The Reliability of a Dialogue Structure Coding Scheme

A new subjective dialogue structure coding scheme is described that differs from previous schemes in three ways: (1) its model-based move categories are relatively task-independent, (2) dialogue structure is classified additionally at the levels of game & transaction structures, & (3) dialogue structure is represented generically. Dialogues are divided into transactions; transactions are divided into conversational games (also called exchanges or interactions) that can nest recursively; & games consist of conversational moves, defined as initiations & responses classified by purpose. Six initiation moves & five response moves are exemplified in 36 discourse fragments. In reliability tests (N = 4), four dialogues of the map task corpus were segmented & coded; measurements include the reliability of move segmentation, move classification, move classification from written instructions, move coding in a different domain (N = 2), game coding, & transaction coding. Results show reliable reproduction across subjects of most tasks at the move level, whereas problems were found in transaction coding; game coding displayed subject-internal stability with reasonable reproducibility. 1 Figure, 27 References. Adapted from the source document