Special issue of the TAL journal on 'NLP and health'

The medical application domain and the discipline of Natural Language
Processing have been interacting for a good half century now, to mutual
benefit. NLP techniques have helped in medical knowledge discovery and
clinical practice improvements.  At the same time, the medical field has
contributed meaningful tasks, sizeable document collections (e.g., MIMIC
for patient records, MEDLINE for scientific abstracts), and detailed
lexical resources (UMLS), and these have helped advancements in the
discipline of NLP in general.

Language is prevalent throughout the care process. It is used to encode
institutional knowledge and health information, patient language
productions can be analyzed for diagnosis (speech and/or voice
disorders, language disorders, mental illnesses), it conveys patient and
health professional interactions (consultations, consensus meetings),
and also encodes clinical descriptions of pathologies and their
management from the perspective of patients and health professionals
(social networks, patient records). These language productions all
address the specialiazed theme of health while exhibiting great
diversity in terms of medium (written or spoken language), register
(written material published in journals, professional note-taking in
clinical documents, spontaneous production on social networks), language
(English for literature, any language for other types of documents),
etc.

As computer science goes through rapid changes (deep learning, big data,
internet of things), and the medical field is seeing its own
opportunities (precision medicine, drug discovery) and pressures
(chronic diseases, an aging population), interactions between these
fields are more relevant than ever.

This special issue of TAL aims to provide an overview of current NLP
research on all aspects of health, including fundamental work and
applications.

Authors are invited to submit papers on all aspects of NLP for health,
in particular regarding, but not limited to, the following issues and
tasks:

- general vs. in-domain pre-trained language models
- general pre-trained vs. in-domain embeddings