<?xml version="1.0" standalone="yes"?>
<Paper uid="C86-1009">
  <Title>I N T E R F A C I L E : Linguistic Coverage and Query Reformulation</Title>
  <Section position="1" start_page="0" end_page="0" type="metho">
    <SectionTitle>
2 Place Jussleu
75005 Paris
I. INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> The experience we have gained in designing \[3,6,7\] and using natural lanauage interfaces has led us to develop a general natural language system, INTERFAC\[LE, involving the following principles : the linguistic coverage must be elementary but must include phenomena that allow a rapid, concise and spontaneous interaction, such as anapho~a (ellipsis, pronouns, etc.) the linguistic competence and limits of the interface must be easily and rapidly perceived by the user.</Paragraph>
    <Paragraph position="1"> the interface must be equipped with strategies and procedures for leading the user to adjust his linguistic competence to the capacities of the system.</Paragraph>
    <Paragraph position="2"> We have illustrated these principles in an application : a natural language (French) interface for acquiring the formal commands of some operating system languages. (The examples given here concern DCL of Digital</Paragraph>
  </Section>
  <Section position="2" start_page="0" end_page="46" type="metho">
    <SectionTitle>
Equipment Corporation).
2. INTERFACILE's ARCHITECTURE
</SectionTitle>
    <Paragraph position="0"> The system is composed of : - a portable syntactic grammar of French written in the formalism of Metamorphosis Grammars \[2\].</Paragraph>
    <Paragraph position="1"> - a domain dependent lexicon-grammar \[4\]. - a domain dependent knowledge base written in a subset of predicate logic (Horn Clauses).</Paragraph>
    <Paragraph position="2"> Operating on these linguistic data and domain knowledge : - a syntactic parser coupled with the grammar and the lexicon-grammar (i) produces intermediate analyses of questions, and (2) generates some messages in case of errors.</Paragraph>
    <Paragraph position="3"> - a semantlc parser transforms the syntactic analysis into a semantic representation by consulting the knowledge base.</Paragraph>
    <Paragraph position="4"> - an evaluator consults the knowledge base and produces answers to questions given their semantic representation. - a general system guides the user and helps him to formulate and reformulate his queries.</Paragraph>
    <Paragraph position="5"> The system is entlrely programmed in PROLOG II and runs on various computers (VAX, SPS and micro-computers). Responses are given instantaneously.</Paragraph>
    <Paragraph position="6"> Eere is an example of a dialogue with  the prompt ..... precedes INTERFACILE's answers. English translations of the dialogues are given in the appendix).</Paragraph>
  </Section>
  <Section position="3" start_page="46" end_page="46" type="metho">
    <SectionTitle>
3. LINGUISTIC COVERAGE
</SectionTitle>
    <Paragraph position="0"> INTERFACILE handles only elementary s~ntactlc structuresdeg In order to ensure a concise and spontaneous dialogue, proforsLs have been introduced such as : 'deg pronouns (le, la, les, en, y, lui, I eur .... ) -possessives (son, sa, ses .... ) demonstratlves (celui-ci, ceux-ci, ce dernier~ . . .) If an ambiguity is detected during the resolui:ion of a profovm, INTERFACILF, asks a question~ as in the fo\].\]owing dialogue : &gt; comment copier un flchier dans un  rSpe~tol~'e ? &amp;quot;- utllisez la commands : COPY nomfichier \[nomrQper'toire\] &gt; comment l'effacer ? * &amp;quot; '~ 1 ' ~' renvoi e ~ : fichier ? (i Pgpertoire ? (2 tapez le num4ro de vo'tre choix. &lt;Dialogue 2&gt; Proform r~soltttlon is based solely on  lexical a~d syntactic information. This strategy Is vital in the framework of a didact.ic system oriented towards beginners as the following dialogue shows : &gt; comment envoyer un message A un  If the proform resolution had been based on the domain semantics, the system would have produced %he answer for&amp;quot; suppressing a message.</Paragraph>
  </Section>
  <Section position="4" start_page="46" end_page="46" type="metho">
    <SectionTitle>
4. QUERY REFORHULATION
</SectionTitle>
    <Paragraph position="0"> Given the plethora of syntactic structures and the complexity of mental processes involved in natural language understanding, the linguistic competence of in%effaces yet remains limited. If we resort to such systems to request ~ service or obtain information that we lack, we generally know much more than the interface itself on the various ways for fo~mulatlng our query.</Paragraph>
    <Paragraph position="1"> Ue argue that the quality of a natural language interface must be estimated not only in terms of its linguistic coverage but also in terms of the procedures for dealing with incorrect, extragrammatical (i.e. correct but not expected b~ the interface) or semantically deviant sentences.</Paragraph>
    <Paragraph position="2"> Knowing that error recovery is a complex task in the &amp;quot;restricted&amp;quot; framework of programming languages (limited syntax and rJ. gorously defined semantics), one can appreciate -the difficulty of endowing natural language interfaces with such capabilities.</Paragraph>
    <Paragraph position="3"> One approach to &amp;quot;the problem of error recoverJng is to &amp;quot;accept&amp;quot; user queries that deviate from the system's expectations and (unbeknown to the user) manipulate them into a form that can be analysed. FoP an overview of this approach, see \[1\]. This method is costly to implement, and moreover }%as 'two theoretical faults : (I) There is no guarantee that the system's interpretation of a &amp;quot;deviant&amp;quot; query will correspond to the user's intention in posing the question.</Paragraph>
    <Paragraph position="4"> (2) There will always be a limit to the degree of deviation the system can tolerate before it gives up trying to analyse a user query. But if the transformation of deviant queries into recognizable structures is performed automatically, the user will *lever know where this limit lles, and may believe that the system can make sense of anything he enters. The distinction between the sentences the system interprets correctly i~nd those which it entirely rejects will seem arbitrary to him.</Paragraph>
    <Paragraph position="5"> The strategy developped and implemented in INTERFACILE is quite different. Our philosophy is not to mislead the user concerning the system's real capabilities, but rather to lead his linguistic competence to that of the interface.</Paragraph>
    <Paragraph position="6"> Here is an example illustrating such a  INTERFACILE explains exactly which part of the question it doesn't understand : an unknown word, a verb with an incorrect preposition or unexpected order of complements, etc. In case oPS spelling errors, solutions for correction are given to the user \[8\].</Paragraph>
    <Paragraph position="7"> To guide the user in reformulating his query, the grammar and the lexicon-grammar are consulte~ as knowledge bases. The user may reformulate his query with the information given by INTERFACILE.</Paragraph>
    <Paragraph position="8"> (At any moment, the user can also obtain the lists of verbs, nouns, prepositions, pronouns, etc.).</Paragraph>
    <Paragraph position="9"> When the user has an erroneous or incomplete view of the application domain, INTERFACILE points out his misunderstanding, as in the  The guidance given here goes beyond linguistic reformulation. It concerns the semantics of the application domain. Note that the information given to the user is not preprogrammed, but is (I) deduced from the knowledge base and (2) generated by the linguistic component (grammar, lexicon-grammar and parsers).</Paragraph>
  </Section>
class="xml-element"></Paper>