Language inference in medicine

State-of-the-art models using deep neural networks have become very good at learning an accurate mapping from inputs to outputs. However, they still lack generalization capabilities in conditions that differ from the ones encountered during training. This is even more challenging in specialized, and knowledge-intensive domains, where training data is limited. To address this gap, MedNLI was released—a dataset annotated by doctors, performing a NLI task, grounded in the medical history of patients.

Chaitanya Shivade explores strategies to leverage transfer learning using datasets from the open domain (e.g., Stanford’s NLI dataset) and to incorporate domain knowledge from external data and lexical sources (e.g., medical terminologies).

The models and data for MedNLI were created using MIT’s Medical Information Mart for Intensive Care (MIMIC) database and are freely available through its derived data repository. The ability to perform NLI in the medical domain can have interesting applications, such as introducing automation in clinical trial screening, checking for compliance of clinical care guidelines, and more.