About the Usefulness and Learnability of Argument-Diagrams from Real Discussions Rutger Rienks and Daan Verbree Human Media Interaction (HMI) University of Twente, Enschede, The Netherlands {Rienks, Verbree}@ewi.utwente.nl, Home page: http://hmi.ewi.utwente.nl/ Abstract. This paper continues the work described in Rienks and Heylen [2005] about argument diagramming of meeting discussions. In this paper we introduce the corpus that we created, discuss a user experiment about the usability of the technique, and show that the units of the diagram- ming method (segmented user utterances) can be learnt and predicted with an accuracy of 88.4% and 82.2% on an unbalanced and balanced set respectively. 1 INTRODUCTION Argumentation has been regarded as our primary means of making progress [van Gelder, 2002]. It is pervasive in everyday life and plays an important role in human communication. Argumentation is inherently related to discussions, conversations and meetings, the arenas where one argues with another and one or more sides are attempting to win the approval of the opponent or of a designated audience. Within organizations the general visible results of conversations or meetings are normally nothing more than what one is able to recall, if lucky from some notes that were taken, or perhaps some more formal meeting minutes or a list of action items. Generally, a lot of energy and information that has been put into the actual outcome is never seen again. In Twente we have tried to find an approach that is able to capture the lines of the deliberated arguments in meeting discussions. This approach, the TAS-schema, was introduced in Rienks and Heylen [2005] and promised to be a valuable technique for capturing organizational memory. The structure that the arguments encapsulate reveals information about the trail or path that has been taken and can show the line of reasoning at specific moments in time. The method can aid querying systems and can be used in meeting browsers (See fig 1). The possibility of preserving the arguments and their coherence relations for future explorations make them potentially valuable documents containing a tacit representation of otherwise volatile knowledge [Shum, 1997, Pallotta et al., 2005]. In this paper we show how we continued our research in this area. Before we elaborate on how we created a corpus of annotations in Section 3, Section 2 will