The Bielefeld Jigsaw Map Game (JMG) Corpus

Spoken language still poses a challenge to mechanisms developed for information processing and retrieval. Applications in this area often require a large amount of annotated data, which is hardly obtainable for spoken language. We present a corpus of 78 semi-natural dialogues (length: ≈20 h) completely transcribed and annotated on various linguistic levels. The dialogues stem from a psycholinguistic, task-oriented coordination game, the Jigsaw Map Game (JMG) (Weiß, Pfeiffer, Schaffranietz, and Rickheit 2008). In order to solve the task of the JMG, the speakers produced spontaneous utterances about objects from a predefined object set, that have to be located according to a map. What makes this data special is the combination of fixed utterance topics and nonetheless unconstrained language use. Primarily developed and annotated to study alignment in communication (Pickering and Garrod 2004), the corpus represents a useful resource for natural language processing and studies on spoken language. We describe 1. the theoretical motivation of the JMG, 2. the experimental setting used for data gathering, 3. the levels of annotation and its correction, as well as 4. the applications of the JMG corpus in the area of lexical alignment.