Language of Bargaining

TL;DR

The paper addresses how linguistic form and content influence bilateral bargaining by collecting a controlled dataset that contrasts numeric alternating offers (AO) with natural-language (NL) audio negotiation, annotated with a bargaining-act taxonomy. It shows that allowing language improves agreement rates and accelerates convergence, while mean agreed prices remain essentially unchanged; linguistic signals, especially LIWC-based features, reliably predict outcomes. The study develops predictive models—primarily logistic regression with linguistic features—outperforms neural encoders in this setting and reveals distinct buyer/seller dynamics, such as early seller-directed interrogation being predictive of favorable outcomes for sellers. The dataset and findings advance understanding of negotiation through language and have implications for designing negotiation-capable dialogue systems and training strategies, while acknowledging limitations of artificial lab settings and ASR-derived text.

Abstract

Leveraging an established exercise in negotiation education, we build a novel dataset for studying how the use of language shapes bilateral bargaining. Our dataset extends existing work in two ways: 1) we recruit participants via behavioral labs instead of crowdsourcing platforms and allow participants to negotiate through audio, enabling more naturalistic interactions; 2) we add a control setting where participants negotiate only through alternating, written numeric offers. Despite the two contrasting forms of communication, we find that the average agreed prices of the two treatments are identical. But when subjects can talk, fewer offers are exchanged, negotiations finish faster, the likelihood of reaching agreement rises, and the variance of prices at which subjects agree drops substantially. We further propose a taxonomy of speech acts in negotiation and enrich the dataset with annotated speech acts. Our work also reveals linguistic signals that are predictive of negotiation outcomes.

Figures (7)

• Figure 1: Gaussian kernel estimates of the distributions of agreed prices among successful negotiations.
• Figure 2: Distribution of Bargaining Acts. Error bars indicate standard error.
• Figure 3: The figure presents the trajectory of new offers in the two treatments. In \ref{['fig:ao_trajectory']} and \ref{['fig:nl_trajectory']}, each line represents a sequence of new offers exchanged between buyer and seller in a single negotiation. Only negotiations ending in agreement are included. Figure \ref{['fig:traj_delta']} presents the absolute differences in consecutive new offers under both treatments. Each dot represents an absolute difference in consecutive new offers within a single bargaining session.
• Figure 4: Overall prediction performance.
• Figure 5: Buyers vs. sellers. Accuracy of Logistic Regression model across different input features, using buyer speech, seller speech, or both. Error bars indicate standard error.