John Hewitt
johnhew [at] stanford.edu
Hi! I’m a fourth-year PhD student in computer science at Stanford University. I conduct research in natural language processing and machine learning. I am grateful to be co-advised by Chris Manning and Percy Liang, and to be supported by an NSF Graduate Research Fellowship.
I aim to design systems that robustly and efficiently learn to understand human languages to the end of advancing human communication and education, and to teach others.
Feel free to look me up on Google Scholar or Twitter, or take my CV.
New! Blog post on initializing new word embeddings for pretrained LMs. New! Blog post on characterizing knowledge in representations.
Current Research Interests
- Memory: what NLP systems store and how they access it.
- Understanding and improving unsupervised learning of language
- Theoretical insights in NLP
- Out-of-domain extrapolation in NLP
- Multilinguality and low-resource language processing
-
-
Publications
Publications
2021
Conditional probing: measuring usable information beyond a baseline.
John Hewitt, Kawin Ethayarajh, Percy Liang, Christopher D. Manning.
EMNLP 2021 (short papers).
(pdf) (blog) (code) (codalab)
On the Opportunities and Risks of Foundation Models.
Bommasani et al (+100 authors). John Hewitt, Co-lead; Interpretability section.
whitepaper.
(pdf)
Probing artificial neural networks: Insights from neuroscience.
Anna Ivanova, John Hewitt, Noga Zaslavsky.
Brain2AI 2021.
(pdf)
Refining Targeted Syntactic Evaluation of Language Models.
Benjamin Newman, Kai-Siang Ang, Julia Gong, John Hewitt.
NAACL 2021 (short papers).
(pdf) (code)
2020
RNNs can generate bounded hierarchical languages with optimal memory.
John Hewitt, Michael Hahn, Surya Ganguli, Percy Liang, Christopher D. Manning.
EMNLP 2020 (long papers).
(pdf) (blog) (code:analytic) (code:learning) (codalab)
The EOS Decision and Length Extrapolation.
Benjamin Newman, John Hewitt, Percy Liang, Christopher D. Manning
BlackBoxNLP 2020. (Outstanding Paper Award).
(pdf) (code)
Emergent Linguistic Structure in Artificial Neural Networks Trained by Self-Supervision.
Christopher D. Manning, Kevin Clark, John Hewitt, Urvashi Khandelwal, Omer Levy
Proceedings of the National Academy of Sciences. 2020.
(pdf)
Finding Universal Grammatical Relations in Multilingual BERT.
Ethan A. Chi, John Hewitt and Christopher D. Manning.
ACL 2020 (long papers).
(pdf) (bib) (code) (viz)
2019
Designing and Interpreting Probes with Control Tasks.
John Hewitt and Percy Liang.
EMNLP 2019 (long papers). (Runner Up Best Paper Award).
(pdf) (bib) (blog) (code) (codalab) (slides) (talk).
A Structual Probe for Finding Syntax in Word Representations.
John Hewitt and Christopher D. Manning.
NAACL 2019 (short papers).
(pdf) (bib) (blog) (code) (nlp highlights podcast) (slides) (talk).
Simple, Fast, Accurate Intent Classification and Slot Labeling for Goal-Oriented Dialogue Systems.
Arshit Gupta*, John Hewitt* and Katrin Kirchhoff.
SIGDIAL 2019.
(pdf)
*: Equal contribution; authors listed alphabetically2018
A Distributional and Orthographic Aggregation Model for English Derivational Morphology.
Daniel Deutsch*, John Hewitt* and Dan Roth.
ACL 2018 (long papers).
(pdf)
*: Equal contribution; authors listed alphabeticallyLearning Translations via Images with a Massively Multilingual Image Dataset.
John Hewitt*, Daphne Ippolito*, Brendan Callahan, Reno Kriz, Derry Tanti Wijaya and Chris Callison-Burch.
ACL 2018 (long papers).
(pdf)
*: Equal contribution; authors listed alphabeticallyXNMT: The eXtensible Neural Machine Translation Toolkit.
Graham Neubig, Matthias Sperber, Xinyi Wang, Matthieu Felix, Austin Matthews, Sarguna Padmanabhan, Ye Qi, Devendra Singh Sachan, Philip Arthur, Pierre Godard, John Hewitt, Rachid Riad, and Liming Wang.
AMTA 2018.
(pdf)2017
Learning Translations via Matrix Completion.
Derry Tanti Wijaya, Brendan Callahan, John Hewitt , Xiao Ling, Marianna Apidianaki, and Chris Callison-Burch.
EMNLP 2017 (long papers).
(pdf)
2016
Automatic Construction of Morphologically-Motivated Translation Models for Highly Inflected Low-Resource Languages.
John Hewitt, Matt Post, David Yarowsky.
AMTA 2016.
(pdf)Invited Talks
A Natural Language Processing perspective on supervised analysis of neural representations.
EvLab, MIT. December 2, 2020.The Unreasonable Syntactic Expressivity of RNNs.
USC ISI NLP Seminar. (video) November 5, 2020.Language Probes as V-information Estimators.
NLP with Friends. September 9, 2020.Probing Neural NLP: Ideas and Problems.
Berkeley NLP Seminar. November 18, 2019.Emergent Linguistic Structure in Neural NLP.
Amazon AI. July 25, 2019.A Structural Probe for Finding Syntax in Word Representations.
NLP Highlights Podcast. May, 2019.Abstracts
RNNs can generate bounded hierarchical languages with optimal memory.
John Hewitt, Michael Hahn, Surya Ganguli, Percy Liang, Christopher D. Manning
2020 Conference on the Mathematical Theory of Deep Learning (abstracts).
Semantic Bootstrapping in Frames: A Computational Model of Syntactic Category Acquisition.
John Hewitt, Jordan Kodner, Mitch Marcus, and Charles Yang.
Conference of the Cognitive Science Society (CogSci), (member posters) 2017. (pdf) (abstract)Patents
Capturing Rich Response Relationships with Small-Data Neural Networks.
John Hewitt.
US Patent App 15/841,963. December 2017. (granted). (application) - Blog
-
Projects
-
Transformers lecture
- I wrote a lecture on Transformers in my role as Head TA for Stanford’s CS 224N: Natural Language Processing with Deep Learning.
-
Pretraining lecture
- I wrote a lecture on Pretraining for the same course!
-
Model analysis and explanation lecture
- I wrote a lecture on analysis and explanation of NLP models for the same course!
-
-
About
Tidbits
This talk by Rajiv Gandhi, to whom I am grateful. For you if you think, like I used to, that research–or any success in STEM–is out of your reach.
Scott Aaronson’s old note on frameworks for reasoning about large numbers, for enjoyment
Kevin Knight’s note on unix commands, to help you with your bash skills
The Fundamental Whiteboard Difficulty (Scott Aaronson):
I figured that chalk has its problems—it breaks, the dust gets all over—but I could live with them, much more than I could live with the Fundamental Whiteboard Difficulty, of all the available markers always being dry whenever you want to explain anything.
I highly suggest Arch Linux for its configurability and the educational experience it provides…
Contact
Take my school email johnhew@stanford, and predict the TLD using your internal knowledge base.