There Is More to Dwell Than Meets the Eye: Toward Better Gaze-Based Text Entry Systems With Multi-Threshold Dwell

Research Background and Issues

• Identified Problems or Challenges: Traditional gaze-based dwell text input systems face a speed bottleneck, typically around 20 WPM (words per minute), and require extensive training to achieve this speed. Additionally, shorter dwell thresholds may lead to frequent false activations (Midas Touch problem), while longer dwell times reduce input efficiency. The literature has yet to systematically explore the factors contributing to this speed limitation or how design innovations can overcome this cap.

• Significance: Gaze-based dwell text input technology is a touchless input method widely used in assistive technology. However, its speed bottleneck limits user experience improvements and broader adoption. Optimizing this technology not only enhances input efficiency but also serves a wider audience in need of touchless solutions.

• Research Motivation and Related Work:

  • Previous studies have primarily focused on adjusting dwell time thresholds to improve speed, but these solutions require extensive user training and have limited effectiveness.
  • Other research has explored reducing false activations, such as dynamically adjusting dwell thresholds or predicting the user's target key.
  • The authors argue that existing technologies have not fully explored the components of dwell selection to optimize the overall input performance of gaze-based dwell systems.

Solution

• Proposed Methods or Solutions:

  • The authors proposed two innovative dwell keyboard designs: the Dual-Threshold Dwell (DTD) keyboard and the Multi-Threshold Dwell (MTD) keyboard.
  • The DTD keyboard introduces a dual-threshold dwell mechanism, where a second selection of the same key requires a longer dwell time to reduce the likelihood of "false clicks."
  • The MTD keyboard further optimizes the DTD design by reducing dwell time thresholds for a few predicted target keys, while providing visual cues (e.g., magnification and highlighting of key targets) to help users quickly select their intended key.

• Innovations:

  • Designing dwell times with multiple thresholds instead of a fixed single time, thereby reducing false activations while improving input speed.
  • Utilizing predictive algorithms to highlight the most likely target keys and assign them shorter dwell times, significantly enhancing input efficiency.
  • Conducting a systematic and granular performance optimization by analyzing different components of gaze dwell time (e.g., pointing time, exit time).

• Implementation Steps and Key Techniques:

  • In the DTD design, the first character selection uses a 300 ms dwell threshold, while an immediate subsequent selection of the same character requires a 500 ms dwell.
  • In the MTD design:
    • Predictive algorithms mark and magnify the three most likely target keys, assigning them a dwell threshold of 200 ms.
    • The first character and non-predicted keys use a 300 ms dwell threshold.
    • The spacebar is assigned a shorter dwell threshold of 100 ms.
  • The design and evaluation experiments were conducted in a virtual reality environment, using gaze-tracking devices to record user input behavior and related data.

Research Outcomes

• Specific Results:

  • The MTD keyboard achieved an average input speed of 18.3 WPM, significantly higher than the DTD's 15.3 WPM and the Constant Threshold Dwell (CTD) keyboard's 12.9 WPM.
  • The new designs also significantly reduced the rate of false activations (e.g., "double-clicking" the same character) and lowered the cost of user input corrections.

• Advantages Compared to Existing Solutions:

  • Without requiring extensive training, the MTD demonstrated the highest average gaze input speed to date in the literature for untrained users.
  • The effective dwell threshold used (233.9 ms) was shorter than the lowest dwell threshold recorded in previous studies (282 ms).

• Experimental or Evaluation Results:

  • Users of the MTD keyboard achieved the fastest input speed for untrained users in this field after only 30 short sentence input exercises (10-15 minutes).
  • Subjective survey results showed that the MTD outperformed in terms of user satisfaction (e.g., speed, interaction ease, physical and mental fatigue).

• Limitations and Future Directions:

  • Limitations:
    • Some users reported that the MTD keyboard's prediction mechanism could lead to false activations or increased cognitive load when adjacent keys were highlighted.
    • The current prediction algorithm, based on fixed parameters, has room for further optimization, such as incorporating auto-correction and more efficient language models.
  • Future Directions:
    • Employing more modern language models (e.g., large-scale language models) to further improve prediction accuracy.
    • Exploring input strategies that combine letter and word prediction while reducing the high dwell threshold for the first letter.
    • Investigating how visual cues and dynamic dwell time adjustments can further enhance the user experience of gaze-based input technologies.

This study adopts a hybrid innovative design, optimizing different components of dwell time to achieve significant performance improvements. It provides new insights and practical evidence for overcoming the speed limitations of gaze-based dwell technology, while also suggesting directions for further research in the field.