This dissertation examines coordinated verbal and nonverbal symmetry in turn-taking behavior during conversation using three modern techniques for analyzing conversational data in nonstationary time series data: windowed cross-correlation, multifractal wavelet analysis, and recurrence quantification analysis. A major goal of this work was to combine the well-established qualitative method of analyzing verbal (and more recently nonverbal) discourse known as conversation analysis (CA) with more recent quantitative methods for analyzing nonstationary time series data. Windowed-cross correlation (WCC) was used to estimate the degree of relationship between head position and speech output within one participant as well as the degree of (verbal and nonverbal) correlation between participants using a correlation matrix-based approach. The multifractal wavelet analysis called wavelet transform modulus maxima (WTMM) examined the underlying structure of each data series including fractal dimension and the Holder exponent. Recurrence quantification analysis (RQA) and cross-recurrence quantification (CRQ) were used to investigate the phase space of each time series, further examining mirroring in verbal and nonverbal behavior (e.g., recurrences) within and between participants at the point where turn-taking occurs during conversation. These three analyses are appropriate human conversation data that is likely to be nonstationary because WTMM, RQA, and CRQ are nonparametric and WCC assumes only local stationarity (about 2 seconds). The two major goals of these analyses were to: 1) perform these three analyses to examine the relationship between an individual participant’s verbal and nonverbal data and the relationship between two people’s verbal or nonverbal data with a focus on the phenomenon of conversational turn-taking and, 2) estimate the predictability of the outcome of each of these three analyses based on gender and dominance were calculated. Finally, in order to demonstrate the real-world applicability of these analyses, the original digitized video of a conversation at salient time segments identified by the RQA/CRQA analyses were transcribed as an illustration of the practical meaning of the analyses using QuickTime (Apple Computers, 1991-2007) to compare to CA transcriptions and plots of the time series data.
|Author||Kathleen Targowski Ashenfelter|
|Degree Level||Doctoral Dissertation|
|Departments and Units|