Text entry for mobile devices using ad-hoc abbreviation

Abstract

This paper presents a new method for improving the number of keystrokes and time required for text entry on mobile devices using ad-hoc abbreviations. The approach is easy-to-use because: users are not required to learn any pre-defined abbreviation rules; abbreviated input phrases are automatically detected and expanded; and it is possible to recover words that may be omitted from phrases either by accident or intention. The paper develops algorithms to detect abbreviated phrases using a Support Vector Machine trained on abbreviation examples and to expand abbreviations into complete phrases using a Hidden Markov Model learned from a text corpus. The abbreviation detector was evaluated on 3,000 word-abbreviation pairs and achieved 90% accuracy. The abbreviation expander was evaluated on 100,000 phrases and achieved 95% accuracy. A user study with 10 participants was performed to measure time and keystroke savings of the new approach compared to the existing iPhone® text entry system. Keystroke savings were consistent amongst users, with an average decrease of 32%. Time for input varied considerably depending on familiarity with the approach, increasing for novice users. However, experienced users achieved an average time saving of 26%. Observations suggest that novice users were spending time thinking about how they wanted to abbreviate words.