For many years fatigue has been a topic of scientific and applied research, especially in relation to evaluating a person’s cognitive state in occupations settings such as factory work, client services, education and other areas where a work routine tends to lead to fatigue. Reading and education are also topics of great scientific interest in their broader sense. A person who is tired is hardly capable of perceiving new information; tiredness reduces cognitive abilities including attention, concentration and a capacity for work. A reader becomes distracted when fatigue develops.
In this study the authors developed algorithms for estimating fatigue during reading based on eye tracking data collected from modern smartphones or tablet PCs. The described method could prove useful for evaluating reading and comprehension effectiveness in order to make predictions about the state of a person’s cognitive capabilities.
Reading Fatigue and its Significance for Assessing a Reader’s Cognitive State
Introduction
The main hypothesis in actual work was the suggestion that fatigue during reading correlates with eye movement parameters. Gaze patterns in reading continuously and dynamically change depending on a reader’s attention, how often he rereads previous sections of text to improve understanding.
Fatigue can't be measured in a moment via observing a few eye movement events. One instead should collect eye-movement data for long periods such as dozens of pages.
The actual study aimed to assess readers' fatigue via analysis of the eye-tracking data collected with the mass market device.
Fatigue could be especially crucial while reading long texts, official documents composed of multiple pages, educational materials, licenses, work instructions, and others.
We show that eye movement data collected during free reading reveals statistically significant trends in two main types of eye movement events, namely fixations duration and saccade amplitudes.
Aim
  • 20 respondents (8 males, 12 females, aged 25+4 years old) participated in the study.

  • A reader set in front of the Pad with the developed application Oken Reader [1].

  • The presentation of the texts was done at a comfortable distance from the respondent's eyes, approximately 50 cm (Fig. 1).

  • Texts were sufficiently long for the occurrence of fatigue escalation (40 min, close to the duration of an academic hour).

  • Eye movement data were recorded utilizing ARKit technology.

  • We used the tools described in our previous studies [1,2] to extract fixations and saccades from eye movement data.
Method
Reading in the Oken Reader application on iPad. The navigation was carried out by finger tapping on the screen.
Figure 1
The view of the screen with Oken Reader application and additional tools for managing a reading process (stop read, move to other document or book, close the application).
Figure 2
The example (Fig. 2) shows one page of solid text on the screen of the iPad. Embedded eye tracker (ARKit) used in the Oken Reader application allows to detect eye gaze. All the participants signed informed consent, got monetary reward in case of proper executing the instructions. The participants passed the test about their feeling good, all the data analyzed impersonally.
Fixations with durations outside the range (50, 1000) ms were not considered. Fixation durations and saccade amplitudes were normalized for each respondent.
References
xn=(x−mean(x))/std(x),
Then the dependency of these normalized values on time were fitted using linear regression (figures 3 and 4, fixation duration and saccade amplitude, respectively). We found that the slopes for both parameters were positive (for most of the respondents, 16 vs. 4) and significantly different from zero (one sample Wilcoxon test, n = 20, p-values are 0.0014 and 0.00017 for fixation duration and saccade amplitude, respectively).
a) Dependence of normalized fixation duration on a reading time. Every horizontal bar relates to particular text. Parameter also depends on the type of a text (for example green text is easier and fixation durations lower for it).
b) Dependence of normalized horizontal amplitude of saccade on a reading time also. Only horizontal saccades were included because only that type of events relates to a reading process. The similar trend is observed on both graphics.
Despite some texts strongly influencing the mean of normalized fixation duration (as well as an saccade amplitude), this parameter linearly increases during reading (Fig. 3).
Figure 3
Based on the obtained data, we can talk about the potential possibility of determining the state of fatigue during reading with assessing eye movements.
Described approach could be the ground for building a complex method for monitoring the reader’s state during his working with documents, books and papers: in case when he deals with a large texts. It could be useful to notice him about his actual functional state and to help him in the situation when he needs the adjustment of his attention and concentration. Could be that in the current moment it is necessary to relax and restore cognitive resources for a reader.
The method could be used in a number of cases: a) in analytics related to reading big amount of official documents, b) in reading difficult technical documentation, c) for users Smartphones and Tablet PC’s who like to read a lot and are interested in getting objective feedback about a dynamics of interaction with his device in terms of perceiving verbal information on the screen. The last one could be a kind of tracking routine for a reader by himself like in the case of fitness trackers and other self management routines.
Discussion
To clarify the factor that leads to a shift in the average distribution, additional analysis is required (a number of factors can influence the distribution skew and consequently change to the approximation line angle). In addition, it is necessary to check the repeatability of the results and the versatility of application of the proposed method to a larger number of different texts.
Limitations
We thank Oken Technologies, Inc. company on the basis of which the research was carried out. We also thank Arsen Revazov for the foundation and fruitful advice to our work.
Acknowledgements
Anisimov, V., Сhernozatonsky, K., Pikunov, A., Raykhrud, M., Revazov, A., Shedenko, K., Zhigulskaya D. & Zuev, S. (2021). OkenReader: ML-based classification of the reading patterns using an Apple iPad. Procedia Computer Science, 192, 1944-1953.

Anisimov, V., Chernozatonsky, K., Pikunov, A., Shedenko, K., Zhigulskaya, D., & Arsen, R. (2021, September). ML-based classification of eye movement patterns during reading using eye tracking data from an Apple iPad device: Perspective machine learning algorithm needed for reading quality analytics app on an iPad with built-in eye tracking. In
2021 International Conference on Cyberworlds (CW) (pp. 188-193). IEEE.
References
© 2022 Oken Technologies Inc.