Investigating Laboratory and Everyday Typing Performance of Blind Users
Hugo Nicolau, Kyle Montague, Tiago Guerreiro, André Rodrigues, Vicki L. Hanson · 2017 · ACM Transactions on Accessible Computing (TACCESS) · doi:10.1145/3046785
Summary
This paper presents the first longitudinal study comparing blind users' touchscreen typing performance in laboratory versus everyday real-world settings. Over 12 weeks, five novice blind smartphone users participated in eight weekly laboratory sessions while also having their everyday typing logged continuously on their personal iPhones running VoiceOver. The laboratory sessions involved transcription tasks using standardized phrases, while field data captured natural typing in messaging, search, and other daily activities. A key methodological contribution is a novel technique for computing "intent" from unconstrained field data — determining what users meant to type when there is no target phrase to compare against, using a combination of word similarity algorithms and manual verification. The study tracked detailed metrics including words per minute (WPM), error rates (uncorrected, corrected, substitutions, omissions, insertions), touch behaviours (landing position, exploration distance, interkey interval), and learning rates across both settings. All participants used the default iOS QWERTY keyboard with VoiceOver's explore-then-lift typing method, where users slide their finger across the screen to hear letters announced via text-to-speech and lift to select. The study collected over 2,100 field trials and data from laboratory sessions, enabling a rich comparative analysis of how blind typing performance differs between controlled and naturalistic contexts.
Key findings
Participants were 1.3 to 2 times faster typing in everyday settings compared to laboratory sessions, averaging about 6 WPM in the field versus 4 WPM in the lab by week 8. However, real-world typing was significantly less accurate, with uncorrected error rates above 7% in the field compared to approximately 1% in the laboratory. This speed-accuracy trade-off is a critical finding — laboratory studies systematically underestimate blind users' typing speed while overestimating their accuracy. Substitutions (incorrect characters) were the dominant error type in both settings and had the greatest impact on entry rates. Participants spent approximately 13% of their total typing time on corrections, and one-third of those corrections were counterproductive — deleting correct characters. The interkey interval was significantly shorter in everyday typing (592ms vs 1060ms), suggesting participants paused less between keystrokes during natural tasks. Over 12 weeks, all participants improved in both speed and accuracy, though learning was slow (approximately 0.2-0.3 WPM improvement per week). Importantly, no participants used autocorrect, autocomplete, cursor positioning, or copy/paste features, possibly due to the difficulty of performing these operations non-visually. 68% of omission errors went undetected and uncorrected.
Relevance
This research has profound implications for how the accessibility community evaluates and designs mobile text input for blind users. The finding that laboratory studies produce a skewed picture of real-world performance — slower but more accurate than actual everyday use — means that keyboard designers relying solely on lab evaluations may be optimizing for the wrong trade-offs. The design implications are directly actionable: keyboards need easier and more efficient correction mechanisms for non-visual use; autocorrect should be adapted to handle substitution errors based on adjacent key distances rather than assuming systematic directional offsets; touch models should leverage landing and movement data (not just lift positions) to predict intended keys; and language-based solutions like spellchecking are essential for the 68% of omission errors that go undetected. For researchers, the intent-computation methodology for analysing field typing data without target phrases is a valuable contribution enabling future in-the-wild studies of accessible text entry.
Tags: visual impairment · text entry · touchscreen · mobile accessibility · longitudinal study · in-the-wild · VoiceOver · input performance