Mobile Typing with Intelligent Text Entry: A Large-Scale Dataset and Results

A large proportion of text is produced using mobile devices. However, very little research looks at the special characteristics of how this happens and, importantly, how it is affected by the design of the language model (LM). The operating systems of modern devices offer a number of LM-based intelligent text entry methods (ITEs) such as Autocorrection (AC) and Suggestion Bar (SB) to aid typing.It is not known how the keyboard and performance of the ITEs influences the typing strategies in the wild.LMs are operating system and language-specific, therefore, in this paper, we release and analyse a large-scale dataset of mobile typing in two languages: English (46755 participants) and Finnish (8661 participants). Typing data was collected with the participants' own iPhone and Android devices resulting a diverse data on the ITE method performance.By analysing the typing speed and the information on which letters and operations happen in each keystroke, we found that iPhone and Android devices encourage the use of two different typing styles. IPhone users used utilise mainly AC and are able to achieve the highest typing speed among the participants when the balance between AC accuracy and threshold to correct user error are adequately balanced. Android users prefer SB to avoid and correct typing errors instead of utilising AC. Especially Finnish participants, who had low ITE accuracy, used SB often to correct typing errors. To develop and evaluate LMs for typing applications, it is essential to study which factors affect the user. The typing dataset that we have prepared and opened for the public, allows for analysing the factors thus aiding the development of the most useful LMs.