The Transformation component decomposes sentences into their
basic canonical sentences. The MLP parser
reads in the grammar procedures and uses them to
transform complex sentence
parse tree filtered by the selection component into basic
canonical assertions or fragments.
Transformation component has three main functions:
- It resolves in-sentence anaphora.
For example, it refers to aspirin, and she refers to Pt,
in the sentence Pt was given aspirin qhs but she did not take it:
- The algorithm searches backward in the parse tree for an antecedent of she—an NHUMAN pronoun (one of H-PT, H-INST or H-FAMILY). It rejects aspirin (NONHUMAN and H-TTMED), but finds a match: Pt (NHUMAN, H-PT, SINGULAR).
- The algorithm searches backward in the parse tree for an antecedent of it—a NONHUMAN pronoun. It rejects Pt (NHUMAN and H-PT), but finds a match: aspirin (NONHUMAN, H-TTMED, SINGULAR).
- It expands conjunctions in the sentence into basic assertions.
the sentence Today, she had no cough, chest pain or shortness of breath is expanded into three basic assertions:
- Today, she had no cough
- Today, she had no chest pain
- Today, she had no shortness of breath
- The algorithm creates three assertions, and makes sure that
each has a subject she, a verb had, a time sentential adjuncts
today and a noun phrase adjunct no.
- The algorithm converts an or conjunction with negation into
- It turns wh-phrases into separate assertions.
- It records indices of noun phrases, tense of sentence, etc.