From Korean Major to Google Gemini: A Linguist's Path to Top AI Research

Background: From Korean Studies to Linguistics

Fangru Lin (林芳如) earned a B.A. in Korean Language and Literature from Shanghai International Studies University (SISU) between 2017 and 2021, achieving a 3.83/4.0 GPA and receiving the first-class undergraduate scholarship. She then pursued graduate studies at the University of Oxford, completing an M.Sc. in Linguistics, Philology, and Phonetics (2023) and continuing toward a Ph.D. in the same fields.

Research Focus: LLM Reasoning, Planning, and Fairness

During her time at Oxford, Lin shifted her research toward large language model (LLM) reasoning and planning capabilities — specifically how models understand logical relations, execute complex steps, and transfer reasoning skills across different problem formulations.

ICML 2024: Graph-Enhanced LLMs for Asynchronous Plan Reasoning

As first author, Lin published "Graph-enhanced Large Language Models in Asynchronous Plan Reasoning" at ICML 2024. The work investigates whether restructuring textual input into graph structures helps models better handle asynchronous planning tasks.

ACL 2025: Dialect Fairness and Robustness in Reasoning

Her second first-author paper, "One Language, Many Gaps: Evaluating Dialect Fairness and Robustness of Large Language Models in Reasoning Tasks" , was accepted at ACL 2025 main conference. This study examines fairness and robustness gaps when LLMs process Standard American English versus African American Vernacular English (AAVE) in reasoning tasks.

ICML 2026: Cross-Representation Generalization of Procedures

A third first-author paper, "Can Large Language Models Generalize Procedures Across Representations?" , was accepted at ICML 2026. It explores whether models can transfer skills learned on symbolic representations (code, graphs) to tasks described in natural language, and how training can improve such cross-representation generalization.

Industry Experience: Microsoft and Google DeepMind

2022: Software Engineering Intern at Microsoft Azure Cloud+AI.

2024: Research Intern at Microsoft Research, working on LLM-related projects.

2025: Research Intern at Google DeepMind's Gemini Deep Research team, gaining direct exposure to frontier large-model research.

On October 28, 2026, Lin announced she had joined Google DeepMind as a Research Scientist focusing on Gemini large-scale pre-training.

Broader Implications: Diverse Paths into AI Research

The article notes a similar trajectory for Lin Junyang (林俊旸), former head of Alibaba's Qwen model, who also holds a linguistics background. As LLMs evolve from language generation toward complex reasoning, planning, and autonomous execution, understanding of language, semantics, and cognitive mechanisms increasingly intersects with computational methods, model training, and engineering practice. Lin's career illustrates that AI talent competition does not follow a single standard route; for aspiring researchers, disciplinary background is a starting point, but continuous learning, interdisciplinary ability, and solid research practice are key to long-term impact.