The current project is concerned with predicting text comprehension from statistical measures of the reading process. The process of skilled reading is usually marked by fast information processing, as well as a systematic coupling between reading process and linguistic text features. The problem is, however, that reading speed has an ambiguous relationship to text comprehension, and reading speed components are not strongly predictive of text comprehension. Also, coupling between reading process and text is usually assessed by correlating process measures (response times, eye movements...) with linguistic text features (lexical, semantical, syntactical and coherence-related text descriptors...). However, the strength of this correlation varies systematically between different languages and different reading tasks
Our project advances the hypothesis that reading time regularity (RTR) can serve as a new metric for predicting reading comprehension from reading process measures. RTR is based on statistical measures of the degree of regularity in processing of a sequence such as time series measures of response times or eye movements during text reading. Particularly, it is hypothesized that RTR captures the coupling strength between reading process measures and linguistic text features, and that the degree of this coupling strength is predictive of reading comprehension. Because the calculation of RTR does not entail linguistic features, but is based on reading process measures alone, it is hypothesized that RTR makes for a language-independent metric of reading process fluency. In the current project, three specific sub-hypotheses are investigated:
- RTR captures coupling between reading process and text.
- RTR of the reading process predicts reading comprehension.
- RTR predicts comprehension invariantly across reading in different languages. To that end, cross-linguistic studies of reading in Chinese, English, German, and Hebrew are being conducted.