Our group develops and applies methods in computational linguistics and natural language processing (NLP) to address a variety of questions about linguistic patterns in human communities. We are interested in creating robust and beneficial language technology, and also in formalizing and testing theories in the language sciences. These goals go hand-in-hand. We use classical algorithms, deep learning models, and experimental data to address them.

We are particularly interested in approaches that:

• Are anchored in what is known about linguistic systems, and about how humans acquire them and deploy them in language processing.
• Succeed in handling language variation across different times, topics, and social contexts.
• Make forecasts or predictions that are consequential and interpretable.

• 11/2021: paper at EMNLP 2021 (Findings)

  • Temporal adaptation of BERT and performance on downstream document classification: Insights from social media

• 08/2021: three papers at ACL 2021

  • Dynamic contextualized word embeddings
  • HateCheck: Functional tests for hate speech detection models
  • Superbizarre is not superb: Derivational morphology improves BERT’s interpretation of complex words

• 06/2021: paper at SwissText 2021

  • Predicting COVID-19 cases using Reddit posts and other online resources

• 11/2020: paper at EMNLP 2020

  • DagoBERT: Generating derivational morphology with a pretrained language model

• 07/2020: two papers at ACL 2020

  • A graph auto-encoder model of derivational morphology
  • Predicting the growth of morphological families from social and linguistic factors