Negation Detection and Antonyms
Negation status identification for findings or diagnoses is an important medical/clinical data mining problem. Negative qualifiers assigned to a medical condition may indicate the absence of the condition, so the ability to reliably identify the negation status of medical concepts affects the quality of results produced by the indexing and search tools. Previous studies include:
- Negation detection includes [2017 Enger]:
- cue detection: detects words that signal negation, such as no, not, without, and unfortunate
- scope detection: identifies the span of the sentence that is affected by the negation
- Antonyms and negation detection:
- Query expansion can be utilized when antonyms with negation are detected in NLP applications to increase performance (precision and recall) by avoiding additional processes of word sense disambiguation (WSD) and scope detection.
- Example-1 (PMID: 11092366): [Pharmacological treatment was not successful, but surgical intervention provided an acceptable result.]
- not successful: not (negation cue word) is detected as a negation, successful has antonym of unsuccessful (bounded antonym pair)
- with query expansion on antonym with negation, the text is converted to [Pharmacological treatment was unsuccessful, ...
- [not successful] is converted to [unsuccessful] and then precisely mapped to a single concept C1272705 for better performance in concept mapping.
- "Unsuccessful" should be detected as negation.
- Without query expansion on antonyms with negation [not successful], the following UMLS concepts are mapped:
- not|C1417836|C1518422
- successful|C1272703|C0597535
- It results in 4 combinations of concept mapping and requires further NLP processes for WSD, while with substituting negated antonyms, results in only 1 concept mapping.
- Exampl-2 (PMID: 17414516): [Even though hormone-refractory prostate cancer is still incurable, it is not untreatable.]
- It is not untreatable = It is treatable
- This is not a negation. It is a false positive negation if only ‘not’ is used for negation detection.
- Example-3 (PMID: 17432562): [this hormonal therapy is not without side effects.]
- not|C1417836|C1518422 (negation)
- without|C0332288|
- side effects|C0879626|C0877248|C0001688
- Two negations, the conclusion is with some side-effect
- not without = with (some)
- Not a negation (false positive)
- Example-4 (PMID: 15326468): [Self-inflicted spinal cord injury with a small needle is difficult, but not impossible.]
- not impossible = possible
- Not a negation (false positive)
- not|C1417836|C1518422 (negation)
- impossible|C4743675
- possible|C0332149
- Example-5 (PMID: 16461473): [renal failure may ensue if medical intervention is not successful] (= unsuccessful)
- Example-6 (PMID: 28770948): [A great number of patients receiving medical intervention is not successfully (= unsuccessfully) treated, ...]
- Example-7 (not from PubMed): [In acute suppuration of the knee, excision is never successful.] (= unsuccessful)