Medical Korean: applying medical term and abbreviation dictionaries

Medical text (medical records, nursing records, consultation notes) is packed with medical terms and abbreviations. A general morphological analyzer treats such terms as out-of-vocabulary words (OOV) or splits them wrongly, which throws off symptom and drug statistics and searches. Bareun recognizes medical terms accurately when you register them in a custom dictionary.

The problem

Medical terms such as 고혈압성심장질환 (hypertensive heart disease) and 급성심근경색 (acute myocardial infarction) get chopped up.
English abbreviations such as EKG and CT are caught only as generic foreign-language tokens, so their meaning is lost.
When terms are unstable, patient record searches and cohort extraction become inaccurate.

How to apply Bareun

Register medical terms in the compound noun dictionary (cp_set) or the proper noun dictionary (np_set) to keep them as single words. If you also register English abbreviations in the dictionary, you can treat them as units of meaning. For the procedure of collecting and registering domain terms, see Domain dictionary workflow, and for why out-of-vocabulary words occur, see Handling out-of-vocabulary words (OOV).

graph TD
  REC[Medical and nursing records] --> DICT[Custom dictionary of medical terms and abbreviations];
  DICT --> TAG[Bareun morphological analysis];
  TAG --> TERM[Accurate recognition of medical terms];
  TERM --> STAT[Symptom and drug statistics, search];

from bareunpy import Tagger

tagger = Tagger("koba-XXXX-...", "localhost")

# Register medical terms and abbreviations
cust = tagger.custom_dict("medical_terms")
cust.copy_cp_set({"고혈압성심장질환", "급성심근경색"})
cust.copy_np_set({"EKG", "CT"})
cust.update()
tagger.set_domain("medical_terms")

res = tagger.tags(["급성심근경색 의심되어 EKG 시행함."])
print(res.nouns())

['급성심근경색', 'EKG']

Both the long medical term and the English abbreviation are each recognized as a single word.

Results and benefits

Item	Without a dictionary	With a medical term dictionary
Long medical terms	Split into several nouns	Kept as one word
English abbreviations	Meaningless foreign-language tokens	Recognized as registered terms
Statistics and search	Inaccurate	Consistent

Privacy caution

Medical text contains sensitive information. If you need an air-gapped network, process it with on-premises (self-hosted) Bareun rather than the cloud. Morphological analysis can be run as a self-hosted installation.

Frequently asked questions

Q. What if an abbreviation overlaps with an ordinary word?

For abbreviations that are interpreted differently depending on context, run separate domain dictionaries, or check for conflicts with CheckConflict and specify the higher-priority dictionary first. For running multiple dictionaries, see Using multiple dictionaries together.

Q. I have thousands of terms. Can I register them all at once?

You can register custom dictionary entries in bulk, by set. Build a set of terms, pass it to copy_cp_set and copy_np_set, and apply it with update().

Q. Sensitive information means I cannot use the cloud.

Morphological analysis can run in an air-gapped network as a self-hosted installation (Linux .deb, Windows .exe, macOS .pkg, or Docker). Process patient data internally without sending it outside.

Q. Can I register medical terms in bulk?

Yes. You can register custom dictionary entries in bulk, by set. Build a set of terms, with long medical terms in the compound noun dictionary (cp_set) and English abbreviations in the proper noun dictionary (np_set), pass them to copy_cp_set and copy_np_set, and apply with update(). Even thousands of terms are registered at once and recognized as single words.

Domain dictionary workflow: registering medical terms in a dictionary
Using multiple dictionaries together: running separate dictionaries per department
Handling out-of-vocabulary words (OOV): why abbreviations and technical terms get missed