/char-iit

A causal intervention framework to learn robust and interpretable character representations inside subword-based language models

Primary LanguageJupyter NotebookMIT LicenseMIT

Inducing Character-level Structure in Subword-based Language Models with Type-level Interchange Intervention Training

This repository contains dataset and models for ACL 2023 Findings paper: Inducing Character-level Structure in Subword-based Language Models with Type-level Interchange Intervention Training.

Code to reproduce results in the paper is also available in Colab.