Xuhui Zhou

I am an incoming PhD student at the Language Technologies Institute at CMU interested in socially intelligent human language technology.

Specifically, I focus on how to enable NLP to better navigate social contexts. Nowadays, NLP models are strong in terms of benchmark performance. However, they produce hatespeech, contain social biases, behave undesiredly, and etc., which hurt, divide, and disappoint people.

Previously, I interned at Apple Machine Intelligence . I was a CLMS student at the University of Washington (UW) advised by Noah Smith. I got my bachelor degree in Statistics at Nanjing University (NJU).

news

Apr 14, 2022	Annotators with Attitudes: How Annotator Beliefs And Identities Bias Toxic Language Detection will appear at NACCL! And Emergent Communication Fine-tuning (EC-FT) for Pretrained Language Models is selected as the runner-up best paper for ICLR Emecom!
Jan 11, 2021	Our paper: Challenges in Automated Debiasing for Toxic Language Detection will appear at EACL 2021!
Dec 14, 2020	Accept two intern offers from Apple! One happens at the Siri Info Intel team for the spring quarter, another happens at the Machine Translation team for the summer quarter
Oct 20, 2020	Our paper: Linguistically-Informed Transformations (LIT): A Method for Automatically Generating Contrast Sets will appear at BlackboxNLP 2020.
Oct 2, 2020	Our paper: Multilevel Text Alignment with Cross-Document Attention will appear at EMNLP 2020.

selected publications

EACL

Challenges in Automated Debiasing for Toxic Language Detection

Zhou, Xuhui, Sap, Maarten, Swayamdipta, Swabha, Choi, Yejin, and Smith, Noah A.

In EACL 2021

Bib HTML Code

@inproceedings{zhou-etal-2020-debiasing,
  title = {Challenges in Automated Debiasing for Toxic Language Detection},
  author = {Zhou, Xuhui and Sap, Maarten and Swayamdipta, Swabha and Choi, Yejin and Smith, Noah A.},
  booktitle = {EACL},
  abbr = {EACL},
  html = {https://www.aclweb.org/anthology/2021.eacl-main.274.pdf},
  code = {https://github.com/XuhuiZhou/Toxic_Debias},
  year = {2021},
  bibtex_show = {true},
  selected = {true}
}

EMNLP

Multilevel Text Alignment with Cross-Document Attention

Zhou, Xuhui, Pappas, Nikolaos, and Smith, Noah A.

In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) 2020

Abs HTML Website

Text alignment finds application in tasks such as citation recommendation and plagiarism detection. Existing alignment methods operate at a single, predefined level and cannot learn to align texts at, for example, sentence \textitand document levels. We propose a new learning approach that equips previously established hierarchical attention encoders for representing documents with a cross-document attention component, enabling structural comparisons across different levels (document-to-document and sentence-to-document). Our component is weakly supervised from document pairs and can align at multiple levels. Our evaluation on predicting document-to-document relationships and sentence-to-document relationships on the tasks of citation recommendation and plagiarism detection shows that our approach outperforms previously established hierarchical, attention encoders based on recurrent and transformer contextualization that are unaware of structural correspondence between documents.
AAAI

Evaluating Commonsense in Pre-trained Language Models

Zhou, Xuhui, Zhang, Y., Cui, Leyang, and Huang, Dandan

In Proceedings of the AAAI Conference on Artificial Intelligence, 34(05) 2020

HTML