Predictability Effects in Natural Reading are Logarithmic, Not Linear: Evidence from an Eye-Movement Replication of Brothers and Kuperberg (2021)

The question of whether the relationship between a word’s predictability and its processing time is linear or logarithmic is of substantial theoretical importance, as it arbitrates between theories of the predictability effect (pre-activation vs. Surprisal) and has implications for sentence processing generally. While several previous corpus studies have obtained evidence for a logarithmic relationship, Brothers and Kuperberg (2021) obtained evidence for a linear relationship in a large self-paced reading study with well-controlled experimental materials. Here, we use Brothers and Kuperberg’s materials in an eyetracking-during-reading experiment. We find clear evidence for a logarithmic relationship between predictability and the eye movement measures of first fixation duration and gaze duration; this relationship is clearest when using predictability estimates from the large language model GPT-2, which can distinguish small differences in predictability at the low end of the scale. We find that this conclusion is robust to log transformation of the reading time measures, and that the relationship between predictability and the log odds of word skipping also appears to be logarithmic. These results support Surprisal as an account of predictability effects in natural reading.Keywords: eye movements, lexical predictability effects, surprisal, preactivation, reading