THE MLP COOCCURRENCE SELECTION


 
Selection is the second component of the MLP. The MLP parser reads in the selection "grammar" (procedures) and medical co-occurrence patterns as filters on the parse tree generated by the parsing component.

Selection has one main function and two support operations:

  1. It passes or rejects a parse based on subtrees, relying on confirmed cooccurrence patterns (in form of cooccurrence selection patterns).
      For example, selection will not reject the sentence Pt had swelling in the arms and fever:
    • the noun phrase swelling in the arms is accepted by List HOST-P-N with the host noun swelling (H-INDIC) modified by the prepositional phrase in the arms (H-PTPART).
    • the conjunction of fever (H-INDIC) with the noun arms (H-PTPART) in the conjoined noun phrases, arms and fever is rejected by List CONJ-EQUIV-CLASSES.
    • the conjoining of swelling (H-INDIC) with fever (H-INDIC) in the conjoined noun phrases, swelling in the arms and fever is confirmed by List CONJ-EQUIV-CLASSES.
    • the pattern for subject Patient (H-PT), verb had (VHAVE) and two objects swelling in the arms and fever (H-INDIC) is confirmed by List SUBJ-VERB-OBJ.
    Thus the sentence is accepted by a series of lists "validated" from the actual medical documents.
  2. It reshapes the sentence "vertically" to maintain the word order of the original input text. In the above example, the noun phrase fever is raised to the level of the noun phrase swelling after the conjoining of arms and fever is rejected, i.e. pattern swelling in fever does not exist in this sublanguage.
  3. See List of Cooccurrence Patterns validated from medical texts.

 
Copyright © 2005 by Medical Language Processing, LLC.