arXiv cs.CL
·
18 hours ago
Researchers introduced Metaphor-Fluid Design, an approach that dynamically adjusts metaphorical representations in voice user interfaces based on conversational context rather than using static metaphors. Two studies with 130 and 91 participants found that adapting metaphors to specific contexts—commands, information seeking, sociality, and error recovery—improved perceived adoption intention, enjoyment, and likability compared to standard assistant-based interfaces. This challenges uniform VUI design and suggests that context-adaptive metaphor selection creates more engaging interactions aligned with user expectations.
arXiv cs.CL
·
18 hours ago
Researchers developed a multi-agent LLM framework to detect OCEAN personality traits from long text narratives by using separate sub-agents conditioned to represent high, low, and neutral perspectives for each trait, combined with a judge model that aggregates their outputs. The framework uses masked language modeling and psychometric supervision to train sub-agents and was evaluated on life narrative datasets with ablation studies and qualitative analyses. This approach reduces individual model biases and provides more consistent personality trait predictions from extended written content.
arXiv cs.AI
·
18 hours ago
Researchers propose a human-centered artificial intelligence framework for AI-assisted lexicography that emphasizes augmenting rather than replacing lexicographers. The framework identifies four dimensions for examining AI integration: the augmented lexicographer, sociotechnical context, bias, and tool design. The approach aims to preserve lexicographer agency and linguistic diversity while enabling beneficial AI integration into lexicographic workflows.
