Keystroke logging is an objective and scalable methodology that has become the gold standard in writing research for modeling writing processes. A particularly significant aspect of this analysis is the examination of features such as pause duration, as pauses are regarded as indicators of underlying cognitive processes. Traditionally, arbitrary pause thresholds that are universally applied to all writers have been established to differentiate between cognitive and non-cognitive pauses. However, this approach presents considerable limitations and fails to account for the complexity and individual variability inherent in the cognitive processes involved in text production. Furthermore, different scholars employ varying approaches to the calculation of between-word pauses. This study is the first to analyze keystroke logs of Russian typed texts utilizing Gaussian mixture models (GMM) to cluster pause duration values at between-word boundaries. By employing keystroke logs collected from 50 university students who described the views from their home windows, we conducted a cluster analysis of pause duration values before words, after words, and between words separately. It was determined that the distribution of pauses between words cannot be characterised by a single distribution. For the majority of participants, two-component distribution provided the best fit for all three types of pauses. Additionally, we observed a high degree of individual variability in the mixing proportions of different components. This paper underscores the necessity of avoiding the imposition of fixed thresholds in pause analysis that are universally applicable to all writers and advocates for individualized and holistic approach to studying the writing process.