Analyzing the Impact of Cognitive Load in Evaluating Gaze-based Typing

Gaze-based virtual keyboards provide an effective interface for text entry by eye movements. The efficiency and usability of these keyboards have traditionally been evaluated with conventional text entry performance measures such as words per minute, keystrokes per character, backspace usage, etc. However, in comparison to the traditional text entry approaches, gaze-based typing involves natural eye movements that are highly correlated with human brain cognition. Employing eye gaze as an input could lead to excessive mental demand, and in this work we argue the need to include cognitive load as an eye typing evaluation measure. We evaluate three variations of gaze-based virtual keyboards, which implement variable designs in terms of word suggestion positioning. The conventional text entry metrics indicate no significant difference in the performance of the different keyboard designs. However, STFT (Short-time Fourier Transform) based analysis of EEG signals indicate variances in the mental workload of participants while interacting with these designs. Moreover, the EEG analysis provides insights into the user’s cognition variation for different typing phases and intervals, which should be considered in order to improve eye typing usability.

💡 Research Summary

The paper investigates how different layouts of gaze‑based virtual keyboards affect users’ cognitive load, measured via EEG, and compares these findings with traditional text‑entry performance metrics. Eye‑typing, which allows users to compose text by fixating on letters, is especially valuable for individuals with severe motor impairments. While prior work has evaluated gaze keyboards mainly through speed‑related measures such as words‑per‑minute (WPM), keystrokes‑per‑character, and backspace usage, the authors argue that these metrics overlook the mental effort required to coordinate eye movements, visual search, and language prediction.

Keyboard Designs
Three keyboard prototypes were built:

• Keyboard A – a conventional layout with a single line of word predictions placed above the key matrix.
• Keyboard B – predictions inter‑spaced between the rows of keys, aiming to reduce the distance between letters and suggestions.
• Keyboard C – combines the top‑line predictions of A with additional per‑key suggestions displayed directly on each key, thereby integrating prediction into the focal area of the user’s gaze.

The designs differ primarily in how they demand visual attention shifts: A requires a look‑away to a separate prediction bar, B introduces more frequent but smaller shifts, and C tries to keep the prediction within the foveal region.

Methodology
Five male participants (ages 22‑26, no prior eye‑typing experience) took part in three separate sessions, each dedicated to one keyboard. Each session comprised five sentences taken from a standard phrase set; participants typed each sentence by gazing at letters and confirming with a physical space‑bar. Eye movements were captured with an SMI REDn tracker (60 Hz) and brain activity with an Emotiv EPOC+ (14 channels, 128 Hz). All streams were synchronized via Lab Streaming Layer.

EEG data were processed using a Short‑Time Fourier Transform (STFT) pipeline: signals were segmented into 8‑second windows (1024 samples) with 50 % overlap, a half‑cosine window was applied, and a discrete Fourier transform computed the spectrogram. Frequency bands were defined as Delta (<4 Hz), Theta (4‑8 Hz), Alpha (8‑14 Hz), and Beta (≥14 Hz). Prior literature links higher Beta power to increased cognitive load, so the authors calculated the ratio of Beta‑band power to total power for each window and used the average ratio as a quantitative load indicator.

Traditional Performance Results

• WPM: A = 9.20, B = 8.60, C = 9.05 (ANOVA, p > 0.05).
• Keystrokes saved (reflecting prediction usefulness): 39.0 % (A), 35.4 % (B), 33.5 % (C) – not significant.
• Backspace usage: 2.92 (A), 6.32 (B), 5.00 (C) – also non‑significant.

Thus, from a conventional standpoint, the three keyboards performed similarly.

Cognitive Load Findings
The Beta‑power ratio revealed clear differences. Keyboard C showed the lowest mean ratio (0.0824), while A and B recorded higher values (0.0865 and 0.0860 respectively). Pairwise t‑tests confirmed that C’s load was significantly lower than both A (p = 0.015) and B (p = 0.00047). Further analysis broke the data into typing phases (prediction selection, character entry, correction) and sentence order. Keyboard C consistently exhibited reduced load, especially during the prediction‑selection phase, suggesting that embedding suggestions on the keys reduces the need for gaze shifts and thus eases mental processing.

Interpretation and Implications
The study demonstrates that layout decisions impact users’ mental workload even when they do not affect observable speed or error rates. By keeping predictive information within the foveal region, Keyboard C minimizes the attentional “cost” of switching between the keyboard and a separate suggestion bar. This reduction is captured by lower Beta activity, which is widely accepted as a marker of task‑related cognitive demand. Consequently, designers of eye‑typing systems should incorporate cognitive‑load measurements alongside traditional metrics to obtain a holistic view of usability.

Limitations

• Small, homogeneous sample (five young adult males) limits generalizability.
• The Emotiv device, while affordable, yields noisier signals than clinical‑grade EEG, possibly affecting the precision of spectral estimates.
• Only the Beta band was used as a load indicator; complementary information from Alpha or Theta bands was not explored.
• No subjective workload assessments (e.g., NASA‑TLX) were collected to triangulate the EEG findings.

Future Directions
The authors suggest expanding participant diversity, employing higher‑resolution EEG or MEG, integrating subjective questionnaires, and exploring real‑time load monitoring to adapt keyboard behavior dynamically. Multi‑modal approaches (eye‑tracking + EEG + physiological signals such as GSR) could further refine the understanding of cognitive demands in gaze‑based interaction.

Conclusion
The paper provides empirical evidence that cognitive load, as measured by EEG Beta‑band power, varies significantly across different gaze‑keyboard designs even when conventional performance metrics do not. Keyboard C, which embeds predictions directly on keys, yields the lowest mental effort, highlighting the importance of considering neurophysiological feedback in the design and evaluation of eye‑typing interfaces. This work paves the way for more user‑centric, cognitively aware gaze‑based input systems.