This study aims to explore the process of reading during writing. More specifically, it investigates whether a combination of keystroke logging data and eye tracking data yields a better understanding of cognitive processes underlying fluent and nonfluent text production. First, a technical procedure describes how writing process data from the keystroke logging program Inputlog are merged with reading process data from the Tobii TX300 eye tracker. Next, a theoretical schema on reading during writing is presented, which served as a basis for the observation context we created for our experiment. This schema was tested by observing 24 university students in professional communication (skilled writers) who typed short sentences that were manipulated to elicit fluent or nonfluent writing. The experimental sentences were organized into four different conditions, aiming at (a) fluent writing, (b) reflection about correct spelling of homophone verbs, (c) local revision, and (d) global revision. Results showed that it is possible to manipulate degrees of nonfluent writing in terms of time on task and percentage of nonfluent key transitions. However, reading behavior was affected only for the conditions that explicitly required revision. This suggests that nonfluent writing does not always affect the reading behavior, supporting the parallel and cascading processing hypothesis.

Keystroke logging has become instrumental in identifying writing strategies and understanding cognitive processes. Recent technological advances have refined logging efficiency and analytical outputs. While keystroke logging allows for ecological data collection, it is often difficult to connect the fine grain of logging data to the underlying cognitive processes. Multiple methodologies are useful to offset these difficulties. In this article we explore the complementarity of the keystroke logging program Inputlog with other observational techniques: thinking aloud protocols and eyetracking data. In addition, we illustrate new graphic and statistical data analysis techniques, mainly adapted from network analysis and data mining. Data extracts are drawn from a study of writing from multiple sources. In conclusion, we consider future developments for keystroke logging, in particular letter- to word-level aggregation and logging standardization.

Moment to moment, a writer faces a host of potential problems. How does the writer’s mind coordinate this problem solving? In the original Hayes and Flower model, the authors posited a distinct process to manage this coordinating—that is, the “monitor.” The monitor became responsible for executive function in writing. In two experiments, the current authors investigated monitor function by examining the coordination of two common writing tasks—editing (i.e., correcting an error) and sentence composing—in the presence or absence of an error and with a low or high memory load for the writer. In the first experiment, participants could approach the editing and composing task in either order. On most trials (88%), they finished the sentence first, and less frequently (12%), they corrected the error first. The error-first approach occurred significantly more often under the low-load condition than the high-load condition. For the second experiment, participants were asked to adopt the less-used, error-first approach. Success in completing the assigned task order was affected by both memory load and error type. These results suggest that the monitor depends on the relative availability of working memory resources and coordinates subtasks to mitigate direct competition over those resources.

Error analysis involves detecting, diagnosing, and correcting discrepancies between the text produced so far (TPSF) and the writers mental representation of what the text should be. The use of different writing modes, like keyboard-based word processing and speech recognition, causes different type of errors during text production. While many factors determine the choice of error-correction strategy, cognitive effort is a major contributor to this choice. This research shows how cognitive effort during error analysis affects strategy choice and success as measured by a series of online text production measures. Text production is shown to be influenced most by error span, that is, whether the error spans more or less than two characters. Next, it is influenced by input mode, that is, whether the error has been generated by speech recognition or keyboard, and finally by lexicality, that is, whether the error comprises an existing word. Correction of larger error spans is more successful than that of smaller errors. Writers impose a wise speed accuracy trade-off during large error spans since correction is better, but preparation times (time to first action) and production times take longer, and interference reaction times are slower. During large error spans, there is a tendency to opt for error correction first, especially when errors occurred in the condition in which the TPSF is not preceded by an auditory prompt. In general, the addition of speech frees the cognitive demands of writing. Writers also opt more often to continue text production when the TPSF is presented auditorially first.