<?xml version="1.0" standalone="yes"?>
<Paper uid="C82-1032">
  <Title>ANALYSIS AND PROCESSING OF COMPACT TEXT</Title>
  <Section position="5" start_page="203" end_page="203" type="evalu">
    <SectionTitle>
RESULTS
</SectionTitle>
    <Paragraph position="0"> In a test set of narrative hospital discharge summaries that were computer-parsed, 49% of the sentences were syntactically incomplete sentence forms (&amp;quot;fragments&amp;quot;).</Paragraph>
    <Paragraph position="1"> The fragments were of six types that can be related to full sentence forms on the basis of the elements which were regularly deleted: (i) deleted verb and object  (or subject and verb), leaving a noun phrase (4); (ii) deleted tense and verb b__ee (5); (iii) deleted subject, tense, and verb be (6); (iv) deleted subj'ect (7); (v) deleted subject, tense, and verb be (passive predicate) (8); and (vi) deleted subject, tense, and verb be (infinitival complement) (9).</Paragraph>
    <Paragraph position="2"> (4) Stiff neck and fever. (63%) (5) Brain scan negative. (22%) (6) Positive for heart disease and diabetes. (8%) (7) Was seen by local doctor. (5%) (8) Treated for meningitis. (2%) (9) To be followed in hematology clinic.</Paragraph>
    <Paragraph position="4"> Viewing these fragments as deletion forms, it is possible to fill them out to full sentences that would be accepted as paraphrastic by &amp;quot;native speakers&amp;quot; of medical language. For example, occurrences (4)-(9) can be related to sentences (10)-(15) respectively.</Paragraph>
    <Paragraph position="5">  (i0) Patient had stiff neck and fever.</Paragraph>
    <Paragraph position="6"> (ii) Brain scan was negative.</Paragraph>
    <Paragraph position="7"> (12) Patlent/test/exam \[depending on context\] is positive for heart disease.</Paragraph>
    <Paragraph position="8"> (13) Patient was seen by local doctor.</Paragraph>
    <Paragraph position="9"> (14) Patient was treated for meningitis.</Paragraph>
    <Paragraph position="10"> (15) Patient is to be followed in hematology clinic.  In a set of 8 analyzed hospital discharge summaries, we found 41 cooccurrence patterns of the subject-verb-object (SVO) type, stated in terms of 18 participating word classes. (There were further patterns of the host-modifier type and some larger patterns involving connective between SVO types.) The SVO patterns could begrouped into six more general types by defining a &amp;quot;superclass&amp;quot; (RESULT) consisting of the classes SIGN-SYMPTOM, LAB-RES, QUANT, NORMALCY, DIAGNOSIS, and DESCRIPTION that occurred with one of the subject classes PATIENT, BODY-P#~T, TEST. For example, sentence (5) is of the type TEST RESULT, where scan is the name of a test (occurring with BODY-PART word brain as modifier), and ~ (in the class NORMALCY in the superclass RESULT) is the finding of the test. Examples of the sentence patterns are provided in Table 2.</Paragraph>
  </Section>
class="xml-element"></Paper>