Yuan Yu

Kent K. Chang, Mackenzie Hanh Cramer, Anna Ho et al.

Conversations are usually structured by roles -- who is speaking, who's being addressed, and who's listening -- and unfold in threads that break with changes in speaker floor or topical focus. While large language models (LLMs) have shown incredible capabilities in dialogue and reasoning, their ability to understand fine-grained conversational structure, especially in multi-modal, multi-party settings, remains underexplored. To address this gap, we introduce a suite of tasks focused on conversational role attribution (speaker, addressees, side-participants) and conversation threading (utterance linking and clustering), drawing on conversation analysis and sociolinguistics. To support those tasks, we present a human annotated dataset of 4,398 annotations for speakers and reply-to relationship, 5,755 addressees, and 3,142 side-participants. We evaluate popular audio-visual LLMs and vision-language models on our dataset, and our experimental results suggest that multimodal conversational structure understanding remains challenging. The most performant audio-visual LLM outperforms all vision-language models across all metrics, especially in speaker and addressee recognition. However, its performance drops significantly when conversation participants are anonymized. The number of conversation participants in a clip is the strongest negative predictor of role-attribution performance, while acoustic clarity (measured by pitch and spectral centroid) and detected face coverage yield positive associations. We hope this work lays the groundwork for future evaluation and development of multimodal LLMs that can reason more effectively about conversation structure.

Steve Mann, Tom Furness, Yu Yuan et al.

The contributions of this paper are: (1) a taxonomy of the "Realities" (Virtual, Augmented, Mixed, Mediated, etc.), and (2) some new kinds of "reality" that come from nature itself, i.e. that expand our notion beyond synthetic realities to include also phenomenological realities. VR (Virtual Reality) replaces the real world with a simulated experience (virtual world). AR (Augmented Reality) allows a virtual world to be experienced while also experiencing the real world at the same time. Mixed Reality provides blends that interpolate between real and virtual worlds in various proportions, along a "Virtuality" axis, and extrapolate to an "X-axis". Mediated Reality goes a step further by mixing/blending and also modifying reality. This modifying of reality introduces a second axis. Mediated Reality is useful as a seeing aid (e.g. modifying reality to make it easier to understand), and for psychology experiments like Stratton's 1896 upside-down eyeglasses experiment. We propose Multimediated Reality as a multidimensional multisensory mediated reality that includes not just interactive multimedia-based reality for our five senses, but also includes additional senses (like sensory sonar, sensory radar, etc.), as well as our human actions/actuators. These extra senses are mapped to our human senses using synthetic synesthesia. This allows us to directly experience real (but otherwise invisible) phenomena, such as wave propagation and wave interference patterns, so that we can see radio waves and sound waves and how they interact with objects and each other. Multimediated reality is multidimensional, multimodal, multisensory, and multiscale. It is also multidisciplinary, in that we must consider not just the user, but also how the technology affects others, e.g. how its physical appearance affects social situations.