Using keystroke logging to explore differences in written language production processes between self-experienced and invented narrative accounts: A forensic linguistic approach

This study aims at investigating if/how writing processes, such as planning and revision, differ between accounts of self-experienced and invented narratives. The underlying assumption is that cognitive load will increase for the writer when s/he is is changing or inventing parts of an otherwise self-experienced series of events. This builds on theories of how limited working memory capacity leads to increased pausing behavior in accordance with increased cognitive demands (Kellogg, 1996; McCutchen 2000), and that the need for revisions will increase when the writer wants to meet the goal of convincing the reader that something is true (cf. the relation between planning, translating and revision described in e.g. the writing model of Hayes and Flower, 1980).
This presentation primarily focusses on comparing written accounts, collected with an experimental design. Participants (n=45) were presented with 4 elicitation videos, depicting misdemeanors (e.g., cheating on an exam, stealing a bike). Each participant performed 4 accounts across the 4 films: two written, and two spoken. For one account in each modality the participant is asked to lie and alter “who did it”. Modality, films and invented/self-experienced accounts are balanced for order. The written data was collected online through keystroke logging (ScriptLog). The participants repeated the experiment 4 times with 2 weeks apart, to allow for comparisons of consecutive accounts of both invented and self-experienced narratives.
The first results showed no differences between time on task between invented and self-experienced narratives, but the invented narratives required overall more pause time. In addition, time on task and overall pause time decreased over the consecutive accounts, indicating that the retelling task became easier independent of condition. There were no overall differences in the amounts of deleted text between the conditions, but during the writing of the invented narratives, less characters were written between pauses, indicating the need to pause (and plan?) more often. The general picture is that there are many individual differences, and that individual baselines may need to be established, as well as including comparisons within subjects in the further explorations of the data.
Continuing analyses will look more closer at the linguistic contexts where the writers need to pause and revise, and will also compare the written accounts to spoken equivalents. The overall picture is however that using keystroke logging to investigate “true” and “false” narratives may be a rewarding avenue for forensic linguistics, and could be used (in addition to other tools) to identify instances where information needs to be further investigated.