<?xml version="1.0" standalone="yes"?>
<Paper uid="E87-1029">
  <Title>AUXILIARIES AND CLITICS IN FRENCH UCG GRAMMAR</Title>
  <Section position="2" start_page="0" end_page="174" type="intro">
    <SectionTitle>
ABSTRACT
</SectionTitle>
    <Paragraph position="0"> French auxilliaries and clitics have been analysed in the flame of U.C.G. (Unification Categorial Grammar).</Paragraph>
    <Paragraph position="1"> Concatenation of a functor sign and an adjacent argument sign is the basic operation of the model ; unification allows (a) to verify ff constraints on concatenation are respected ; (b) to produce a flow of information between the functor sign and the argument sign.</Paragraph>
    <Paragraph position="2"> The rules of the grammar and the design structure of the sign allows to express : (a) the concatenation between French auxilliaries (are and avoir) and the participle verb form within a single pattern, (b) transitions between clitics in a systematic way. Two complex questions of French syntax are thus covered in a fairly simple way.</Paragraph>
    <Paragraph position="3"> The UCG Model U(nification). C(ategorial) G(rammar) is a new grammatical model proposed by an Edinburgh team headed by Ewan Klein in \[CALDER 86\] and \[ZEEVAT 86\]. UCG is a feature grarnmar incorporating some basic insights from GPSG \[GAZDAR 85\] and HPSG \[POLLARD 84\]. Functional application applies in UCG as in categorial grammars; it allows for concatenation of a functor with an adjacent argument. Unification is a basic operation which allows (a) to verify if constraints on concatenation are respected; (b) to produce a flow of information between functor and argument. This information together with some defined aspects of the information carried by the functor, will be finally inscribed in the resulting concatenated sign.</Paragraph>
    <Paragraph position="4">  arglist -&gt; agrl ..... argn The work reported here was carried out as part of ESPRIT Project 393 (ACORD), 'The construction and Interrogation of Knowledge Bases using Natural Language Text and Graphics&amp;quot; order -&gt; pre \[ post phonology-&gt; &lt;lexical_item&gt; In graph notation, a UCG sign can be represented (in a slightly simplified form, relevant to this paper) as in figure I.</Paragraph>
    <Paragraph position="5"> In this figure the leaves of vertical branches columns (i) through (viii) - denote the values of the corresponding labels in its upper portion.</Paragraph>
    <Paragraph position="6"> We have :  (i) to (vi) Simple categories : sent~_:nil, noun~:nil. (ii) Features on (i) (see below) (iii) values for the CL(itics) label are: prod (dialogue pronouns, for me , re, noua, vous); protob (third person object pronouns : le, la, /es); prota (third person dative pronouns : lui, leur) ; se, en and y, (for se, en and y pronouns respectively); n is a barrier symbol (see below).</Paragraph>
    <Paragraph position="7"> (iv and v) values for morphological aspects of the sign: (v) categorizes signs in lex(ical) and pron(ominal) ones, (iv) in maac(ulin) and fem(inin), in sing(ular) and pl(ural), and introduces values for the 3 persons. (vi) the subcaflist : the label C will denote the typical variable for it.</Paragraph>
    <Paragraph position="8"> (vii) the index sort system (which is not exhibited here)  allows selection on semantic features while a special field (pr res) contains information (agreement, class and COR) for the pronoun resolution component.</Paragraph>
    <Paragraph position="9"> COR(eference) is intended to prepare the semantic representation for pronoun resolution. The corresponding values are : obl(igatory), for bound anaphora as in se ; ind(ependent), for NP nominals of indicative sentences and dep(endent) for NP nominals of subjunctive ones (the algorithm for pronoun resolution will not be presented here, but the semantic representation specified by the proposed grammar is intended to carry all the required relevant information).</Paragraph>
    <Paragraph position="10"> (viii) post and pre are the values for order ; they are essential for handling word order and for the application of the grammar roles.</Paragraph>
    <Paragraph position="11"> In the unification process and in the generation of the subsequent flow of information, the labels Class, Ge, Nb and Pe denote variables for the corresponding values, Clo the variable for clitic placement value and O, the variable for order values.</Paragraph>
    <Paragraph position="12"> The two following are the French UCG signs for a/me and Mar/e :  We added two rules to these, inspired by functional composition as described in \[STEEDMAN 86\].  BC is designed to deal with free-order of riparguments of verbs Forward application must be interpreted as follows :  If a sign of string Wl and category HF:CF/(HA:CA) unifies with a sign of string W2 and category HA:CA, W1 concatenates with W2; the resulting sign, with string \[-W1,W2\], is of category HF:CF, where HF:CF is the category inherited from the functor as resulting from unification with its argument, and stripping HA:CA. Mutadis mutandis, analogous interpretations must be given to (5) through (7).</Paragraph>
    <Paragraph position="13"> By definition (3) HA:CA in HF:CF/(HA:CA) of (4) must be a sign; it is the active part of the functor. The final concatenated sign is obtained by stripping the active part of the functor as instantiated by the argument.</Paragraph>
    <Section position="1" start_page="174" end_page="174" type="sub_section">
      <SectionTitle>
Example
</SectionTitle>
      <Paragraph position="0"> For example: (8) is the instantiation by BA of (2b) as the functor with respect to (2a) as the argument of the rule; (9) is the resulting sign, obtained from (8) by stripping; (10) represents the sign of the whole sentence  The semantics of UCG incorporates the basic insights of Kamp's DRT \[KAMP 81\] but the introduction of indexes greatly increases the expressive power of semantic representations (cf. \[ZEEVAT 86\]).</Paragraph>
      <Paragraph position="1"> To resume : The whole model is based on :  information and to verify conditions : unification. * similar ways to combine a functor and its argument to give a resulting sign.</Paragraph>
      <Paragraph position="2"> The French sentence simple verbs They accept left-placed arguments (as clitics) and rigth-placed ones (as lexical ones).</Paragraph>
      <Paragraph position="3"> composed verbal forms No argument can be inserted between the auxiniary and the participle form.</Paragraph>
      <Paragraph position="4"> Whereas in English only one auxiliary is used to construct perfect tenses, French uses avo/r and ~tre depending on the main verb. Furthermore, ~tre is also used for passive constructions.</Paragraph>
      <Paragraph position="5"> The most important problem, however, is due to the agreement of the past participle with the subject of the main verb when used with ~tre, but with the object -only if it precedes the auxiliary- when used with avoir. However, we succeeded to maintain a single lexical entry for a verb, allowing for the different order of arguments. This is made possible by the introduction of forward and backward composition rules.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>