While reading text, the eyes move forward in small jerks. These jerks are called saccades, and the pauses between saccades are called fixations. The length of the fixations varies, and they do so especially depending on whether the words in the text are easy or difficult to read. When a reader encounters words that is difficult to read, there are different patterns of behavior to be traced. Some readers keep fixating on difficult words, while others seek help with decoding by looking at pictures, other in the context in the surrounding text, or something completely different. Eye movements thus reveal something about the reading process of each individual reader.
One of the hypotheses that two PhD students at the University of Copenhagen Joachim Bingel and Maria Barret have investigated is whether a computer algorithm can see a pattern in the gaze behavior of readers when the text becomes difficult – and even read incorrectly. A peer-reviewed research article has emerged from the collaboration between EyeJustRead and the University of Copenhagen, and the article will be presented at the BEA workshop (Workshop on Innovative Use of Natural Language Processing for Building Educational Applications) in New Orleans in early June 2018. An exciting conclusion from the article is:
We showed that despite the noisy conditions under which this data was obtained, features we extract from the gaze patterns are predictive of reading mistakes among children. Besides the immediate application in automating some parts of reading teaching, this could be exploited in personalized text simplification, where gaze could be used as feedback to the system.
Bingel and Barret have found opportunities for eye movements in readers to be used to automatically detect reading errors. Thus, the manual work of continuously calculating the accuracy percentage can be automated when students read with EyeJustRead. Furthermore, such automation opens up to “detect” when previously misread words are decoded correctly.
The full article can be found here