Accelerating Text Communication via Abbreviated Sentence Input

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 6574–6588 August 1–6, 2021. ©2021 Association for Computational Linguistics 6574 Accelerating Text Communication via Abbreviated Sentence Input Jiban Adhikary1 jiban@mtu.edu Jamie Berger2 jamieberger16@gmail.com 1Michigan Technological University, Houghton, Michigan, USA 2Washington Leadership Academy, Washington, DC, USA Keith Vertanen1 vertanen@mtu.edu Abstract Typing every character in a text message may require more time or effort than strictly neces- sary. Skipping spaces or other characters may be able to speed input and reduce a user’s phys- ical input effort. This can be particularly im- portant for people with motor impairments. In a large crowdsourced study, we found workers frequently abbreviated text by omitting mid- word vowels. We designed a recognizer opti- mized for expanding noisy abbreviated input where users often omit spaces and mid-word vowels. We show using neural language mod- els for selecting conversational-style training text and for rescoring the recognizer’s n-best sentences improved accuracy. On noisy touch- screen data collected from hundreds of users, we found accurate abbreviated input was pos- sible even if a third of characters was omitted. Finally, in a study where users had to dwell for a second on each key, sentence abbreviated in- put was competitive with a conventional key- board with word predictions. After practice, users wrote abbreviated sentences at 9.6 words- per-minute versus word input at 9.9 words-per- minute. 1 Introduction Experienced desktop and touchscreen typists can often achieve fast and accurate text input by simply typing all the characters in their desired text. How- ever, for some users, such quick and precise input is difficult due to a motor disability. Such users may use a virtual touchscreen keyboard, but their touch locations may be slow and inaccurate, e.g. peo- ple with Cerebral palsy. Other users may need to click keys by pointing at them with a head- or eye- tracker and dwelling for a fixed time, e.g. people with amyotrophic lateral sclerosis (ALS). When a person’s typing is slow or inaccurate, word completions may provide more efficient in- put. Word completions predict the most probable words based on the current typed prefix. However, monitoring predictions carries a cognitive cost and may not always improve performance (Trnka et al., 2009). Further, monitoring predictions can be diffi- cult without visual feedback. Eyes-free text input can be slow for users who are visually-impaired (Nicolau et al., 2019), and even slower for users who are motor- and visually-impaired (Nel et al., 2019). Finally, eyes-free text input may be needed in future augmented reality (AR) interfaces where visual feedback is limited or non-existent (e.g. due to lighting or device limitations). In audio-only AR, it is still possible to type on an invisible virtual keyboard (Vertanen et al., 2013; Zhu et al., 2018). All these cases motivate our interest in explor- ing alternatives to conventional word completion. Here we investigate accelerating input by allowing users to skip typing spaces and mid-word vow- els. We decided to abbreviate …

Content truncated to 500 words