over three issues: food, water, and firewood.…”

“…MIBT is a generic framework that can be useful for many negotiation tasks beyond these datasets as well, for instance, salary negotiations, or negotiations between art collectors distributing the items among each other. It is extensively used in NLP (Lewis et al, 2017;Chawla et al, 2021;Yamaguchi et al, 2021), beyond NLP (Mell and Gratch, 2017), and in the industry as well (e.g. iDecisionGames 5 ).…”

“…Figure 1: A simplified view of a multi-issue negotiation based on the scenario in CaSiNo (Chawla et al, 2021). The negotiation involves 3 issues: Food, Water, and Firewood, each with 3 items that must be divided among the two players.…”

“…Consider the scenario presented in Figure 1. Two participants role-play as campsite neighbors and engage in a multi-issue negotiation (Fershtman, 1990) over three issues: food, water, and firewood (Chawla et al, 2021). Each negotiator has their own priority order depending on the relative importance assigned to each issue.…”

“…For a comprehensive evaluation that serves multiple downstream applications, we propose three evaluation metrics for this task (Section 4). Our experiments are based on two dialogue datasets in English: CaSiNo (Chawla et al, 2021) and DealOrNoDeal (Lewis et al, 2017), showing the utility of the proposed methodology with complete or partial dialogue as input in full, few-shot, and zero-shot scenarios (Section 5).…”