Geneva, Switzerland – A methodology has been developed by researchers at the University Hospitals of Geneva and the University of Geneva, addressing a pressing need in the realm of Natural Language Processing (NLP). The lack of substantial, high-quality annotated datasets, particularly in specialized areas like medicine, has been a longstanding challenge. This recent innovation offers a viable solution.
The researchers’ method revolves around crosslingual annotation projection. By employing a sophisticated, language-agnostic approach built on BERT technology—a state-of-the-art tool in NLP—they were able to transfer annotations from one language to another with significant precision. This process, known as annotation projection, is pivotal for efficiently creating datasets in languages where such resources are scarce or non-existent.
A tangible outcome of this methodology is the creation of FRASIMED, a comprehensive French annotated resource. This corpus comprises over 2,000 synthetic clinical cases and stands as a landmark achievement, being the most extensive open annotated corpus with integrated medical concepts in the French language available to the research community.
The depth of FRASIMED is further exemplified by its integration of datasets such as CANTEMIST and DISTEMIST. While CANTEMIST is centered on tumor morphology terms, DISTEMIST provides a wealth of information related to various diseases. Both these datasets play a pivotal role in enriching the corpus, ensuring it is a robust resource for medical NLP tasks in French.
One of the standout features of this research is the versatility of the methodology. While its primary application was to convert Spanish medical datasets into French, its design allows it to be adaptable across any bilingual corpus. This flexibility could potentially be a game-changer for researchers working in myriad languages, especially those that have traditionally been under-resourced.
The increasing intersection of healthcare and NLP, where the latter is utilized for analyzing medical records, making treatment recommendations, and other critical tasks, amplifies the importance of resources like FRASIMED. With the public release of FRASIMED, the research team hopes to accelerate advancements in the medical domain of French NLP.
For those in the academic and research sectors keen on delving deeper into the intricacies of this methodology and its applications, comprehensive details of the study are available. This initiative by the Geneva team underscores the importance of linguistic adaptability in research, setting the stage for future developments in the NLP arena. Read paper.
