Computation and Language for Society

Research focuses on misinformation and fact-checking, AI integration in the public information sphere, and causal inference with noisy text-based variables.

Examines how nonverbal facial cues contribute to social bias and stereotypes (especially ageism) and how subtle linguistic biases shape cognitive processing.

Pavel's research combines humanistic inquiry and computational analysis to examine how large language models summarize popular media. He develops validation methods to identify LLM biases, assess their political stakes, and ways we can address them.

Yingjia's research interests include verifiable reasoning, multimodality, and responsible generation in foundational models, with a focus on evaluation, inference-time scaling, and post-training alignment.

Thomas works on (i) multimodal and cross latent reasoning and (ii) verifiable reward models for tabular data, and (iii) multi-agent RAG system for database, while focusing on improving table representation in language model agents through both inference-time scaling and post-training alignment methods.

Siyi's research focuses on the socio-cognitive foundations of cooperation, communication, and innovation. Her recent projects examine the communicative mechanisms behind pantomime and investigate the role of social learning in driving communicative innovation.

Joyce's research examines how the cognitive mechanisms of machines align with and diverge from human intelligence in multimodal contexts — such as image-to-text and text-to-image generation — and how these (mis)alignments shape human-AI interactions with generative content.