Prediction as a Basis for Skilled Reading: Insights from Natural Language Engineering Models
Reading is not an inborn human capability, and yet, English-speaking adults read with impressive speed. This study considered how predictions of upcoming words impact on this skilled behaviour. We used a powerful computer model from natural language engineering (GPT-2) to derive predictions of upcoming words in text passages. These predictions were highly accurate, and showed a tight relationship to fine-grained aspects of eye-movement behaviour when adults read those same passages, including whether to skip the next word and how long to spend on it. Strong predictions that did not materialise resulted in a prediction error cost on fixation durations. Our findings suggest that predictions for upcoming words can be made based on relatively superficial statistical information in reading, and that these predictions guide how our eyes interrogate text. This study is the first to demonstrate a relationship between the internal state of a modern natural language engineering model and eye-movement behaviour in reading, opening substantial new opportunities for language research and application.