Indirect Text Entry Using One or Two Keys
Melanie Baljko, Andrew Tam · 2006 · Proceedings of the 8th International ACM SIGACCESS Conference on Computers and Accessibility (Assets '06) · doi:10.1145/1168987.1168992
Summary
This paper introduces a new descriptive model for indirect text entry — text composition methods where the number of input devices (switches) is significantly smaller than the number of selectable characters. This is the primary text entry paradigm for users of augmentative and alternative communication (AAC) devices, particularly those with severe motor impairments who may only be able to operate one or two switches (buttons, puffs of breath, eye blinks, or EMG signals). The paper's key theoretical contribution is the containment hierarchy (CH), a formal model based on directed acyclic graphs that generalises how indirect text selection interfaces organise characters into groups for scanning. In scanning, groups of characters are highlighted sequentially; the user activates a switch to select the highlighted group, then sub-groups within it are highlighted, continuing until a single character is reached. The paper demonstrates a novel application of Huffman coding to derive optimal containment hierarchies that minimise mean encoding length — the average number of switch activations needed to select a character, weighted by character frequency. Two variants are compared empirically: Row-Column scanning (the conventional approach with a fixed grid layout) and a Huffman-coded variant (where character groupings are optimised based on letter frequency). Twelve non-disabled participants typed target text using each variant across four sessions, with two dwell periods (750ms and 1250ms).
Key findings
Contrary to the theoretical prediction that Huffman coding would yield faster entry rates due to shorter mean encoding lengths, the Row-Column variant was consistently and significantly faster in practice. Row-Column started at 2.048 words per minute (wpm) and reached 3.075 wpm after four sessions, while Huffman started at 1.445 wpm and reached 1.809 wpm. The Row-Column variant's learning curve closely followed the power law of learning (R² = 0.96), while the Huffman variant did not (R² = 0.79), suggesting users were still in early learning stages after four sessions. Error rates were significantly higher for Huffman (18.50%) than Row-Column (6.60%). User feedback confirmed the finding: participants described the Huffman variant as "torture" and noted "you always have to be looking," citing its unpredictable scanning pattern and high cognitive load. The key insight is that the predictability of the scanning pattern matters enormously — the Row-Column layout's regular, predictable structure allows users to anticipate where focus will move, reducing cognitive load even though the theoretical encoding is suboptimal. Shorter dwell periods (750ms) produced faster entry but more errors. A hybrid variant combining Row-Column's predictable structure with Huffman's frequency-based encoding was proposed as a compromise.
Relevance
This paper provides essential evidence for AAC device designers: theoretically optimal interfaces are not necessarily usable. The Huffman variant's failure despite its mathematical optimality demonstrates that human factors — cognitive load, pattern predictability, learnability — can completely override information-theoretic advantages. For practitioners designing switch-based interfaces, the containment hierarchy model offers a rigorous framework for comparing and optimising scanning layouts. The finding that predictability trumps encoding efficiency has broad implications for any interface designed for users with limited input capabilities: reducing the cognitive burden of the interaction pattern may be more important than minimising the number of actions required. The study also highlights methodological concerns — the participants were non-disabled undergraduates, and the authors acknowledge that evaluation with actual AAC users who have physical disabilities is essential but challenging.
Tags: AAC · switch access · text entry · scanning · Huffman coding · information theory · speech generating devices · motor impairment