Reading as a core skill

Reading is a uniquely human skills upon which educational and professional success is predicated. But even skilled readers differ among themselves.

Although the goal of reading is for the reader to process and understand a text, most research on reading has been devoted to unravelling the principles underlying reading mechanics, that is, mapping letters to sounds. But meaning plays a role in reading, too, and meaning has received much less attention in the literature.

Form and meaning in reading

In our paper we ask: what role do orthographic and semantic information play in the behaviour of skilled readers?

We ran an experiment with native speakers of Russian who read through sentences, word by word, at their own pace. The orthographic and semantic information they had access to was modelled using the Naive Discrimination Learner. We found that the reading latencies from this self-paced sentence reading experiment were well predicted by a combination of orthography and semantics.

Differences between readers

There are, however, some interesting differences in the extent to which reading performance relies on orthography and semantics when you look at it from the perspective of mental speed.

Only the uptake of orthographic information was modulated by individual differences: readers who recorded slower mental speeds, as evidence by their performance on a serial reaction time task, were more dependent on orthographic support during reading, and read those words faster for which there was strong orthographic support. Readers who recorded fast mental speeds did not change their reading speed depending on the amount of orthographic support there is for a word.

All readers responded in the same way to semantic information; if there was little context support for a word, that word was read slowly, and they read faster as soon as some support was available. Interestingly, there is an additional effect: while initially participants read strongly supported words more slowly, halfway through the experiment their behaviour changed and by the end of the experiment they read words for which there is semantic support more quickly. We explained the observed differences in the role that semantics plays in reading in terms of the exploration/exploitation hypothesis from reinforcement learning, where initially slower and more variable behaviour leads to better performance overall.

Follow-up

We explore the usefulness of the exploration/exploitation hypothesis from reinforcement learning for modelling participant behaviour in experiments further in Divjak & Milin (2020): Exploring and Exploiting Uncertainty in Language Processing