It's about time: distributional analyses of psycholinguistic data.

• Applicant: Peter Hendrix, Ph.D.
• Subject Area: General and Comparative Linguistics, Experimental Linguistics, Typology, Non-European Languages
• Term: from 2017 to 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 387774888

Analyses of psycholinguistic data with temporal response variables (e.g. reaction times and fixation durations) have traditionally focused on the mean of response variable distributions. Although analyses of the mean are invaluable for assessing main trends, predictor effects need not be constant across the distribution of the response variable. Some predictors primarily influence short response times, whereas the effects of other predictors may be most prominent for slower responses. Insight into the temporal dynamics of the effects of lexical predictors promises to contribute substantially to our understanding of the mechanisms that underlie the language processing system.Recent advances in the development of statistical methods have created novel opportunities for distributional analyses of psycholinguistic data. The research proposed here investigates specifically the potential of two different statistical methods: time-to-event analysis and generalized additive quantile regression. Time-to-event analysis allows for an investigation of the relative timing and the temporal dynamics of lexical predictor effects. Generalized additive quantile regression allows for an investigation of both linear and non-linear predictor effects at different quantiles of the response variable distribution.Time-to-event analysis and generalized additive quantile regression can be applied to a wide range of temporal response variables in psycholinguistic research, including reaction times in chronometric tasks such as lexical decision or word naming and fixation durations in eye-tracking experiments. Initial explorations of such response variables through distributional analyses shed surprising new light on experimental data. For instance, whole-word frequency effects for complex words (e.g. "bookworm") turn out to temporally coincide with or precede frequency effects of constituents (e.g., "book", "worm") in both English and Mandarin Chinese. This pattern of results challenges models of lexical processing that posit that access to constituents precedes full-form access, while providing support, from an entirely new perspective, for semantics-first theories.The objective of the research proposed here is, first, to establish best practices for the application of distributional analyses of temporal response variables in psycholinguistic research, and second, to gain further insight into the nature of lexical processing through the re-analysis of psycholinguistic data sets and the re-evaluation of models of lexical processing on the basis of results from the distributional analyses. Results will be presented at academic conferences, published in academic journals, and written up in an introductory monograph on the application of distributional analyses for (psycho)linguistic data. In the interest of reproducibility, statistical analyses underlying all publications will be made available through the Potsdam Mind Research Repository.

• DFG Programme: Research Grants
• International Connection: Canada, France, United Kingdom
• Cooperation Partners: Paul Blanche, Ph.D.; Matteo Fasiolo, Ph.D.; Daniel Schmidtke, Ph.D.