EFG - 04

Fluent transcription – the ability to perform the motor, spelling and syntactic operations necessary to construct a sentence without undue hesitation – is important for maintaining students’ motivation and skill development. As importantly, where and when a young writer pauses provides valuable information about what they have and have not learned: An accurate completed sentence can hide significant underlying bottlenecks.

Teachers in typical classrooms cannot systematically monitor students writing processes. Writing timecourse data, from keyboard or writing tablet, offers potential for automatic extraction of important diagnostic information, thus providing earlier and more precise identification of writing difficulties.

This potential will not be realised, however, without step-change in the computational and statistical methods used to interpret writing timecourse data. These present at least four challenges. (1) Statistical inference from writing timecourse data must handle complex multilevel and multi- distributional data. (2) Real-time handwriting traces require extensive pre-processing and this is difficult to perform automatically.(3)Routine collection of keystroke or digitised handwriting gives for “big data”, with associated analytical challenges (and potential benefits). (4) Computational- linguistic methods for syntactic markup must handle noisy and fragmented text.

The proposed project will bring together field experts whose current work shows promise for providing solutions to one or more of these problems. Our goals in the first two years of the project will be to

• Explore the potential of various approaches to the analysis of writing time course data, detailed in the next section, that go substantially beyond current state of the art.

• Give educational researchers access to these new developments in the form of published worked examples.

Approaches will be linguistically / psycholinguistically informed. Outputs will be open-access / open-source, and accessible to literacy researchers without advanced statistical expertise. The project is broad and exploratory, and hence risky. However, we guarantee disseminating an honest summary of our successes and failures.