CfP | RTCA @ NeurIPS 2026 [R]

The emergence of the Real-Time Conversational Agents (RTCA) workshop at NeurIPS 2026 signals a crucial shift in the trajectory of conversational AI. Moving beyond the familiar landscape of text-based chatbots, RTCA focuses on the far more complex challenge of creating agents capable of natural, multimodal interaction in real-time. This focus addresses a critical constraint in current conversational AI – the persistent reliance on offline generation models that simply aren’t suited for dynamic, responsive dialogues. As Intuit's experience shows [Intuit scrapped its own AI agent architecture twice in four months. At VB Transform 2026, its AI VP called that the fast path], building truly functional agents isn't always a smooth process, and RTCA directly tackles a core obstacle. The call for papers highlights the need for shared benchmarks and methodologies, acknowledging a gap in the field that is vital to overcome as we move towards more sophisticated and integrated AI experiences.

The workshop’s emphasis on real-time generation, naturalness, and evaluation is particularly compelling. The technical hurdles outlined – latency, turn-taking, multimodal alignment – are significant and demand innovative solutions. It’s fascinating to see this focus on even seemingly minor aspects of interaction, such as backchannels and interruptions, being elevated to first-class problems. The inclusion of topics like prosody, emotion, and paralinguistic generation underscores the ambition to create agents that feel genuinely human. The consideration of safety, identity, and trust, particularly in the context of deepfakes and persuasion, is also vital. Roblox’s move to incorporate AI-powered game creation [Roblox launches an AI-powered game-creation feature in its mobile app] demonstrates the expanding accessibility of AI tools, and ensuring these tools are ethically sound and trustworthy becomes increasingly important as they become more integrated into our lives. We've also seen the continued emphasis on short-paper submissions [short-paper at ACL/EMNLP/EACL [R]], showing a need for rapid iteration and sharing of research in this quickly evolving space.

The scope of the topics calls for a truly interdisciplinary approach, bringing together researchers from speech, vision, language, HCI, and ML systems. This cross-pollination of expertise is essential to address the multifaceted challenges of real-time multimodal conversation. The workshop’s commitment to being non-archival – allowing authors to publish elsewhere – is a pragmatic decision that encourages rapid dissemination of findings. The focus isn't just on building impressive demos, but on developing robust, evaluable systems, as evidenced by the inclusion of rigorous evaluation methods like perceptual studies and interactive Turing-style tests. The call for datasets and benchmarks specifically tailored for interactive evaluation is a critical step towards fostering progress and enabling meaningful comparisons between different approaches.

Ultimately, the RTCA workshop represents a significant investment in the future of conversational AI. The shift from offline to real-time processing is not merely a technical upgrade; it’s a fundamental reimagining of how we interact with AI. As these agents move beyond text and into the physical and digital worlds, their ability to engage in natural, fluid conversations will be paramount. The challenge now lies in translating the research presented at RTCA into practical, deployable systems that can truly empower users and enrich our interactions – what novel interaction paradigms will emerge, enabled by these increasingly responsive and multimodal agents, and how will we ensure these interactions are both beneficial and ethically sound?