<?xml version="1.0" standalone="yes"?>
<Paper uid="C92-3163">
  <Title>J.P.DESCL\]~S CH.JOUIS H.G.OH D.REPPERT</Title>
  <Section position="1" start_page="0" end_page="0" type="metho">
    <SectionTitle>
INTERACTION BETWEEN LEXICON AND IMAGE:
LINGUISTIC SPECIFICATIONS OF ANIMATION
Maryvonne ABRAHAM *~, Jean-Pierre DESCL\]~S i
</SectionTitle>
    <Paragraph position="0"> atelier &amp;quot;Formalisation'&amp;quot; du GDR 957 (CNRS) Sciences Cognitives de Paris. France t</Paragraph>
  </Section>
  <Section position="2" start_page="0" end_page="0" type="metho">
    <SectionTitle>
ABSTRACT
</SectionTitle>
    <Paragraph position="0"> Certains textes d'une langue peuvent 6tre considdrds comma des spdcifications linguistiques capables d'engendrer des animations qui simulent la comprdhension des taxies d'entr6e. Pour rdafiser un tel programme, il nous sernble indispensable de construire des reprdsentations intermddiaires compatibles, d'un c6t6 avec les descriptions sdmantiques d'unitds linguistiques, et d'un autre c6td avec des spdcifications d'animarion d'images.</Paragraph>
    <Paragraph position="1"> Comprendre un texte relatif h des mouvements spatiaux revient: i) it reprdsenter sdmantiquement le texte; ii) it engendrer une animation mettant en jeu des images; caste animation vise it reproduire ce qui eat compris lots de la lecture du texte. Ce programme ambitieux suppose qu'au prdalable chaque verbs sit des reprdsentations qua nous appelons sch6mes sdmantico-cognitifs (SSC), correspondant aux diffdrentes significations dn verbe. Tout schdma de mouvement associ~ it la signification d'un verbs repr6sente trois types de connaiesances: i) des ddclarations et des relations invariantes pendant le mouvement; ii) la description cindantique du mouvement qui fait passer d'une phase initiale (Sih) it une phase finale (Sit2); iii) les conditions dynamiques dventuelles qui rendent possibles ou qui contrSlent le mouvement cindmatique.</Paragraph>
    <Paragraph position="2"> Nous reprenons l'approche de R.Schank ou de J.Sowa qui remet en cause une association directe mot-concept pour adopter une d6composltion de la signification des unitds linguistiques en 61dments de Usens&amp;quot; plus petits. Notre formalisms a pour but d'une part, de mieux traiter les probl/~mes de temps, et d'autre part, de fonder notre choix de primitives, dana le cadre d'une th6~rie qui articule its niveaux de reprdsentation linguistiques et cognitifs.</Paragraph>
    <Paragraph position="3"> Introduisons quelques distinctions: I- le verbs est une entitd lexicale qui peut ~tre polysdmique, ex: circular.</Paragraph>
    <Paragraph position="4"> 2- le verbe syntaxique entre dans une construction syntaxique, ex: NI circular prep Nloe. 3- le prddicat logique n-airs est aesocid au verbe syntaxique, ex: CIROULER est l'expression du prddicat l-airs CIRCULER1.</Paragraph>
    <Paragraph position="5"> 4- A un prddicat correspondent un ou plusieurs sens, d6signds par /verbei/, reprdsent~s chacun par un SSC. Un SSC est organisd g par-Sir de primitives cognitives; il conduit it une reprdsentation cognitive dAs qu'il est instanci6. 5- l'archdtype cognitif notd/verbe/, s'il exists, sc prdsente alors comae &amp;quot;le sans abstrait&amp;quot; commun it tous Its SSC d'une m6me entrde lexicah polys~mique. La construction de l'arch6type repose sur une organisation des SSC en rdseau.</Paragraph>
    <Paragraph position="6"> Ella utilise une description analogue it cells de J.Picoche.</Paragraph>
    <Paragraph position="7"> A partir de l'exemple d'un texts fran~ais qui reprdsente des mouvements spatio-temporels, nous donnerons pour quelques verbe du texte lea SSC et les animations correspondantes. L'animation complete ddclenchde par la comprehension de texts ndcessite cette premiere phase de recherche. Les sch6mes que nous propesons sont int6grables darts un dictionnalre s~mantique du fran~ais.</Paragraph>
    <Paragraph position="8"> Notre 6tude presenters /t partir de I'exemple choisi la mdthoclologie appuyde par une rdalisation. mots-cl~s: { comprehension de tezte, repr&amp;entation des connaissances dynamiqttes, leziqaC/ verbs/, m~thodolofie, cholz dt primitives sdmantlco-toqnitiees, eerbes de mouvement, analyst d'un It=St, dictionaaire s~mantiqtte, sp~cifitation linfeistiqae st animation } *D6partement Math~matique~ et Syet~m~ de Communication, LIASC, Telecom Bretagne, BP 832, 29285 Bre~t cedex, tel: 98 00 11 11 - FAX: 98 00 10 98 e-mail: ABRAHAMObrtst.enat.bretasne.yr tC.A.M.S., unitd mixte CNRS - EHF~S - Universit6 Paris-Sorbonne, 96 Boulevard Raspail, 750~ Paris - tel (16-1) 45 48 08 04 - FAX : (16-1) 40 49 07 98 - e-mail : ishaO\[rci651.bitnel Acr~ DE COLING-92. NANTEs. 23-28 AOt~' 1992 1 0 4 3 PRoc. OF COLING-92. NAN'rES. AUG. 23-28. 1992 key.words: understanding a text, dynamic know--l~ge representation, lexicon of verbs, methodology, choice of semantico-cognitive primitives, verbs of motion, text analysis, semantic dictionary, linguistic specification and animation.</Paragraph>
    <Paragraph position="9"> 1 understanding a text and a moving picture Certain texts of a given langage can be considered as linguistic specifications able to produce motion pictures simulating the understanding of entry texts. To achieve such a program, we think that it is necessary to build intermediate representations which are compatible, on the one hand with the semantic description of linguistic units, and on the other hand, with specifications for the moving of pictures.</Paragraph>
    <Paragraph position="10"> other approaches contest this direct association word- concept, and start from a breakdown of the signification of the linguistic units l, following R.Schank in the 1975's. The meaning of the words is no longer considered as primitive but can be split up into smaller siguifica~ tion units. More recently, Sown has presented a formalism of representation based on graphs with two sorts of nodes representing concepts and relations. However, many researchers feel that the choice of the primitives looks arbitrary (\[SABS8\], pp.241-243). We propose a formalism similar to Sowa's graphs \[SOW84\], but we try to process time better and establish our choice of primitives z, using a theory which arranges links between the levels of linguistic and cognitive representations.</Paragraph>
    <Paragraph position="11">  age Understanding a text relative to spatial motion consists of: i) a semantic representation of this text; ii) movement with pictures; this animation aims to reproduce what is understood during the reading of the text. As a first step, this ambitions program requires that each verb should have representations which we call semantico-cognitive schemas (SCS). These representations correspond to the different meanings of the verbs. Each scheme of motion associated with the meaning of a verb represents three kinds of knowledge: i) declarations and relations which are invariant during the motion; ii) the cinematic description of the motion from an initial phase (Sitl) to a final phase (Sit2); iii) the dynamic conditions which make or control kinetic motion.</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
1.1 semantico-cognitive representa-
</SectionTitle>
      <Paragraph position="0"> tions Although several formalisms of knowledge representations (e.g. semantic networks) suppose that the word is the elementary semantic unit,  rico-cognitive schemes (SCS) The cognitive archetypes (or semantico-cognitive schemes, written SCS, see figure 2), were introduced in 1985 \[DES85\] in order to represent the verbal knowledge underlying the language. Our method to discover them is to collect linguistic data which allows us to characterize prototypic schemes of meanings from which an abstract invariant may appear, in some cases.</Paragraph>
      <Paragraph position="1"> The following levels can be taken into consideration: null  creases S.Schaumyan's initial model with more cognitive levels where we aim to discern operations about cognitive categories; We have described the maln primitives (MOUVT, FAIllE, CONTROL/g, ...) which are used in several publications DES90 ; see \[JPD911 for time processing, and \[DES91 for the theory underlying the choice of the primitives. ACl~ DE COLING-92. NANTES, 23-28 AOUT 1992 1 0 4 4 PROC. OF COLING-92. NANTES. AUG. 23-28, 1992 of a syntactic construction (e.g., N1 sortir prep I%,). 3- The n-ary predicate is associated with the syntactic verb (e.g., SORTIR1) a.</Paragraph>
      <Paragraph position="2"> 4- With a predicate, one or several meanings are associated, written /verb//4. A SCS is organised from cognitive primitives. It produces a cognitive representation the moment it is instanciated. null 5- Then, if it exists, the cognitive archetype (written/verb/) appears as the &amp;quot;abstract meaning&amp;quot; common to each of the SCS of one lexical polysemic entry. Its construction is organised from SCSs. It uses a description similar to the one in J.Picoche's method \[PIC86\] 5.</Paragraph>
      <Paragraph position="3"> From a sample text representing spatio-temporal motion, we give the semantics-cognitive schemes and the corresponding motions for each verb of the text. This first step is required in order for an understanding of the text to trigger complete animation.</Paragraph>
      <Paragraph position="4"> 2 A few verbs to represent a text The schemes that we propose are integrated in a semantic dictionary of French. From a chosen example, our study presents the methodology based on demonstration. To illustrate our method, we analyse the verbs in the following French text and we give cognitive representations of them: Madame Dupont circulalt dana use petite rue ~ sees unique od stationnaient des voitures de ehaque c6t&amp; Un vdhicule ap_r3dt_rd d'un parking surgit brusque.</Paragraph>
      <Paragraph position="5"> meet sur sa droite.</Paragraph>
      <Paragraph position="6"> which means: Mrs Dupont__was driving along a small one-way street where cars were parked on each side. A vehicle leaving a car park shot out on her leff .</Paragraph>
      <Paragraph position="7"> sin agreement with the conceptual language of R.Jackendoff \[JAC83\] 4i indexes one of the prototypical meanings of the polyeemes.</Paragraph>
      <Paragraph position="8"> SSee two examples a~eociated with the different meaningB of the French verbs fouler \[ABR92\] and raonter (J.P.Descl~!s, article to be published). We need syntactic descriptions such as those of M.Grosa (\[GRO81\]). For identifying the SCSs, we use work8 on semantics such a~ 1COL83\] and \[MEL84\].</Paragraph>
      <Paragraph position="9"> 6Notice that we make more than a lexical translation from French into English: cognitive laws of the highway code are ~translated ~ too.</Paragraph>
      <Paragraph position="10"> As we have not enough room in this paper to describe each verb of this text, we have chosen to describe the French verbs stationner, and surer.</Paragraph>
      <Paragraph position="11">  In x stationne, we have a static positional relation. This situation is implicitly bound by a constraint (which can be more or less explicit). The signification of/stationner/in this proposition is given by the dynamic scheme in figure  3. The quantification which binds the variable y blocks every instanciation of y. In our example, tile SCS is instanciated with:</Paragraph>
      <Paragraph position="13"> A constraint is needed to keep a car parked. It is expressed here using the primitive &amp;quot;FAIRE&amp;quot; which keeps the movement from Sit1 to Sit2 in the same place. We say that /x stationne/ if the behaviour of x implies a constraint on its own movement which keep it in the same place.</Paragraph>
      <Paragraph position="14"> This constraint is expressed by the equality between the localisations loci of Sit1 and loc2 of Sitz.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2.2 SCS of the French verb surgir
</SectionTitle>
      <Paragraph position="0"> x surgit means a sudden transition, a movewent entering the perception zone of an entity able to perceive. The perception zone is generally taken from the speaker's point of view. We have a rule (IDES901 pp.285-287, \[ABR91\]SS2.1) concerning localisation: AcYgs DE COLING-92, NANTES, 23-28 AOfrr 1992 I 0 4 5 PROC. OF COLING-92, NANrES, AUG. 23-28, 1992 d~clarations: indivldu: x, y; lieu:</Paragraph>
      <Paragraph position="2"> rule: if z is an entity typed lndlvldu z determines a localisation loc(z} including the entity Z.</Paragraph>
      <Paragraph position="3"> In fact, to each x is associated a family of localisations, among which a proper loealisation, a peripheral localisation depending on cultural habits, and a neighbouring localisation dependent on the context. The localisation loe (x) can be determined by taking its internal area, its external area, or its frontier. The family of localisations can be extended to the localisations perceived by the individual entity. definition of perception: The localisation PER-CEP(x) is defined as a result of the intersection 7 of an external area to be perceived and of the properties of perception of the person perceiving (its senses).</Paragraph>
      <Paragraph position="4"> x surgit if x moves from Sitx to Sit2. In Sitl, x is situated in any localisation; in Sit2, x is situated in the place of perception of the witness, or, by default, in the area of perception of the speaker.</Paragraph>
      <Paragraph position="5"> /surgir/can be associated with a preposition, for example, sur (on), which partially specifies the place of arrival. The French relation ETRE-SUR can be represented formally by the following set of relations s: definition of the relation ETRE-SUR: i) The one referred to is outside a domain D (closed localisation) ii) The one referred is in contact with the external frontier of the domain D.</Paragraph>
      <Paragraph position="6"> iii) An czternal action (by default, gravitation) keeps the referred one in contact with this external frontier.</Paragraph>
      <Paragraph position="7"> VPERCEP(x) can be considered ae the result of the interjection of the area SENSE(x) (of an entity that x can perceive) and of the external domain which is examined. VIS(A) is the vhdbility area of A.</Paragraph>
      <Paragraph position="8">  Concerning visual space perception (figure 5), the notions of right and left are relative: they can be organised from the point of view either of the subject observed, or of the speaker. Let us take an orientation from the subject observed A, here, Madame Dupont, which has an intrinsic orientation. A defines FRONT(A) REAR(A) LEFT(A), RIGHT(A), VIS(A). The intersection of these areas defines frontiers. In our example, we can consider that /to be on the left of Mrs Dupont/means /to be on the external frontier between the areas VIS(Mrs Dupont) and LEFT(Mrs Dupont)/.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>