The recent surge of interest in explainability in artificial intelligence (XAI) is propelled by not only technological advancements in machine learning, but also by regulatory initiatives to foster transparency in algorithmic decision making. In this article, we revise the current concept of explainability and identify three limitations: passive explainee, narrow view on the social process, and undifferentiated assessment of understanding. In order to overcome these limitations, we present explanation as a social practice in which explainer and explainee co-construct understanding on the microlevel. We view the co-construction on a microlevel as embedded into a macrolevel, yielding expectations concerning, e.g., social roles or partner models: Typically, the role of the explainer is to provide an explanation and to adapt it to the current level of understanding of the explainee; the explainee, in turn, is expected to provide cues that guide the explainer. Building on explanations being a social practice, we present a conceptual framework that aims to guide future research in XAI. The framework relies on the key concepts of monitoring and scaffolding to capture the development of interaction. We relate our conceptual framework and our new perspective on explaining to transparency and autonomy as objectives considered for XAI.

Decontextualized talk is assumed to be used only rarely when children are younger than 30 months. Motivated by Bühler's (1934/1999) linguistic theory that describes different dimensions of (de-)contextualization, we provide evidence that this kind of input can already be found in caregivers’ talking to their 12-month-old children. Such early input is characterized by being decontextualized on some dimensions while being grounded in the immediate context on others. In this way, parents may scaffold understanding of talk about the there-and-then. We also examined whether caregivers adapt decontextualized verbal input to individual trajectories in language development. We observed 59 parent–child interactions within a decorated room when children were 12 months old, and assessed the children's linguistic development at 12 and 24 months of age. However, we did not find differences in the input directed toward children with different trajectories in language development.

Pointing is one of the first conventional means of communication and infants have various motives for engaging in it such as imperative, declarative, or informative. Little is known about the developmental paths of producing and understanding these different motives. In our longitudinal study (N = 58) during the second year of life, we experimentally elicited infants’ pointing production and comprehension in various settings and under pragmatically valid conditions. We followed two steps in our analyses and assessed the occurrence of canonical index-finger pointing for different motives and the engagement in an ongoing interaction in pursuit of a joint goal revealed by frequency and multimodal utterances. For understanding the developmental paths, we compared two groups: typically developing infants (TD) and infants who have been assessed as having delayed language development (LD). Results showed that the developmental paths differed according to the various motives. When comparing the two groups, for all motives, LD infants produced index-finger pointing 2 months later than TD infants. For the engagement, although the pattern was less consistent across settings, the frequency of pointing was comparable in both groups, but infants with LD used less canonical forms of pointing and made fewer multimodal contributions than TD children.

In explanatory interactions, explainees are expected to continuously provide feedback to explainers by signaling whether they understand an ongoing explanation. The study presented in this paper is based on the hypothesis that explainees use a set of multimodal cues, including vocalizations, facial expressions, and movements of the torso, head, and hands, to do so. We test this hypothesis by building a random forest classifier based on a multimodal corpus of dyadic explanations (21 explainers and explainees), in which windows of understanding or non-understanding were identified by participants in a retrospective video recall task. Results show that sequences of understanding can indeed be differentiated from those of non-understanding, and that a diverse set of predictors covering a wide range of modalities contributes to this classification. Due to data sparsity and a high degree of individual variation, the generalizability of our results is currently limited, but they support our hypothesis of the relevance of multimodal display in explanatory interactions.

Everyday explanations are interactive processes with the aim to provide a less knowledgeable person with reasonable information about other people, objects, or events. Because explanations are interactive communicative processes, the topical structure of an explanation may vary dynamically depending on the immediate feedback of the explainee. In this paper, we analyse topical transitions in medical explanations organised by different physicians (explainers) related to different forms of multimodal behaviour of caregivers (explainees) attending an explanation about the procedures of an upcoming surgery of a child. The analyses reveal that explainees’ multimodal behaviour with gaze shifts (and particularly gaze aversion) can predict a transition from an elaborated topic to a new one, whereas explainees’ forms of multimodal behaviour with static gaze cannot be related to changes of the topical structure.