Byung-Doh Oh (오병도)

I am a Faculty Fellow at the Center for Data Science at New York University, where I collaborate with Tal Linzen. I received my PhD in computational linguistics from The Ohio State University, where I worked with William Schuler.

My work aims to advance our understanding of language processing in machines and humans by drawing on techniques from machine learning and psycholinguistics. I am particularly interested in developing computational models that capture the real-time processing behavior of human language users, and interpretability techniques for studying the predictions and representations of neural networks.

selected publications

EMNLP

Leading whitespaces of language models’ subword vocabulary pose a confound for calculating word probabilities

Byung-Doh Oh, and William Schuler

In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (forthcoming), 2024

Bib PDF

@inproceedings{ohschuler24emnlp,
  author = {Oh, Byung-Doh and Schuler, William},
  title = {Leading whitespaces of language models' subword vocabulary pose a confound for calculating word probabilities},
  year = {2024},
  booktitle = {Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (forthcoming)},
  section = {papers},
}

EACL

Frequency explains the inverse correlation of large language models’ size, training data amount, and surprisal’s fit to reading times

Byung-Doh Oh, Shisen Yue, and William Schuler

In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics, 2024

Bib PDF Slides

@inproceedings{ohetal24eacl,
  author = {Oh, Byung-Doh and Yue, Shisen and Schuler, William},
  title = {Frequency explains the inverse correlation of large language models' size, training data amount, and surprisal's fit to reading times},
  year = {2024},
  booktitle = {Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics},
  pages = {2644--2663},
  section = {papers},
}

EMNLP Findings

Transformer-based language model surprisal predicts human reading times best with about two billion training tokens

Byung-Doh Oh, and William Schuler

In Findings of the Association for Computational Linguistics: EMNLP 2023, 2023

Bib PDF Slides

@inproceedings{ohschuler23emnlp,
  author = {Oh, Byung-Doh and Schuler, William},
  title = {Transformer-based language model surprisal predicts human reading times best with about two billion training tokens},
  year = {2023},
  booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2023},
  pages = {1915--1921},
  section = {papers},
}

ACL

Token-wise decomposition of autoregressive language model hidden states for analyzing model predictions

Byung-Doh Oh, and William Schuler

In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics, 2023

Bib PDF Poster

@inproceedings{ohschuler23acl,
  author = {Oh, Byung-Doh and Schuler, William},
  title = {Token-wise decomposition of autoregressive language model hidden states for analyzing model predictions},
  year = {2023},
  booktitle = {Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics},
  pages = {10105--10117},
  section = {papers},
}

TACL

Why does surprisal from larger Transformer-based language models provide a poorer fit to human reading times?

Byung-Doh Oh, and William Schuler

Transactions of the Association for Computational Linguistics, 2023

Bib PDF Slides

@article{ohschuler23tacl,
  author = {Oh, Byung-Doh and Schuler, William},
  title = {Why does surprisal from larger Transformer-based language models provide a poorer fit to human reading times?},
  year = {2023},
  journal = {Transactions of the Association for Computational Linguistics},
  volume = {11},
  pages = {336--350},
  section = {articles},
}

EMNLP

Entropy- and distance-based predictors from GPT-2 attention patterns predict reading times over and above GPT-2 surprisal

Byung-Doh Oh, and William Schuler

In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, 2022

Bib PDF Poster

@inproceedings{ohschuler22emnlp,
  author = {Oh, Byung-Doh and Schuler, William},
  title = {Entropy- and distance-based predictors from GPT-2 attention patterns predict reading times over and above GPT-2 surprisal},
  year = {2022},
  booktitle = {Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing},
  pages = {9324--9334},
  section = {papers},
}

FAI

Comparison of structural parsers and neural language models as surprisal estimators

Byung-Doh Oh, Christian Clark, and William Schuler

Frontiers in Artificial Intelligence, 2022

Bib PDF

@article{ohetal22fai,
  author = {Oh, Byung-Doh and Clark, Christian and Schuler, William},
  title = {Comparison of structural parsers and neural language models as surprisal estimators},
  year = {2022},
  journal = {Frontiers in Artificial Intelligence},
  volume = {5},
  pages = {777963},
  section = {articles},
}