LitTerra, Bukvik, and ColaboFlow - An AI augmented workflow platform for "solidifying" the hybrid creative processes in Digital Humanities

In this paper, we evaluate usage of workflows in the domain of digital humanities (DH), primarily in the hybrid (a fusion of distant and close reading) multi-lingual reading practices. We present the ColaboFlow (a workflow framework) in the context of LitTerra platform that supports this kind of DH practices. ColaboFlow is integrated in LitTerra through its specialization Bukvik. In other words, Bukvik is an instance of ColaboFlow that supports common research workflows addressing distant reading, stylometry and parallel multilingual corpora. Through practical context of LitTerra and Bukvik, we explore the requirements ColaboFlow as a workflow framework and platform needs to provide both from the perspective of reproducible DH research and from the perspective of the creative processes digitalization. We argue that workflows as inherent enablers of reproducible research are critical for keeping human-in-the-loop or rathet \emph{human-on-the-loop}, where in the addition to the human-in-the-loop there is an aim of keeping them at the top of human+AI symbiotic research and decision making process. We presentx and argue the key features in ColaboFlow and LitTerra that contributes to abovementioned goals. Having that identified, we address the process of \emph{"solidification"} which transforms the declarative creative process (or work) into a solid, executable, and reproducible workflow. We augment solidification with semantically decorated workflow artefacts (tasks and datasets primarily) and with the help of generactive AI. The solidified workflows are placed in the ColaboFlow framework and evaluated how they can be used in the LitTerra platform. We argue that the presented approach helps keeping human in the loop, it documents and clarifies the creative process and work, and enables the trasparency of the created workflows, and the possibility of ongoing collaborative and AI interventions on them, and their reproducible execution in distributed environment. Finally, we evaluate the solidification process and the ColaboFlow framework through both human generated data and partially sintetic data generated from human-generated data.