Topic models help digital humanists explore large archives, but their outputs are sparse word lists that still require expert interpretation. Large language models (LLMs) can draft topic interpretations, yet their usefulness for end users and exploratory sensemaking remains unclear. We report a feasibility study on a German twentieth-century cultural heritage archive, Ortslinien (OL). Using BERTopic, we identified recurring topics and produced three German interpretations per topic: a peer-proofread digital humanities (DH) expert baseline, GPT-4o, and LLaMA-3. In an online survey (N = 111), participants rated interpretations on coherence, relevance to top words, informativeness, readability, interpretive value, and theme, and ranked the best overall match. GPT-4o led on most dimensions, while the human expert outperformed LLaMA-3 on interpretive value; disciplinary background moderated several judgments. We discuss implications for designing human–AI co-interpretation workflows for scalable, user-facing archive exploration.

Intelligent assistants (IAs) in virtual reality (VR) environments offer novel ways of interaction between humans and artificial intelligence (AI). Human-like conversational abilities change the way we interact with AI. In human communication, multimodal referential interaction, such as using eye gaze or pointing with deictic expressions, or by using descriptive speech, is essential for achieving a shared understanding of the space surrounding the interaction partners. By facilitating efficient and intuitive referential interaction among users and IAs, user engagement and overall effectiveness could be enhanced.This paper compares three referential interaction methods in VR to determine user preferences: (i) verbal communication, (ii) gaze, and (iii) pointing. We conducted a user study with 39 participants in VR, where information about scene objects had to be gathered using these different referential interaction methods. The results reveal that gaze performed well in terms of performance (both time and number of prompts) and usability measures (as assessed by the UEQ and NASA-TLX). While verbal interaction was less physically straining than pointing, it required greater mental effort. Although there were no significant differences in the overall preferences, a qualitative analysis indicated differences in individual inclination. Enriching the referential methods with feedback and enabling multimodal interactions are promising methods to improve usability and decrease task load further.

Figure 1: Illustration of familiarity in intelligent virtual agents: (A) picture of real human, (B) the familiar agent created from it, and (C) the unfamiliar agent created from a non-existent person with similar visual appearance, such as hairstyle, skin color, eyes color, or clothes.

In virtual reality (VR), using the hands instead of controllers has enormous potential for intuitive user interactions, e.g., direct grasping, touching, or moving virtual objects. In this context, the realism of the hands plays an important role in the body ownership illusion and the user’s feeling of presence. Yet, most applications use generic 3D hand models, so they do not include individual user characteristics, like wrinkles, moles, or skin features. In this work, we present a pipeline to include fully articulated, photorealistic, fully tracked hand scans for VR applications. Our pipeline captures photos of the hand from different perspectives simultaneously using a customized hand scanner. Using open-source photogrammetry software, a rigged 3D model of the hand is created and integrated into a Unity application. At last, we provide practitioners and researchers with a fast and easy way to integrate a user’s real hands into a VR application. Finally, we present guidelines for generating a fully articulated, personalized virtual hand model within 30 minutes from a set of 2D images.

The increasing relevance of artificial intelligence (AI) agents for people’s everyday professional and private lives leads to a great need for identifying efficient and user-friendly modalities for human-agent interaction. Such interaction usually takes place via text or voice. While chatbots and voice assistants each rely on only one modality, the focus of this paper is on asymmetric modality combinations. A user study was conducted in the context of information retrieval, and different input and output modalities were evaluated in a 2 × 2 within-subject design. It was found that efficient modality combinations are not always preferred and that usability factors, such as user control, have a greater influence on preference. For example, the combination of voice input and text output can prove to be very efficient, but at the same time be less preferred due to weaknesses in the user experience. Pure text-based interaction was found to be the preferred combination as it ensures a high degree of control and freedom for the user. Dialog-based applications for information retrieval should therefore use the text modality as a basis and supplement this with voice-based input and output options.

Medical virtual agents (VAs) hold great potential to support patients in achieving their health goals, especially at times or in regions where the demand for physiological and psychological therapy exceeds the capacity of medical services. To create an accepted complement to on-site diagnosis, treatment, and counseling, it is critical to understand the impact of factors such as the agent’s visual representation, behavior, and responsibilities on creating a trustworthy human-agent relationship. To gain insights into these factors, we conducted a systematic literature review including 59 papers on embodied VAs in the medical domain. Our review focused on the application fields and the role of VAs in medicine, as well as the technology used to display them. Using thematic analysis, we discuss our findings in terms of user preferences, as well as potentials and barriers faced in the interaction with medical VAs. Concerning the visual representation, the users’ wish for customization in terms of appearance and communication modalities was pointed out. It was also important that the agent’s information builds up on trustworthy sources, that they are motivating and adapted to the users’ knowledge. Finally, our results identify research gaps, in particular regarding the technological implementation and the use of artificial intelligence.

Avatars in virtual reality (VR) with fully articulated hands enable users to naturally interact with the virtual environment (VE). Interactions are often performed in a one-to-one mapping between the movements of the user’s real body, for instance, the hands, and the displayed body of the avatar. However, VR also allows manipulating this mapping to introduce non-isomorphic techniques. In this context, research on manipulations of virtual hand movements typically focuses on increasing the user’s interaction space to improve the overall efficiency of hand-based interactions.

Cognitive-physical exercises can reduce the progression of dementia. However, traditional methods often induce problems (e.g., lack of motivation), whereas the success of recent virtual reality (VR) exergames such as Beat Saber may provide a playful, motivational, and immersive alternative. Yet, until now, it remains unclear which game mechanics, concepts, and designs work best for people with dementia, and how to implement exergames for and with this user group. In this paper, we adapted a human-centered design approach to address the specifics of developing VR exergames for people with dementia. This includes semi-structured interviews with stakeholders and contextual inquiries to better analyze the user requirements. Based on our analysis, we present Memory Journalist VR - a novel VR exergame specifically designed for people with dementia in a participatory design process. We report the qualitative evaluation based on the feedback gathered in five focus group sessions. Finally, we discuss the lessons learned, which provide important insights for the design of future VR exergames for people with dementia: (i) creating social gaming activities with a focus on shared aspects, (ii) support of an inverse game flow channel addressing decline and variance in cognitive-physical abilities, and (iii) ensuring a safe VR exergame experience.

Redirected walking allows users to explore immersive virtual environments by real walking even when the physical tracking space is limited. Redirected walking is usually implemented via translation gains, rotation gains, and curvature gains, while previous research was focused on identifying detection thresholds for such manipulations. To our knowledge, all previous experiments were conducted without a visual self-representation of the user in the virtual environment, in particular, without showing the user's feet. In this paper, we address the question if the virtual self-representation of the user's feet changes the detection thresholds for translation gains. Furthermore, we consider the influence of the holisticness of the visual stimulus, i. e., the type of virtual environment. Therefore, we conducted an experiment to identify detection thresholds for translation gains under three different conditions: (i) without visible virtual feet and (ii) with visible virtual feet both in a high fidelity visually rich virtual environment, and (iii) with visible virtual feet in a low cue virtual environment. The results revealed the range of detection thresholds for translations gains, which cannot be detected by the user when the feet are visible. Furthermore, the results show a significant difference between the two types of environment. Our findings suggest that the virtual environment is more important for manipulation detection than the visual self-representation of the user's feet.