← All reviews

SituFont: A Just-in-Time Adaptive Intervention Interface for Enhancing Mobile Readability in Situational Visual Impairments

Jingruo Chen, Kexin Nie, Mingshan Zhang, Chun Yu, Zhiqi Gao, Kun Yue, Yuanchun Shi, Chen Liang · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3791020

Summary

SituFont is a smartphone reading interface that treats situational visual impairment (SVI) as a first-class design target and dynamically adjusts font parameters in response to real-time context. The authors argue that static accessibility settings and one-off manual adjustments cannot keep up with the way mobile reading conditions actually change: lighting swings from a 50,000 lux street to a dim bedroom, users read on subway vibrations or while walking, and personal factors like fatigue and distraction fluctuate within a single session. Building on Sears et al.'s Situationally-Induced Impairments and Disabilities framework and the just-in-time adaptive intervention (JITAI) paradigm from health behaviour research, SituFont is designed through a three-study sequence. Study 1 is a semi-structured interview with 15 participants (including people with myopia, astigmatism, presbyopia, and normal vision) that maps the space of environmental, personal, and information factors users recognise as impairing their mobile reading. Study 2 is a controlled experiment in which 18 participants read isomorphic Chinese passages across six motion/light conditions (indoor/outdoor x standing/walking/running) while adjusting font size, weight, line spacing, and character spacing to their preferred level - providing population-level priors for a regression model. The resulting system sits on an Android phone and a cloud backend, uses the front camera (for reading distance via MediaPipe face landmarks), light sensor, and accelerometer, and organises context into a three-layer label tree (movement/posture, environmental scene, personalised self-report). A long-press gesture triggers an ML-driven font suggestion; users can accept or override, feeding the human-in-the-loop update. A comparative user study with 12 Mandarin readers across eight simulated SVI scenarios (combinations of intense brightness, high vibration, distraction, and fatigue) benchmarks SituFont against a competitive manual-adjustment baseline where users first tuned the static interface to their preferred configuration.

Key findings

SituFont significantly improved reading goodput (characters read correctly per minute) over the manually optimised baseline in six of eight SVI scenarios, with the largest effects in multi-factor conditions: intense brightness + high vibration + distraction (Cohen's d = 1.63) and fatigue-dominated conditions (d = 2.92). The two single-factor scenarios (intense brightness alone, high vibration alone) showed only marginal trends, suggesting that manual tuning can approximate a good static configuration when one impairment dominates, but breaks down as factors compound. Comprehension accuracy was statistically indistinguishable between systems, confirming the speed gains did not come at the cost of understanding. NASA-TLX showed significantly lower mental demand (distraction, intense brightness + high vibration, fatigue conditions) and physical demand (intense brightness, intense brightness + high vibration, high vibration), and lower effort in distraction, intense brightness + high vibration + distraction, and fatigue. Study 2 quantified the typographic mechanism: font size correlated positively with vibration and motion (r up to 0.4, p<0.01), and font weight correlated positively with light intensity (r = 0.37, p<0.01) - users asked for bigger, bolder text as conditions degraded, but character spacing changed little. On UEQ and SUS, SituFont was rated significantly more efficient, supportive, inventive, leading-edge, and easier to use; only initial learning effort was rated slightly higher than the baseline. Qualitative feedback highlighted that participants valued automatic adjustment most in cycling, running, and subway contexts, and flagged unintended long-press activations and post-adjustment vibrations as usability issues to fix.

Relevance

SituFont is one of the more rigorous recent demonstrations that accessibility engineering should extend beyond permanent impairments to the situational impairments that affect almost everyone. It provides practitioners with three concrete takeaways: (1) the adaptation space that matters for mobile reading is multi-parameter - font size, weight, line spacing, and character spacing together - and tuning only size or only zoom is insufficient when multiple SVI factors co-occur; (2) human-in-the-loop ML that starts from a population prior and refines per user is a workable pattern for accessibility personalisation without a long cold-start period; (3) activation design is a first-class accessibility concern - long-press gestures interrupted reading flow for some users, and interventions that fire unpredictably can themselves become an impairment. Important limitations for practice: the study used native Chinese readers reading Chinese text, and findings may not transfer unchanged to Latin scripts, right-to-left scripts, or long-form reading. Participants were young adults, not older adults with age-related visual decline. The binary yes/no personalisation signal is coarse. Extending the approach to screen-reader users, low-vision readers with diagnosed conditions, and other scripts would sharpen its accessibility impact.

Tags: situational visual impairment · situationally induced impairments and disabilities · adaptive typography · just-in-time adaptive intervention · human-in-the-loop · mobile accessibility · readability · context-aware computing · machine learning · personalization · Chinese text