Explicit negations are triggered by negation terms that may precede the finding being negated, as in "The chest x-ray revealed no abnormalities,'' or may follow the observation, as in "The patient is tumor free.'' Consistent with Zipf's law (Manning and Schutze, 1999), which states that there exist a few very common words, a middling number of medium-frequency words, and many low-frequency words, very few negation phrases account for the majority of negation in patient reports. Two studies on automated negation (Mutalik et al., 2001, Chapman et al., 2001a) found that a few negation phrases accounted for approximately 90% of negation in different report types: "no,'' "denies/denied,'' "without,'' and "not.''
The other 10% of negated observations are triggered by a potentially huge number of low-frequency negation phrases.
Once a human identifies a negation term, he must decide whether a relevant finding in the sentence is being negated by that term, that is, whether the finding is within the scope of the negation term. For example, in sentences (1) and (2) the words "source'' and "change'' are being negated by "not'' instead of the findings "infection'' and "pain.''
(1) This is not the source of the infection.
(2) There has not been much change in her pain.
Was this article helpful?