<?xml version="1.0" standalone="yes"?>
<Paper uid="C92-4198">
  <Title>A Solution hw the Problem of Interactive Disambiguation</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
Herr6 BLANCI ION
GETA, Institut IMAG (UJF &amp; CNRS)
</SectionTitle>
    <Paragraph position="0"> BP 53X, 38041 Grenoble Cedex, France e-mail : blanchou@imag.fr Extended abstract After the experiences of dialogue based MT systems with ITS \[91, N-Tran 16\] mid KBMT-89 \[5\], the LIDIA project aims at the construction of a mock-up of a personal MT systeln for a monolingual user. One major aspect of rite 1.1DIA project is thus, the study of a dialogue of standardization and disambiguation between the system and the user so as to produce a high quality translation. This dialogue satisfies two properties: its questions are explicit, so no linguistic knowledge is required; its questions are monolingual, ~ no foreign language knowledge is needed. Here, we focus on one part of the di~mbigtmtion process: the disambiguation of the structure prtMuced by the analyser.</Paragraph>
    <Paragraph position="1"> The structure produced by our analyser is called MMC (Multisolution, Multilevel and Concrete).</Paragraph>
    <Paragraph position="2"> Multisolution means that the analyser produces every analysis fitting with the syntagmatic, syntactic and logico-semantic model of the grammar (an example is shown fig. 1). Multilevel means that the same structure consists of three levels of linguistic interpretation, namely rite level of syntactic and syntagmatic classes, the level of syntactic functions and the level of logic and semantic relations. Finally, the stracture is said to be concrete because the original utterance can be found back by a simple left-to-fight reading of the structure. We have taken iuto account three kinds of differettees between the solutions produced for one sentence, and each kind of difference is associated with the name of an ambiguity. We lmve defined ambiguities of syntactic classes (cf fig. 2), ambiguities of geometry (cf fig. 3) and ambiguities of syntactic, logic and semantic decoration (cf fig. 4). We have also defined three principles (SS IlL 1) to order the questkms if there is more titan one to be asked. &amp;quot;Ille first principle is: first of all, fiod out the right segmentation into simple sentences. The second principle is: for each common predicate in the MMC structure, find out the right subject, objects and adjuncts. The last principle is: for each simple sentence, find the right structure.</Paragraph>
    <Paragraph position="3"> With those principles we are able to define a strategy (cf fig. 5). We have also isolated some patterns in the three classes of ambiguity. The class of ambiguities of syntactic classes needs no refinement (SS I11.3.1). On the other hand we create four patterns of ambiguity of geometry (SS I11.3.2) called: verbal coordination, argument structure of rise verb, non verbal coordination, subordination; and three patterns of ambiguity of syntactic, logic attd semantic decoration (SS I11.3.3) called: logico-semantic labelling, argument order of direct transitive verbs, syntactic labellhlg.</Paragraph>
    <Paragraph position="4"> Here is an example with the interpretations for each pattern we have chosen: Problem of cla~'s. Le pilote ferme la I~rte: The firm pilot carries her. The pilot shuts the dcv,)r, Problem of verbal coordination. 11 regarde h~ photo et la classe: He looks at the photograph and file class. lie looks at the photograph and files it.</Paragraph>
    <Paragraph position="5"> Problem of the argument structure of the verb. 11 parle de l'fcole de cuisine: He talks about the cooking school. He talks l?om the cooking school. He talks front the school al~ont cooking.</Paragraph>
    <Paragraph position="6"> Problem of non-w'a'bal coordination. I1 l)rend ties crayous et des cahie~s noirs: He rakes pencils and black notebooks. He takes black peacils and black notelx~oks.</Paragraph>
    <Paragraph position="7"> Problem of subordination. L'6cole de cuisine lyonnaisc est fermOe: The lyounaise cooking school is closed. The school of lyonnaise cooking is closed.</Paragraph>
    <Paragraph position="8"> Problem of Iogico-semantie labelling. Pierre fair porter des ehocolats ,5 Lncie: Pierre lets l,ocie carry chocolates. Pierre gets chocolates to t~e delivered to Lucie.</Paragraph>
    <Paragraph position="9"> Problem of argument order of direct transitive verbs. Quel auteur cite ce couffrencier: Which author this lecturer is quoting7 Which lecturer this author is quoting? Problem of .~yntactic labelling. 11 parle de ta tour Eiffel: Ile is talking about the Eiffel Tower. He is talkiug from the Eiffel Tower.</Paragraph>
    <Paragraph position="10"> For each pattern we have defined a methml to produce the appropriate dialogue (SS 111.3). These methods use two kinds of processing: i)rojection and paraphrase. To build paraphrases we use basically three operators: an operator of semantic replaceluent of occurrence, an o|}erator of iK;runttation of groups of occurrences attd an operator of distribution of occurrences. The examples (SS IV) give an idea.</Paragraph>
    <Paragraph position="11"> In conclusion we can say flint our method is quite simple but Iixed.once attd lbr all. We are goiug to study two points in the uear future. The first one is to reduce rise number of analysis and titus, by getting infinmation ffoln thc user, reduce the tittle to speud ou the disambiguation. &amp;quot;the second is to try to build tools which will allow the linguist, designer of the linguistic part of the LIDIA system, to define its owtt metbt~ls of dis~nnbiguation.</Paragraph>
  </Section>
class="xml-element"></Paper>