Do frequency and frequency-related measures signal turn completion? An exploratory corpus study

Christoph Rühlemann^*

• University of Freiburg, Freiburg, Germany

Speakers in conversation have access to word frequency information stored in the mental lexicon. This paper examines whether word frequencies play a role as a turn-completion cue in conversation. Based on the Freiburg Multimodal Interaction Corpus (FreMIC) frequencies and frequency-related measures are compared in turn-constructional units (TCUs) from two types of action/turns that are systematically complementary with regard to turn transition: question TCUs, which exert pressure for a next speaker to take over, and storytelling TCUs, which largely resist transition. Based on these systematic tendencies, the focus is on question TCUs that result in speaker change and story TCUs that result in speaker continuation, thereby tying turn-transition inevitably to social action. We address two research questions: RQ #1 - Do word frequencies in the TCUs follow an S-shaped pattern? and RQ #2 - Which frequency-related measures predict that a TCU will be followed by turn transition or continuation? To address RQ #1, a mixed effects model showed the same S-shape found in prior research in large corpora. To address RQ #2, a mixed-effects model was computed, with Turn Transition (TT) as a binary outcome variable. The model suggested that turn finality in question TCUs co-occurs with a more pronounced drop in word frequency toward the TCU end than in story TCUs. A follow-up analysis revealed a more asymmetrical (right-leaning) distribution of nouns in turn-final question TCUs. Information extracted from word frequencies may hence serve listeners in conversation as cues to anticipate turn completion in questions as opposed to turn continuation in stories.