Measuring non-linearity of multi-session writing processes

When (professional) authors work on their texts, they frequently 'jump' around their document to make textual changes and create new content at a wide range of locations. Currently, a range of linearity measures are available to capture this, some of which requiring time-intensive manual coding. Linearity metrics are commonly calculated based on the leading edge and are mostly used for short texts and single writing sessions. However, especially for longer, multi-session writing processes, text can often be created at various spaces, not necessarily including the leading edge. Accordingly, the leading edge is not enough to distinguish between linear production and non-linear text alterations. Therefore, in the current study, we propose a novel, more flexible, automatized non-linearity analysis, which does not solely rely on the leading edge. In this approach, all backwards and forwards cursor and mouse operations from the point of utterance are extracted from keystroke data, and characterized both based on duration and distance. This results in a detailed list of characteristics per writing episode, allowing us to compare and group episodes of writing at various scales. We illustrate this approach by analysing the writing process of a complete novel based on close to 400 writing sessions totalling 276 h of writing. The results show that the current non-linearity analysis allows us to successfully cluster writing sessions using the non-linearity characteristics. This analysis can be used to find patterns in non-linearity over time, allowing us to chart interactions with the text-produced-so-far and session management strategies in multi-session writing.