Classification of Modal Meaning in Negotiation Dialogues

This paper addresses modality classification for multi-issue bargaining dialogues in order to model human-like negotiation behaviour and to efficiently compute negotiation strategies. We propose a modality annotation and classification scheme comprising semantically distinguishable categories applied reliably by humans and machines. Our classification of modality varieties is based on both the traditional dichotomy of epistemic and root modalities, and on the analysis of the available corpus data. Our modality scheme has been used for annotating human-human dialogues and training SVM-based classifiers. We built predictive models that show accuracies in the range between 73.3 and 82.6%.