William Merrill

Young Investigator (Postdoc) @ Ai2

About

I'm starting as an Assistant Professor at the Toyota Technological Institute at Chicago later this year, currently doing a postdoc at the Allen Institute for AI. Previously, I did my PhD at NYU and also worked at Google Research My research leverages linguistics and the theory of computation to better understand large language models. I've worked on characterizing the computational power of language models: what kinds of algorithms can they implement, and what kinds of computation are beyond them? Another mysterious thing about language models is they learn simply by "autocompleting" text. Theoretical research of mine has touched on the question of whether learning in this way can give you a complete picture of language and the world, or whether some information must fundamentally be missing.

Country

United States

City

Seattle

Industry

Research

Skill

Machine Learning, Research, Python, Artificial Intelligence (AI), Linguistics, Game Development, Java, Natural Language Processing (NLP), Computational Linguistics

Experience

Young Investigator (Postdoc)

Ai2

2025-8 - Present · 1 yr 2 mos

Seattle, Washington, United States

Research Intern

Ai2

2024-5 - 2025-8 · 1 yr 4 mos

Open language model pretraining

Doctoral Student

NYU Center for Data Science

2021-9 - 2025-8 · 4 yrs

New York, New York, United States

Research Intern

Ai2

2023-5 - 2023-8 · 4 mos

Seattle, Washington, United States

Student Researcher

Google

2022-6 - 2022-12 · 7 mos

New York, New York, United States

Pre Doctoral Young Investigator

Ai2

2019-9 - 2021-9 · 2 yrs 1 mo

Seattle, Washington, United States

Undergraduate Learning Assistant – Natural Language Processing (CPSC 477/577)

Yale University

2018-1 - 2019-5 · 1 yr 5 mos

I worked for Dragomir Radev's NLP course twice (spring 2018 and 2019). I hosted office hours for students and helped with the design and grading of homework assignments. In 2018, I created an assignment about CCG parsing from scratch. This involved writing skeleton code, instructions, a full solution, and an auto-grader.

Research Assistant

Yale University

2016-5 - 2018-2 · 1 yr 10 mos

I worked as a research assistant to two academic workshops on medieval Central Asian languages and history. The first (summer 2016) was about the Kitan language, and the second (spring 2018) was about Tangut. My responsibilities included filming lectures, editing and publishing the recorded footage, and technical support for typesetting text in esoteric scripts.

Undergraduate Learning Assistant – Artificial Intelligence (CPSC 470/570)

Yale University

2017-9 - 2017-12 · 4 mos

I designed a homework assignment on computational agents for the game Othello and oversaw its implementation and grading for 200+ students. Other responsibilities included 3 hours of office hours per week.

Software Engineer Intern

Google

2018-5 - 2018-8 · 4 mos

New York City

Designed and implemented a deep learning project related to DoubleClick ads analytics, as well as a full-stack pipeline to deploy it. Received a full-time return offer. Project involved research, experimentation and implementation of several different kinds of ML models, and final deployment of the best model. I also developed the front-end to interact with the model. The project involved writing production code in Python, Java, and JavaScript and utilized TensorFlow, numpy, sklearn, and in-house frameworks.

Research Assistant

Boston College

2017-6 - 2017-8 · 3 mos

Greater Boston Area

Working under the guidance of the lab supervisor and a senior researcher, I applied machine learning to two related NLP tasks: classifying the native language of an English learner from an essay they have written, and grading the fluency of essays written by students learning a foreign language. The models I created were able to achieve state-of-the-art accuracy, and they are being used currently by Boston College in two ways: 1) to automate the placement of students in foreign language classes and 2) to score the fluency of subjects in large crowdsourced experiments about language acquisition.