We present ACLBot, an interactive chatbot designed to support literature exploration in the ACL Anthology by combining structured knowledge graph querying with large language model (LLM) generative AI. ACLBot integrates a Neo4j-based knowledge graph constructed by extracting data on publications, authors, topics, and research trends from the ACL Anthology, and automatically generates knowledge graph queries to retrieve relevant information in response to user questions. Retrieved results are re-injected into the LLM to produce concise, contextually grounded summaries. We describe the system’s architecture, including its query generation pipeline, knowledge graph integration, and visualization components for highlighting temporal trends in research. To assess usability and effectiveness, we conducted a user evaluation with researchers, collecting qualitative and quantitative feedback on response accuracy, informativeness, and utility for literature discovery. Results indicate that ACLBot effectively supports exploratory search, helps identify relevant works and trends, and offers a promising framework for integrating structured information with generative AI for scientific information retrieval.

Robust understanding of user input is a core requirement for voice assistants deployed in real-world environments. In practice, these systems encounter heterogeneous fallback situations caused by noisy audio input, transcription errors, ambiguous requests, incomplete utterances, or unintended activations. Existing systems typically respond with generic fallback messages, which do not resolve the underlying interaction failure and can degrade user experience. We study fallback handling in a deployed smartwatch-based voice assistant for general health support in everyday environments. Our analysis is based on six months of real-world usage data from more than 500 users, yielding a dataset of 3,030 anonymized, naturally occurring fallback-triggering utterances. We contribute (1) an operational taxonomy and the annotated VoxFallbacks dataset of these interactions, (2) a comparative evaluation of different models within a classification pipeline under practical deployment constraints, and (3) practical lessons for designing robust and cost-efficient fallback mechanisms. Results show that lightweight embedding-based classifiers outperform larger generative models on most classification tasks while requiring substantially fewer computational resources.