A Latent-Variable Model for Intrinsic Probing


  • Karolina Stańczak University of Copenhagen
  • Lucas Torroba Hennigen Massachusetts Institute of Technology
  • Adina Williams Meta AI Research
  • Ryan Cotterell ETH Zürich
  • Isabelle Augenstein University of Copenhagen




SNLP: Interpretability & Analysis of NLP Models, SNLP: Machine Translation & Multilinguality


The success of pre-trained contextualized representations has prompted researchers to analyze them for the presence of linguistic information. Indeed, it is natural to assume that these pre-trained representations do encode some level of linguistic knowledge as they have brought about large empirical improvements on a wide variety of NLP tasks, which suggests they are learning true linguistic generalization. In this work, we focus on intrinsic probing, an analysis technique where the goal is not only to identify whether a representation encodes a linguistic attribute but also to pinpoint where this attribute is encoded. We propose a novel latent-variable formulation for constructing intrinsic probes and derive a tractable variational approximation to the log-likelihood. Our results show that our model is versatile and yields tighter mutual information estimates than two intrinsic probes previously proposed in the literature. Finally, we find empirical evidence that pre-trained representations develop a cross-lingually entangled notion of morphosyntax.




How to Cite

Stańczak, K., Torroba Hennigen, L., Williams, A., Cotterell, R., & Augenstein, I. (2023). A Latent-Variable Model for Intrinsic Probing. Proceedings of the AAAI Conference on Artificial Intelligence, 37(11), 13591-13599. https://doi.org/10.1609/aaai.v37i11.26593



AAAI Technical Track on Speech & Natural Language Processing