Yuan Tian

I work across linguistics, HCI, and social computing. My recent projects study AI companionship, bilingual ambiguity, multilingual interaction, and the social lives of language in both contemporary and historical settings.

Researcher trained in linguistics with corpus, field, and behavioral research experience. Recent work focuses on AI-mediated interaction, multilingual communication, and human-AI differences in language use and interpretation.

Research themes

• Harm reduction in AI companionship: Designing and evaluating conversational interventions for risky dependency, support-seeking, and attachment patterns.
• Pragmatics under ambiguity: Studying how humans and language models navigate modifier scope, bilingual ambiguity, and non-canonical interpretation.
• Language in social context: Connecting corpus analysis, fieldwork, and social theory to understand identity, stigma, and linguistic change.

Current directions

• SafeCompanion: Evaluating Pushback/Friction as an Intervention for Harmful AI Interaction: Investigating mitigation strategies for high-risk AI companion interactions (e.g., emotional dependence) by evaluating user responses to varying levels of conversational friction.
• Bilingual Ambiguity Mental Models: Communication Strategies in Human vs. AI Interaction: Comparing communication strategies in human-human versus human-LLM interactions to understand how cognitive constraints shape strategy adoption in resolving bilingual ambiguity.
• AI Tutoring for Zhongkao Preparation: Audited limitations in existing Chinese exam-prep apps, especially shallow explanation flows and weak support for students without reliable offline tutoring resources.
• Pragmatics-Aware Evaluation of Degree Adverbs in Chinese LLM vs. Human Text: Comparing Chinese degree adverb usage in LLM-generated vs. human text, motivated by Chinese word-class heterogeneity and pragmatics-driven POS ambiguity.

Current questions

• Current question: How can language-sensitive design reduce harm in AI interaction without flattening the nuance people actually need?
• Through-line: From classical corpora to modern chat interfaces, the focus stays on interpretation, context, and how meaning changes in use.