Building Croatian Medical Dictionary from Medical Corpus

Abstract

The overall objective of this project is to define linguistic models at the lexical and syntactic levels that appear in the health domain, depending on the type of corpus. In the first phase of the project, the texts forming the medical corpus A – MedCorA (2,232 pharmaceutical instructions for medicaments available in Croatia) were prepared. The terminology found in this corpus was analyzed and the semantic subdomains (anatomy, condition, microorganism, chemistry, etc.) within the medical domain were defined and added to the dictionary entries. These dictionary resources were used as the foundation for the second phase in which NooJ morphological grammars were built allowing annotation of medical terminology in the corpus. Said grammars were built to allow for recognizing Latinisms, as well as Latin expressions written with Croatian case endings, not only Croatian words. Prepared resources are made available to a broader scientific community via Sketch Engine for further research in the field of medicine enabling additional research and development of algorithms for, among others, medical documents classification, medical texts’ information retrieval or machine translation of medical documentation, taking into account quality and reliability as well as terminology variability.

Keywords

• language processing
• semantic annotations
• medical domain
• NooJ
• Croatian