A Model of Individual Differences in Gaze Control During Reading

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1810–1815, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics A Model of Individual Differences in Gaze Control During Reading Niels Landwehr1 and Sebastian Arzt1 and Tobias Scheffer1 and Reinhold Kliegl2 1 Department of Computer Science, Universit¨at Potsdam August-Bebel-Straße 89, 14482 Potsdam, Germany {landwehr, sarzt, scheffer}@cs.uni-potsdam.de 2 Department of Psychology, Universit¨at Potsdam Karl-Liebknecht-Straße 24/25, 14476 Potsdam OT/Golm kliegl@uni-potsdam.de Abstract We develop a statistical model of saccadic eye movements during reading of isolated sentences. The model is focused on rep- resenting individual differences between readers and supports the inference of the most likely reader for a novel set of eye movement patterns. We empirically study the model for biometric reader identifica- tion using eye-tracking data collected from 20 individuals and observe that the model distinguishes between 20 readers with an accuracy of up to 98%. 1 Introduction During skilled reading, the eyes of a reader do not move smoothly over a text. Instead, read- ing proceeds by alternating between brief fixations on individual words and short ballistic movements called saccades that move the point of fixation to a new location. Evidence in psychological research indicates that patterns of fixations and saccades are driven partly by low-level visual cues (e.g., word length), and partly by linguistic and cognitive pro- cessing of the text (Kliegl et al., 2006; Rayner, 1998). Eye-movement patterns are frequently studied in cognitive psychology as they provide a rich and detailed record of the visual, oculomotor, and linguistic processes involved in reading. Com- putational models of eye-movement control de- veloped in psychology, such as SWIFT (Engbert et al., 2005; Schad and Engbert, 2012) or E-Z Reader (Reichle et al., 1998; Reichle et al., 2012), simulate the generation of eye movements based on physiological and psychological constraints re- lated to attention, visual perception, and the ocu- lomotor system. Recently, the problem of mod- eling eye movements has also been approached from a machine-learning perspective. Matties and Søgaard (2013) and Hara et al. (2012) study con- ditional random field models to predict which words in a text are fixated by a reader. Nilsson and Nivre (2009) use a transition-based log-linear model to predict a sequence of fixations for a text. A central observation made by these studies, as well as by earlier psychological work (Erdmann and Dodge, 1898; Huey, 1908), is that eye move- ment patterns vary significantly between individ- uals. As one example of the strength of indi- vidual differences in reading eye movements, we cite Dixon (1951) who compared the reading pro- cesses of university professors and graduate stu- dents of physics, education, and history on read- ing material in their own and the two other fields. He did not find strong effects of his experimen- tal variables (i.e., field of research, expertise in re- search) but “if there is one thing that this study has shown, it is that individual differences in read- ing skill existed among the subjects of all depart- ments. …

Content truncated to 500 words