Esteves, A., Bakker, S., Antle, A., May, A., Warren, J. and Oakley, I. 2015. The ATB Framework: Quantifying and Classifying Epistemic Strategies in Tangible Problem-Solving Tasks. In Proceedings of the 9th International Conference on Tangible, Embedded and Embodied Interaction (TEI ’15). ACM, New York, NY, USA, 13-20. [download]
The ATB (Artifact, Tool and Body) framework contributes to our understanding of how epistemic actions are used in human problem-solving tasks, providing researchers with a video-coding tool to more systematically assess this complex type of behavior in tangible interaction. In terms of HCI, this tool has two objectives. Firstly, it is intended as a mechanism to evaluate tangible systems in terms of the type, diversity and appropriateness of the epistemic actions they support, and in terms of the impact these actions can have on more traditional metrics such as performance time or errors. Secondly, in the long term, we argue that a series of such evaluations will result in a corpus of knowledge describing the use of epistemic actions in real tasks. This data can be used as the basis for grounded, practical design knowledge on how to create novel systems that truly support epistemic actions, and thus, improve our ability to design tangible interaction that is natural and meets the real needs of the user.
The ATB framework builds on the action classification framework presented by Antle et al. In their work, an action can be classified as either a direct placement (DP), an indirect placement (IP), or as exploratory (EXP). In the puzzle task they studied, a DP action corresponded to those situations where users already know where to place a piece before picking it up, leading to a fast and direct transition between acquiring a piece, moving to the final destination and correctly placing it. IP represented similar outcomes but described situations in which users are not initially certain of where to position the pieces they pick up. As such, they translated or rotated the piece while searching for its correct destination. Finally, EXP represented those actions where pieces do not end in their final and correct position. As with IP, if these intermediary actions make the task easier for the user they are considered epistemic (e.g., if a user organizes pieces into different piles for subsequent identification and retrieval).
The ATB framework was developed through an extensive literature review with the goal of capturing a wide range of epistemic activity descriptions. A set of keywords was used to conduct a literature search on both Google Scholar and Science Direct. The search terms were ‘epistemic action(s)’, ‘complementary action(s)’ and ‘complementary strategies’. The first 60 results from each of these searches were kept for further inspection. Additionally, papers referencing seminal work in the area and including the keywords defined above were also retained. Ultimately, 78 papers were obtained through this process – a typical number for meta-analysis papers in the area of HCI. Each paper was then inspected for any mention of actions that could be interpreted as epistemic, or were directly treated as epistemic, and quotes such as: “(…) preparing the workplace, for example, by partially sorting nuts and bolts before beginning an assembly task in order to reduce later search (…)” were extracted. These represented concrete examples of epistemic actions from research literature in a range of fields (such as mathematics, cognitive science, HCI and design) from the last three decades. [download publication list]
In total, 335 quotes were compiled. Two of the authors then worked collaboratively to create an affinity diagram that identified different clusters of epistemic actions. Quotes judged to depict actions with unclear epistemic value were discarded. This process led to the identification of 20 types of epistemic action based on a subset of 225 of the original quotes. These were then grouped by actions performed with 1) task artifacts (e.g. objects marked with fiducials), 2) tools (e.g. a pencil that can be used for annotations) or 3) the users own bodies. These categories are then used as the basis for classifying behaviors through video-analysis, according to the following procedure. Firstly, raters should categorize actions as being either DP, IP, or EXP, as in Antle et al’s framework. After this process is completed, raters should review each action that can contain epistemic activity (i.e. those coded as either IP or EXP) and match these to one (or more) of the 20 types of epistemic actions identified. [download ANVIL coding scheme]