Combining Computational and Human Analysis to Study Low Coherence in Design Conversations

This paper presents a mixed computational and manual procedure to systematically probe for distinct low coherent turns in design conversations. Existing studies indicate that focus shifts and their linguistic equivalent, low coherent turns, positively influence ideational productivity. Because coherence is a versatile phenomenon, we contribute a classification of low coherent turns to enable future research to further investigate the influence of low coherence turns on creativity. We analyze the DTRS11 corpus, comprising 16 sessions of design conversation that contain 9830 sentences, with automated Latent Semantic Analysis (LSA) to identify potential low coherent turns. We argue that an additional manual coherence analysis with the Topic Markup Scheme (TMS) further qualifies preselected turns. This mixed method procedure constitutes a promising pragmatic instrument for locating low coherent turns in large corpora. 297 distinct low coherent turns out of a total amount of 6072 turns were successfully retained. The selected data contains twice as much turns that shift the focus of attention within an existing design issue as turns that interrupt and introduce a new design issue. Based on an interpretative analysis of low coherent turns, we suggest distinguishing between turns that interrupt the focus of attention and turns that shift the focus of attention through either diversifying, reframing, or selective tendencies.