Natural Language Processing · Large Language Models · Interpretability

Linfeng Liu

I am a PhD student in the Department of Computer Science at the University of Cincinnati, advised by Prof. Tianyu Jiang in the cincyNLP Lab. My research explores how language and multimodal AI systems represent meaning beyond surface form.

I am especially interested in evaluation settings where literal pattern matching is not enough: non-compositional language, indirect visual reference, multilingual figurative generation, and geometric views of model-question interactions.

Linfeng Liu

Research Interests

  • Natural language processing
  • Large language models
  • Machine translation
  • Multimodality
  • Interpretability

Experience

  • University of Cincinnati, PhD student, 2025-present
  • University of Cincinnati, MS student, 2024-2025
  • University of Cincinnati, undergraduate student, 2019-2024
  • Chongqing University, undergraduate student, 2019-2024

Selected Publications

2026

Evaluating the Impact of Verbal Multiword Expressions on Machine Translation

Linfeng Liu, Saptarshi Ghosh, Tianyu Jiang

ACL 2026 Main Conference

Evaluates how verbal multiword expressions affect machine translation quality across seven language pairs and eight machine translation systems.

evaluation · multiword expressions · machine translation

2026

A Computational Approach to Visual Metonymy

Saptarshi Ghosh, Linfeng Liu, Tianyu Jiang

EACL 2026 Main Conference Oral

Introduces ViMET, a visual metonymy benchmark for testing whether vision-language models can interpret indirect associative cues rather than only literal depiction.

metonymy · semiotic theory · dataset · cognitive reasoning

2026

A Geometric Lens on LLM Abilities through Joint Embedding Item Response Theory

Louie Hong Yao, Nicholas Jarvis, Tiffany Zhan, Saptarshi Ghosh, Linfeng Liu, Tianyu Jiang

Transactions on Machine Learning Research (TMLR), accepted, to appear

Introduces JE-IRT, a framework for analyzing LLM evaluation through shared geometric representations of abilities, item semantics, and difficulty.

large language models · interpretability · item response theory · joint embeddings

Preprints

2026

Cross-Lingual Steering for Figurative Language Generation

Linfeng Liu, Tiffany Zhan, Louie Hong Yao, Saptarshi Ghosh, Tianyu Jiang

Uses activation steering to test whether figurative-language generation signals in multilingual LLMs are language-specific or reusable across languages.

figurative language · multilingual large language models · activation steering