This special issue presents the expanded and improved versions of selected papers from a workshop on Formal and Informal Information Exchange in Negotiations. The workshop took place on May 26–27, 2005, at the University of Ottawa. It was designed to bring together researchers who work on various aspects of interaction in negotiations and those who work in Natural Language Processing or Machine Learning, on problems that might be of interest to negotiation specialists. Such problems include sentiment analysis. Recognizing the sentiments that negotiators express in language (for example, in messages exchanged during electronic negotiations) could offer insight into the negotiation process and a glimpse of the changes in feelings throughout the course of the negotiation. While external factors may influence users at any time, it is likely that they will react fairly consistently to the tone of the partner’s messages and offers. Analysis of texts to recognize attitudes, sentiments, and opinions has been receiving more and more attention in recent years. Hearst (1992) proposed direction-based text interpretation to complement topic analysis. Such interpretation showed where a text lies on the axis from opposed to in favor. Later work on sentiment analysis abandoned the continuous axis in favor of a binary view: recognize positive and negative (or favorable and unfavorable) attitudes. Most research has an economic motivation, as a result of companies wanting to monitor the consumers’ satisfaction with their products (Cognitrend,1 Tenorio Research).2 Opinion recognition may also increase the usefulness of search engines by allowing the user to compare the number and content of positive and negative opinions about a variety of products and services (Das and Chen 2001; Dave, Lawrence, and Pennock 2003; Hu and Liu 2004) or as part of recommender systems which collect, analyze, and summarize user opinions (Tarveen et al. 1997; Mooney, Bennet, and Roy 1998; Tatemura 2000). Hearst (1992) and Sack (1994) propose cognitively inspired models for sentiment analysis. Hearst’s model is inspired by Talmy’s theory of force dynamics (Talmy 1985), which describes the lexical and grammatical expressions of the interaction between two opposing entities—the agonist and the antagonist. Each entity expresses an intrinsic force, tending either toward motion or toward rest. The balance between these forces determines the resulting state of the interaction. In a variation on this idea, the focus is on one entity and on the events that affect it during encounters with other entities. This can be imagined as an entity following a path toward a goal or destination, and meeting barriers or facilitators along the way. This path model is what Hearst applies, with minor modifications, to queries that have a directional component, which imply finding whether an agent or event opposes or is in favor of another event. Sack briefly describes SpinDoctor, a system designed to identify the point