Fangru Lin

Fangru Lin works as a research scientist at Google DeepMind, studying how language models reason and plan across different representations. Their research includes methods for improving procedural generalization and evaluations of how language-model reasoning performs across dialects; they completed a DPhil in Linguistics, Philology, and Phonetics at the University of Oxford.

Career

Cosigned byLoading...

CosignedSign in to view

Elsewhere

Google DeepMind

Oct 2026 — Present

Research Scientist

University of Oxford

DPhil in Linguistics, Philology, and Phonetics

University of Oxford

MPhil in Linguistics, Philology, and Phonetics

Shanghai International Studies University

BA Korean

Products & media

icml.cc

Can Large Language Models Generalize Procedures Across Representations?

An ICML 2026 poster on whether language models can transfer procedural reasoning between code, graphs, and natural language, and a curriculum designed to improve that ability.

Media appearance

proceedings.mlr.press

Graph-enhanced Large Language Models in Asynchronous Plan Reasoning

A study of how language models handle asynchronous plans and a graph-based approach that combines graph structure with natural-language prompts.

Writing

arxiv.org

One Language, Many Gaps: Evaluating Dialect Fairness and Robustness of Large Language Models in Reasoning Tasks

Research introducing the ReDial benchmark and examining how language models perform on reasoning prompts in African American Vernacular English.

Writing

github.com

ReDial dialect robustness and fairness dataset

The project repository describes a dataset of paired Standardized English and AAVE reasoning prompts for studying language-model fairness and robustness.

Profile name

Company name

Company name

Profile name

Company name

Company name

Profile name

Company name

Company name

Profile name

Company name

Company name

Profile name

Company name

Company name

Fangru Lin

Google DeepMind

University of Oxford

University of Oxford

Shanghai International Studies University

Products & media

Can Large Language Models Generalize Procedures Across Representations?

Graph-enhanced Large Language Models in Asynchronous Plan Reasoning

One Language, Many Gaps: Evaluating Dialect Fairness and Robustness of Large Language Models in Reasoning Tasks

ReDial dialect robustness and fairness dataset

Curated Lists

List Inclusions