<?xml version="1.0" standalone="yes"?>
<Paper uid="P84-1102">
  <Title>Disambiguating Grammatically Ambiguous Sentences By Asking</Title>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
NUMBER ?
</SectionTitle>
    <Paragraph position="0"> The technique to implement this, which is described in the following sections, is called Explanation List Comparison.</Paragraph>
  </Section>
  <Section position="4" start_page="0" end_page="478" type="metho">
    <SectionTitle>
2. Explanation List Comparison
</SectionTitle>
    <Paragraph position="0"> The basic idea is to attach an Explanation Template to each rule.</Paragraph>
    <Paragraph position="1"> For example, each of the rules &lt;a&gt; - &lt;g&gt; would have an explanation template as follows:  (1) is a noun (1) is a determiner of (2) (1) is (2) (1) is a preposition of (2) (2) is an object of the verb (1)  tThi: lesearcn was sponsored by the Defense Advanced Research Projects :~ger',:y {('~O~3), ~.PP.,'~ Order No. 3597, monitored by the Air Force Avionics Lahor;llor~, !JnOer Contract F3.3615-81 K-1539. The views and conclusions c~,lte,:l~J in fi=is d~cumnnt are those ef the authors and should not be interpreted as reor.e~,~ntinq the official L)olicies. eilher expressed or implied, of the Defense AdvanceO Rgsearch Projects Agency or the US Government. 2personal communication.</Paragraph>
    <Paragraph position="2"> Whenever a rule is employed to parse a sentence, an explanation is generated from its explanation template. Numbers in an explanation template indicate n-th constituent of the right hand side of the rule. For instance, when the rule &lt;f&gt;</Paragraph>
    <Paragraph position="4"> matches &amp;quot;with a telescope&amp;quot; (*prep = &amp;quot;WITH&amp;quot;; NP = &amp;quot;s  te'lescope&amp;quot;), the explanation &amp;quot;(with) is a preposition of (a telescope)&amp;quot; is uenerated. Whenever the system builds a parse tree, it also builds a list of explanations wnich are generated from explanation templates ot all rules employed. We refer to such a list as an explanation list. the explanation lists of the parse trees in the example above are:  In order to disambiguate a sentence, the system only examines these Explc, nation Lists, but not parse trees themselves* This makes our method independent from internal representation of a r~a~se tree. Loosely speaking, when a system produces more than erie parse tree, explanation lists of the trees are &amp;quot;compared&amp;quot; and the &amp;quot;diliere,~ce&amp;quot; is shown to the user. The user is, then, asked to select the correct alternative.</Paragraph>
    <Paragraph position="5"> 3. The revised version of ELC Ur, fortunately, the basic idea described in the preceding section does not work quite well. For instance, the difference of the two explanation lists in our example is  despite the fact that the essential difference is only 1) The action (Mary saw a man) takes place (with a telescope) 2) (a man) is (with a telescope)  Two refinement ideas, head and multiple explanations, are introduced to solve this problem.</Paragraph>
    <Paragraph position="6"> 3.1. Head We define head as a word or a minimal cluster of words which are syntactically dominant in a group and could have the same syntactic function as the whole group if they stood alone. For example, the head of &amp;quot;VERY SMART PLAYERS IN NEW YORK&amp;quot; is &amp;quot;PLAYERS&amp;quot;, and the head o! &amp;quot;INCREDIBLY BEAUTIFUL&amp;quot; is &amp;quot;BEAUTIFUL&amp;quot;, but the head of &amp;quot;1 LOVE CATS&amp;quot; is &amp;quot;1 LOVE CATS&amp;quot; ilk, elf. The idea is that. whenever the system shows a part of an input sentence to the user, only the ilead of it is shown. To implement this idea, each rule must hove a head definition besides an explanation template, as follows.</Paragraph>
    <Paragraph position="7">  For instance, the head definition of the rule &lt;b) says that the head of the construction &amp;quot;NP + VP + PP&amp;quot; is a concatenation of the head of 1.st constituent (NP) and the head of 2-nd constituent (VP). The i~ead of &amp;quot;A GIRL with A RED BAG saw A GREEN TREE WITH a telescope&amp;quot; is, therefore, &amp;quot;A GIRL saw A TREE&amp;quot;, because the head of &amp;quot;A GIRL with A RED BAG&amp;quot; (NP) is &amp;quot;A GIRL&amp;quot; and the head of &amp;quot;saw A GREEN &amp;quot;IREE&amp;quot; (VP) is &amp;quot;saw A TREE&amp;quot;. in our example, the explanation (Mary) is a subject of the action (saw a man with a telescope) becomes (Mary) is a subject of the action (saw a man), and the explanation (a man with a telescope) is an object of the verb (saw) becomes (a man) is an object of the verb (saw), because the head of &amp;quot;saw a man with a telescope&amp;quot; is &amp;quot;saw a man&amp;quot;, and the head of &amp;quot;a man with a telescope&amp;quot; is &amp;quot;a man&amp;quot;. The difference of the two alternatives are now:  should have only one explanation template: (1) &amp;quot;Is a subject of the actton (2).</Paragraph>
    <Paragraph position="8">  With the idea of head and multiple explanations, the system now produces the ideal question, as we shall see below.</Paragraph>
    <Section position="1" start_page="476" end_page="478" type="sub_section">
      <SectionTitle>
3.3. Revised ELC
</SectionTitle>
      <Paragraph position="0"> To summarize, the system has a phrase structure grammar, and each rule is followed by a head definition followed by an arbitrary number of explanation templates.</Paragraph>
      <Paragraph position="1">  &lt;g&gt; \[t 2\] (2) is an object of the verb (1) With the ideas of head and multiple explanation, the system builds the following two explanation lists from the sentence &amp;quot;Mary saw a man with a telescope&amp;quot;.</Paragraph>
      <Paragraph position="2">  The difference between these two is The action (Mary saw a man) takes place (with a telescope) and (a man) is (with a telescope).</Paragraph>
      <Paragraph position="3"> Thus, the system can ask the ideal question:  1) The action (Mary saw a man) takes place (with a telescope) 2) (a man) is (with a telescope) Number?.</Paragraph>
      <Paragraph position="4"> 4. More Complex Example  The example in the preceding sections is somewhat oversimplified, in the sense that there are only two alternatives and only two explanation lists are compared. If there were three or more alternatives, comparing explanation lists would be not as easy as comparing just two.</Paragraph>
      <Paragraph position="5"> Consider the following example sentence: Mary saw a man in the park with a telescope. This s~ntence is ambiguous in 5 ways, and its 5 explanation lists are shown below.</Paragraph>
      <Paragraph position="6">  The action (Mary saw a man) takes place (with a telescope) The action (Mary saw a man) takes place (in the park) : : With these 5 explanation lists, the system asks the user a question twice, as follows:  1) (a man) is (in the park) 2) The action (Mary saw a man) takes place (in the park) NUMBER? 1 i) (the park) is (with a telescope) 2) (a man) is (with a telescope) 3) The action (Mary saw a man) takes place (with a telescope)  NUMBER? 3 The implementation of this is described in the following. We refer to the set of explanation lists to be compared, {/1' L2 .... }, as A. If the number of explanation lists in A is one ; jusl return the parsed tree which is associated with that explanation list. If there are more than one explanation list in A, the system makes a Qlist (Question list). The Qlist is a list of explanations Qlist = { e I, e 2 ..... en} which is shown to the user to ask a question as follows:</Paragraph>
      <Paragraph position="8"> Qlist must satisfy the following two conditions to make sure that always exactly one explanation is true.</Paragraph>
      <Paragraph position="9"> * Each explanation list / in A must contain at least one explanation e which is also in Olist. Mathematically, the following predicate must be satisfied.</Paragraph>
      <Paragraph position="10"> VL3e(e E L A e E Qlist) This condition makes sure that at least one of explanations in a Qlist is true.</Paragraph>
      <Paragraph position="11"> * No explanation list L in A contains more than one explanation in a Qlist. That is,  ~(gLgege'(L E AAeEL Ae'EL A e G Qlist A e' E Qlist A p =e') This condition makes sure that at most one of explanations in Qlist is true.</Paragraph>
      <Paragraph position="12"> The detailed algorithm of how to construct a Qlist is presented in Appendix.</Paragraph>
      <Paragraph position="13"> Once a Olist is created, ~t is presented to the user. The user is asked to select one correct explanation in the Qlist, called the key explanation. All explanation lists which do not contain the key explanation are removed from A. If A still contains more than one explanation list, another Olist for this new A is created, and shown to the user. This process is repeated until A contains only one explanation list.</Paragraph>
    </Section>
  </Section>
  <Section position="5" start_page="478" end_page="478" type="metho">
    <SectionTitle>
5. Concluding Remarks
</SectionTitle>
    <Paragraph position="0"> An experimental system has been written in Maclisp, and running on Tops-20 at Computer Science Department, Carnegie-Mellon University. The system parses input sentences provided by a user according to grammar rules and a dictionary provided by a super user. The system, then. asks the user questions, if necessary, to disambiguate the sentence using the technique of Explanation List Comparison. The system finally produces only one parse tree of the sentence, which is the intended interpretation of the user. 1he parsor is implemented in a bottomup, breath-first manner, but the idea described in the paper is independent from the parser implementation and from any specific grammar or dictionary.</Paragraph>
    <Paragraph position="1"> The kind of ambiguity we have discussed is structural ambiguity.</Paragraph>
    <Paragraph position="2"> An ambiguity is structural when two different structures can be bui!t up out of smaller constituents of the same given structure and type. On the other hand, an ambiguity is lexical when one word can serve as various parts of speech. Resolving lexical ambiguity is somewhat easier, and indeed, it is implemented in the system. As we can see in the Sample Runs below, the system first resolves lexical ambiguity m the obvious manner, if necessary.</Paragraph>
    <Paragraph position="3"> Recently, we have integrated our system into an English-Japanese Machine Translation system \[3\], as a first step toward user-friendly interactive machine translation \[6\]. The interactive English Japanese machine translation system has been implemented at Kyoto University in Japan \[4, 5\].</Paragraph>
  </Section>
class="xml-element"></Paper>