Agreement detection in multiparty conversation

This paper presents a system for the automatic detection of agreements in multi-party conversations. We investigate various types of features that are useful for identifying agreements, including lexical, prosodic, and structural features. This system is implemented using supervised machine learning techniques and yields competitive results: Accuracy of 98.1% and a kappa value of 0.4. We also begin to explore the novel task of detecting the addressee of agreements (which speaker is being agreed with). Our system for this task achieves an accuracy of 80.3%, a 56% improvement over the baseline. General Terms MEASUREMENT, PERFORMANCE, EXPERIMENTATION Keywords agreement detection, multi-party conversation Categories and Subject Descriptors I.2 [Natural Language Processing]: Discourse; I.2 [Natural Language Processing]: Text Analysis

• Type: Conference
• Year: 2009
• Where: ICMI
• Authors: Sebastian Germesin, Theresa Wilson
• Keywords: Biometrics, ICMI 2009, Multi-party Conversation, Multi-party Conversation Categories, Natural Language Processing