HOW TO VISUALIZE TIME, TENSE AND ASPECT? COMMENT VISUALISER LE TEMPS ET L'ASPECT?

Transcription

1 HOW TO VISUALIZE TIME, TENSE AND ASPECT? COMMENT VISUALISER LE TEMPS ET L'ASPECT? GErard LIGOZAT - Michael ZOCK LIMSI, Langage & Cogniion B.P. 133, Orsay Cexiex / France courier 61eeronique : limsi.fr, ou, limsi.fr Nous allons decrire ci-dessous une exension d'un sys~me ineracif de generaion de phrases. Apr~s une descripion de SWIM,--acronyme de See Wha I Mean?-- nous allons discuer la fa~on don on communique au sys~me des noions relaives au emps e ~ l'aspec. Avec SWIM nous poursuivons rois objecifs : (a) assiser un Eudian dans rapprenissage d'une langue, plus prdcisemen aider l'apprenan h passer des idees aux phrases (producion du langage); (b) visualiser la relaion sens-forme.; (c) fournir un environnemen voisin des condiions dans lesquelles on apprend naurellemen une langue. Savoir parler une langue n'es pas inn, c'es le resula d'un apprenissage. Selon le cadre on parle d'apprenissage naurel (on apprend dans le pays mme 05 es uilisee eee langue) ou d' apprenissage insiuionnel (enseignemen). Une difference fondamenale enre les deux ypes d'apprenissage se siue au niveau du conr61e. Dans un cas rapprenan es ac ± f (apprenissage naurel), - il uilise la parole pour rsoudre un problsme le concernan (il cherche h obenir des informaions, il veu verifier quelque chose, ec.) - andis que dars raure cas il es esseniellemen reac i f, c'es-h-dire qu'il ne parle que lorsqu'011 le lui demande. L'inconvEnien majeur de 1 'apprenissage naurel es lid au fai que l'apprenan se rouve dans des siuaions off ou change consammen : la siuaion discursive, les bus ou les inenions de communicaion, les conenus e les formes linguisiques correspondanes. Cela es normal, car l'objeeif es la communicaion -- faire passer un message en vue d'un bu donne ~ e non la fixaion, rexplicaion ou rillusraion sysemaique d'lm poin precis de la langue. Quanh l'apprenissage dans un conexe scolaire (enseignemen), les inconveniens en son les suivans: (a) la siuaion es arificielle (on s'exerce pour apprendre la langue e non pour communiquer un message en rue d'un bu donn6); (b) le professeur ignore gndralemen la naure des besoins de l'e16ve. Si l'e16ve sai ce qu' 11 veu dire, en revanche, il ignore souven coranen le dire. Quan au professeur, il sai conmwn le dire, mais il ignore ce que veu dire reudian. Pour rsoudre ce probl~me de communicaion enre le professeur e reudian nous avons con~u un sysme oh l'elsve peu poser des quesions e off le professeur repond. On noera, que c'es reudian qui a riniiaive du dialogue. C'es lui qui pose des quesions, e il les pose en foncion de ses besoins. A l'beure acuelle le sysme a plusieurs faiblesses e un des grands problmes es celui de l'inerface concepuelle : commen lui comnmniquer le conenu du message, puisqu'on ne parle pas encore cee lmlgue? Nous monrons dans la deuxime parie de ce papier la fa~on don l'dlsve communiqne des informaions relaives au emps e h l'aspec. Ces infonnaions son d61icaes h ransmere pour au moins rois raisons : (a) il n'y a pas d'isomorphie enre les emps concepuels e les emps grammaicaux (le "presen" peu aussi bien exprimer des 6vEnemens qui on eu lieu que des 6vnemens qui von avoir lieu); (b) il n'y a pas de correspondance srice enre les emps des differenes langues (c) la enninologie uilisee dans les manuels es rop absu'aie. Elle pose des problmes d'une manisre generale, e h plus fore raison lorsque la caegorie mdalinguisique uilisee n'a pas de correspondan dans la langue cible (le passd simple, par exemple, n'exise pas dans oues les langues). Paran des ravaux de Reichenbach e de Vendler nous avons conqu un langage ic6nique pour communiquer les informaions relaives au emps. Pour ce faire, nous avons complee leurs iddes ou en les rendan plus accessibles ~ l'uilisaeur (mea) linguisfiquemen innocen. Mos clefs : enree concepueile, generaion de phrases, relaions fonne-sens, EIAO ACRES DE COLING-92. NANTES, AO{n' $ PRo(:. OF COLING-92. NAI'CrEs. AUG , 1992

2 1 Inroducion The major goal of he sysem described here is o provide a naural envirc~'~nen I for learning o produce senences in French. In order o achieve his goal, we have inegraed he suden ino he process. He should be acive raher han reacive, ha is, he should alk no only when being asked. Afer all, he knows wha he wans o say, wha he lacks are he means of ranslaing his houghs ino language. SWIM 2 is an exploraory environmen which has been buil o simulae naural learning. The suden asks he quesions and he sysem answers hem. Any of he following quesions could occur in a naural seing : 3 1) How does one say ±c~a.9 2) Can one say linguisic fo~m? 3) How should one say idea.9 4) Why does one say fen'n-1 and no form-2? 5) Wha would happen if concepual recificaion? 6) Wha would happen if synacic modificaion? 7) Wha would happen if word y insead of word x? 8) Wha is he difference beween form xand fozaxa y? leaning-represenafio: lia~uhl decimlv p~l Ix gargon.'a pas mgarde~ ies filles.., g= o. n'a pa* ~ Figure 1 les filles. 1) L&~ ~r S~dclafOe. [21 IX' ilalpm ~" PU i~8~d4 I fi[~ ] J ) IL ~6~ "'a pas regard# I~ fi/lo. 1 A compuerized learning environmen is said o be naural if he ransfer of knowledge is achieved essenially in rie same way as could be he case beween wo human agens, where a learner (child. suden) asks quesions and he eacher (naive speaker, paren/eacher) answers hem. 2 SWIM is he acronym of See Wha I Mean? I has been conceived by Michael Zock and implemened by A. Laroui on a Macinosh. 3 For he ime being only he firs and las four ypes of quesions are implemened for a small subse of French (simple senences). The firs communicaion-mode (how does one say <idea>?) raises an ineresing problem : how o ell he compuer wha one wans o say (message). The fundamenal quesion here is in wha erms o code hough and access i. Obviously, here are several ways. One could ask in anoher naural language (ranslaion), one could use images, or one could use some form of mealanguage. In his paper we discuss some of he problems ha arise by using a mealauguage. We hen show how a special caegory of hough, ense, can be communicaed by nsing a hybrid form of represenaion. Bu, before doing so le us give a shor descripion of he sysem. For more deails see [ 29, 30, 31 ]. 2 Descripion of he sysem The dialogue is iniiaed by specifying he communicaion mode. Le us suppose ha he user had sared wih he firs quesion: How does one say <idea>? In ha case he sysem presens a menu (ype hierarchy) from which he user has o choose: Speech ac : saemen, quesion, command Acions - processes: axonomy of verbs ense : presen, pas, fuure, ec. Discourse objecs : axonomy of nouns number : singular, plural communic, saus : definie, indefinie, Aribues: axonomy of aribues By choosing specific values from a se of alernaives, he user ells he sysem wha he wans o say. 4 As he dialogue develops, he sysem builds he underlying meaning in he form of a semanic nework. I hen invies he user o ry o express his meaning, afer which i oupus is own form. As one ca* see from Figure 1, he screen is divided ino five pars. The large window a he op represens he underlying meaning of he message a suden is rying o convey. The nex wo windows conain respecively he user's aemp o express his meaning and he sysem's version. Possible mismaches beween hese wo versions are highlighed on he screen. This allows he sysem o draw he suden's aenion o errors. Acually, by looking a Figure 1 you can see ha he suden made a misake in he verb agreemen "regard6". Finally, he wo windows a he boom represen, from lef o righ, a snapsho of he sysem's rcorozy of meaning and he user's mexmry of fesa. The former is a device o display, hence o recall he underlying concepual represenaion of a specific senence chosen from he r~m~ry of fern window, whereas he laer is an incremenally buil daabase (race) of all he senences encounered so far. The idea behind his separaion is o allow he user o make a conrasive analysis of meaning and form beween 4 For more deails concerning he inerface conrolling he concepual inpu (Wha o say-componen), see Zoek (1991). ACRES DE COLING-92, NANTES, AO~' PREC. OV COL1NG-92, NANTES, AUO , 1992

3 wo seneuces. 5 Choosiug a seaeucc ill hc l-6~lorr y Of foran window gives a represenaion of he seuence's ueaniug in he mc~'ory of n~..anin9" winlow. By comparing he surface foru and he underlying meaning of wo senences, file user can appreciae he relaionship beween meauing aud rural. The criical feaure, dig one ha is respousible for he difference of form, is highlighed by he sysem. In our exanple i is he value "singular". Pas his poin, he user has various opions : eihel he coninues in he basic communicaion m<rde (How doe~ one nay<idea>?), or he changes he kind of quesion he wans o ask. Le us suppose ha he wans o build a compleely differen senlence. In his case he. could eiher go hrough he whole rouine, which i,~ quie cumbersome, or he could perfom he concepual changes direeely on he graph. Obviously, his laer melod is much faser. 6 Acually, everyhing you see on he graph, excep deepcase relaions, is considered by he sysem as a vaiiable whose value may be changed direcely by he user. h order o do so, he clicks on any of he showu aribues (verb, ense, mode, ec.) and chooses a new value. For example, if he clicks on he verb "regarder'" (o wach) he sysem answers by displaying a lis of cmdidaes from which we mus choose.. By clicking on he feaure "presen", rie sysem shows differen values ( presen, pas, fuure, ec.) of he variable "ense", ec. Everyiue he sysem is given some piece of informaion i will change he meaning represenaion accordingly and, il asked, oupu rie corresponding form. For example, if one sared from senence (a asking he sysem o chm~gc he number of he direc objec l om "singular" o "plural" i would produce (b). If one asked o change he ense from "simple pas" o "presen perfec", i would produce (e), ec. a) Le garcon regarda In lille (he boy wached ile J;jr!) b) Le garcon regarda!ees fill e~s (file boy wached he gir~. ) c) Le garcon a re rrrrrrrr~d.~ los filles (he Ir~,y has wached he girls) AS one can see, his mehod allows h)r local as well as for global changes. Acually, he dialogue described corresponds o he commuicaion mode 5. If one changes only one value each ime, asking he sysem o ell righ away how his meaning change is refleced in form, one can very quickly build ad explore a large search space. I should be noed, ha his kind of dialogue beween man and machine is much faser ad 5 See communicaion mode 8 : Wha is he difference beween <form x> and <form y>? 6 Acually, ha is one of he major advanages of absrac represenaions compared o, le us say, icons. Anoher advanage is ha his kind of represenaim shows on-line how concepual or pragmaic choices affec he inermediae srucure, hence more or less direcly ile final form. In consequence, by displaying on-line he way how he inermediae srucure changes afer each choice, we have a means of moving from a black box o a glass box. The former only shows how changes of he inpu '(concepual choices) are refleced in he oupu, whereas he laer also shows how discourse choices (opicalisaion, word choice, ec.) affec he inermediae levels. less edious han i would be beweeu a sudeu ani a eacher. The change of rieauiug ad rie sysem's generaion of file corresponding forrn ake abou a,second, cvea lor ml unffaiued u~r. Auoher conmunicalim mode he sysem allows for is callod: wua wo~q d ~,~ i~ <synacic modificakm>? This node allows rie user o perforu cerain rausformaions such as passive voice or prnnominalizaiou. In he laer case, he sysem will ask he user o specify which elemen he wans pronominalized (he boy, he girl, or boh). A&sume lu, saring from he concepual srucure underlying senence (a, he user waus o proaoniualize respecively he agen, he objec, and finally boh argunens. In hese case,s he would ge he following oupux : II regarde la fille. (he waches he gkl) I.e gargon la regarle. (he I W waches her) 11 h regarde. (lie waches her) By exploring and conrasing he difleren possibiliies, he user is nean o lean how variaions of neani,g are reflec~.l in hirer. 3 Discussion We have described a compuerized learning euviromnen ha has several iueresig feaures : (a i allows a poenial user o learn French on his own. (b) i fosers learning hrough explorakm (hypohesis-esing, inducive learning); (c) i allows he user o ask quesions raher I]lan simply answer hem. As he knows bes wha his needs are, he is uns likely o ask he righ quesions a he rigid momen; On he oher hand, he sysem has also various shorcomings: (a he linguisic coverage is exremely small : only very simple srucures c~n be used; (b) he concepual inerface is a surface- raher han a deepsrucure compouen (i already conains he lexieal iems, i.e., he progran hinks ill erms of he arge language); (c) he size of he discourse universe : one can duly alk abou a very limied range of facs; (d) he rigidiy of he concepual couponeu : he order in which lhe meaning is conveyed is defined a priori. The choice of verbs always precedes he choice of he uouns; (e) he erminology used (mealanguage) may be ilmccessible o he linguisically innocen user. In he remainder of his paper we will only be eonceneal wih he las poin. More precisely, we will focus on he problem ol conveying he noious of ime, ense ad aspec. For a proposal of how o increase he power and flexibiliy of he concepual compouea see [33, 34]. 4. Senence generaion and ense specificaion While much aeuiou has been devoed o rie problem of represening and inerpreing ense [2, 13, 13a, 13b], lile, if any aeuion has been devoed o he problem of generaing ense ill he conex of naural language generalim. A noable excepion is he work of V. Ehrich [6] who describes he properies of a siuaion (a convenien neural erm for making indiscriminaely reference o eveals, acious, processes, saes) in erms of caegory, aspoc, relaion, and posiion. Senence geaeraion consiss of ranslaing a concepual swucure ino a surface form. Obviously, informaion concerning ime, ense aud aspec is par of he ACrE.S DE COLING-92, NANTES, ao(rr Pico. ol: COI.ING-92, NANTES, AUG , 1992

4 concepual inpu. The problem is in wha erms a naive user could communicae hese absrac noions. 4.1 The problem of choosing he righ form Tense can be communicaed by choosing among a se of meulinguisic erms (pas, presen, fuure, ec.). These erms are acually used in schools, in ex books and even in inelligen uoring sysems [7, 8]. Ye his approach has several shorcomings : (A) Sudens are no linguiss. While noions such as "pas, presen, fuure" seem o be meaningful when we deal wih simple enses, hings change rapidly as soon as we move o more complex enses like presen coninuous, pas perfec, ec. The suden is cerainly shor of inuiion when he comes across enses for which here is no equivalen in his moher ongue. (B) There is no one-o-one correspondance beween grammaical enses and semanic values. The "presen forms" are quie ofen used in order o convey he meaning of he pas, he presen, or he fuure (1-4). In a similar vein, pas forms can be used wih a presen value (3) 1) J'arrive fi l'insan. (I have jus arrived.) pl~, 2) Je r~fl~chis. (I'm hinking.) pi'g~,, 3) J'arrive duns un insan. (I'm coming righ now.) 4) J'arrive. (l jus arrived.) PKST, - (I an coming.)exylx~: 5) Je venais vous demander une augmemnion (I jus waned o ask for a salary increase) Pr~.qEh~;, (C) The possibiliy of using a cerain form depends on he conex of he senence (presence of a emporal adverbial, naure of he subjec and complemens). 7 Consider he following examples : 6) Je suis conen (I am very glad.); 7) * J'dais conen (I was glad.); 8) * J'ai ~~ conen.(i have been glad.); 9) Hier, j'~ais conen. (Yeserday, I was very glad.); l O) * Hier, j'ai $$ conen. (Yeserday, I have been glad.); ll) Hier, j'ai ~~ conen d'apprendre on reour (Yeserday, I wa.v glad o learn ha you were coming back). While (6) is quie accepable, is ranslaions ino pas -- be i he simple pas (7), or pas perfec (8),-- need an explici reference. Moreover, he form of he pas perfec in French can only be used if one refers o an even, -- he ingression of a sae,-- as in (11). 5. How o visualize ense? Icons provide a very naural way of represening ense. Several grammars for English [21] and for French [10] make use of hem. We draw on hese sources and adap hem for SWIM. In order o do so we provide he user wih a basic se of elemens for ranslaing ense ino icons : (a) a ime axis; (b) a represenaion of he poin of speech (NOW) on he ime axis; (c) a se of ypes of 7 When using a specific ense one should also be sensiive o he kind of inferences he addressee migh draw. For insance, he following wo senences have compleely differen implicaions. : (a) I've been hinking abou you. (b) I hough abou you. While (a) suggess ha one cares for he person one is alking o, (b) implies ha one doesn' care any more. (We owe his example o Nicolay Vazov from he universiy of Sofia) siuaions (sae, even, ec.); (d) a frame of reference (a window); and/or (e) a poin of reference. The ime axis represens he inuiive noion of ime as an oriened linear srucure. The ime of verical arc locaes he ime of uerance: on he lef hand side is he pas, on he righ he fuure. The siuaion o be referred o can be communicaed by choosing an icon from a se of alernaives (see figure 7). The las wo componens, he frame and he poin of reference represen he way (poin of view) he speaker looks a he siuaion. We sar by describing hese basic elemens and illusrae heir use hrough examples. We hen explain how hese elemens are relaed o form a coheren meulinguisic vocabulary. 5.1 The basic elemens The minimal machinery necessary for generaing ense is composed of a ime axis wih a poin of speech (now) and a means of locaing an acion/even wih respec o his poin of speech (Figure 5). ime axis Figure 5 o o O poin of speech Reichenbach used he following conceps o characerize ense: a poin of speech (S), a poin of reference (R) and a poin of even (E). 8 Furhermore he used wo kinds of connecors in order o express he fac ha wo evens coincide wih one anoher C,"), or ha an even preceded anoher C_"). The following formulas shall illusrae his classificaion. I,',) (b) (c) (a) S,R,E (b) R,E_S (c) E S,R (d) E_R_S (e) S,RE (0 S_E_R R.E S,R.E E R S I Idl I I I S,R,,.,, E S,R S E R I I O) ' ' ' Figure 6 presen ense simple pas presen perfec pas perfec simple fuure fuure perfec I see Paul I saw Paul I have seen Paul I had seen Paul I will see Paul I will have seen Paul Since hen he sysem has been exended or re-inerpreed by a number of auhors [3a, 12a, 12b, 24, 24a, 25a,28a]. (A) Types of siuaion and corresponding icons: A senence refers o some poin in ime (emporal siuaion). A siuaion will be represened by any of he following icons 8 Voin of speech refers o he very momen in which he uerance akes place (now). The poin of reference is he perspecive from which he even is considered. The r~in of even is he very momen a which he even ook place (yeserday). AcrEs DE COLING-92, NANTnS, AO~r PROC. OV COLING-92, NArcrl~s, AUG , 1992

5 Pul likes lmimn coling. Pul u.em o puy mn~. Pul wong he nce. Pul doav o l,cmdml. I hu been r~ininll. Psul i piying ~nm~ Paul is dnvbg o l~dm, Figure 7 : Types of siuaion and corresponding icons A sae is considered as covering a srech of ime wih no definie boundaries. The icon reperesening his suggess his open-endedness by having fuzzy delimiing zones. Boh ongoing processes and aciviies have a well-defined saring poin bu no deermined poin of erminaion. Noe however, ha aciviies can be inerruped a any ime and be repored as evens. Le us suppose ha "Paul was swimming", and hen sopped. This can be repored by ~Paul swam,. Telie processes do no have his propery, because hey have some naural poin of compleion. If <<Paul was swimming across he Channel~, and hen sopped swimming, we could describe his siuaion as Paul has been swimming, bu we could no say ~Paul swam across he Channel.~> In ha case, he poin of compleion can be referred o, even if i is never reached: ~Paul was swimming across he channel, bu he never made i>>. The siuaion can also be considered as indeeffninae:,paul was swimming across he Channel when I las heard abou him>~. An even has boh a saring poin and a poin of erminaion. I can be puncual or no. According o Reichenbach i is represened as a poin or au inerval, whereas we allow for he addiional use of a frame. An even can also be considered wih regards o he resul of is occurrence. There are evens wih a resuling sae. In he corresponding icon, we represen his sae as adjacen o he even which has brough i abou. Usually, he precise naure of his resuling sae needs o be inferred (cf. [10]): I has been raining; in consequence he pavemen is we, i go comer, here are puddles in he sree, my hairdress is spoiled, ec. Iabiual evens are an undeermined umber of evens. The icon of habiual evens is mean o sugges heir resemblance o saes. I should "also be noed ha he number of evens may be concepnally indefinie. Compare: (a)l played ennis many imes las year.(b) 1 used o play ennis when I was a young man In gus case we are dealing wih a sequence of individual evens, no wih a habiual even. Hence i is quie possible o coninue he dialogue by asking he following quesion "how many imes did you play?", while his quesion.seems quie odd in he second case. (B) Frame of reference, poin of reference: Opional elemens are a frame of reference and a poin of reference, boh of which can be manipulaed by he user. The former consiss of a window which represens he period of ime considered by he speaker. I is usually associaed wih a emporal adverbial (Yeserday Paul visied he Eiffel Tower). The poin of reference can be used in order o represen some insan which plays he role of a ranslaed poin of speech (h 1980 Paul had already visied Paris wice.). 5.2 Using hese elemens n produce a ime scenario The procedure is as follows: (a) Choose a suiable ype of siuaion (icon) o represen he emporal srucure of he siuaion; (b) Posiion he icon relaive o he poin of speech; (c) If necessary, inroduce and posiion a frame and/or poin of relerence and posiion i.(d)in he overall represenaion, some par of he siuaion should be adjacen o, or included in an elemen of reference (poin of speech, poin of reference, frame of reference). If no, reexamine sep (c). Suppose we have an unensed represenaion of : ~Paul aime la cuisine ialienne, (Paul likes Ialian cooking). In order o generae he ense, we choose a sae and he posiion of he poin of speech on he ime axis (Figure 8a). From his represenaion he corresponding ime (presen ense) can be generaed. _ (a) ~ml imlml Pmul mhne ik cuisine i~licnnc (Paul likc~ lalim cooking),... / L'mm~ pas~, Paul aimai h cuisine ialiem~ (c) ha. j'fic~ivmi~ mor uucle (I was wriing my paper ye*erday) ]... ill...! (d) llier, j'6civis mma irick (1 w~ wriing my ~ ye6~'djy) Figure 8 Figure 8b conveys he same mes~ge as 8a in he pas. Since no direc reference o he poin of speech is available, a frame of reference is necessary. Here i plays he same role as he poin of speech in figure 8a. If he user adds he adverbial ~<l'an pass6,~ (las year) as a marker of reference, we ge ~<l'an pass6, Paul aimai la cuisine Z n ACTES DE COLlNG-92, NANTES ho~r PROC. OF COLING-92, NANTES, AUG , 1992

6 ialienne~, (Las year Paul liked ialian cooking.) The inerpreaion of maimer la cuisine ialienne,(o like ialian cooking) in erms of a sae should be disinguished from is inerpreaion as an even. The laer would be he insananeous change from one sae o anoher. Tha's why we need furher specificaions: ~Quand Paul a renconrd Gina, il a aimg la cuisine ialienne.~ (When Paul me Gina, he began o like Ialian cooking). Figures 8c and 8d show wo possible represenaions of similar, ye differen siuaions.the compleion of he work is no precisely deermined in his case. Among he difficulies faced by a learner of French is he dual value of "pass6 compos6". Consider, (lier), j'ai dcri mon aricle - [ (yeserday) 1 wroe my paper] In one inerpreaion (figure 9a), he wriing of he paper is considered as a pas even. Locaing he acion wihin a reference frame requires he use of an adverbial like ~hier>> (yeserday). 9)11i ~ jqu &:i non riele (I wroe my paper yeserday)... I ii :~ ili T" I ~ii ili i [ Figure 9 This case is o be differeniaed from he siuaion depiced in Figure 9b. Here he implici reference is now. The pas even is considered wih respec o he sae resuling from i (presen relevance). In fac, he same ense is used in his ease in French, whereas <<Yeserdayl wroe my paper~ would be used in English in case (9a), and e.l have wrien my paper~ in case (9b). Anoher ypical case of ambiguiy is he use of he "presen ense" in French. The Figures 9c and 9d, while differen in English, yield he same form in French. The firs one (9e) signals habiual or repeiive use of he presen ense, corresponding o <~I play ennis~. The second siuaion corresponds o "1 am playing ennis", implying an ongoing aciviy a he ime of speech. Please noe ha he firs siuaion does no imply any acual aciviy a he ime of speech, bu only a general validiy of a kind of propery. As he icon suggess, i looks very much like a sae. 6. The underlying emporal onology and is use for generaion 6.1. Types of siuaion and corresponding icons Our classificaion of ypes of siuaion (cf. also [2]) is based on conceiving he emporal srucure of a siuaion as a resul of (a) he lexical meaning, he ense and he aspec of a given verb, (b) he naure of he subjec, he complemens and he ime adverb'rals. Lexical meaning : We basically subscribe o Vendler's ypology [26] who classifies siuaions in erms of saes, aciviies, accomplishmens and achievemens. Nakhimovsky, in a more recen classificaion [17], inroduces addiional feaures (generic vs. specific resources for aclic processes) o Vendler's puncualiy, saiviy, and elieiy. Our classificaion is based on four classes of lexical meanings, corresponding o combinaions of he feaures: puncualiy, saiviy, eliciy. puncual evens (puncual, non saive, aelic); saes (non puncual, saive, aelic); elic processes. aelic aciviies; Aspec is relaed o he perspecive from which a given siuaion is viewed. Following Comrie [4], we disinguish beween perfec, perfecive and imperfeeive aspec. In French, he imparfai (coninuous paso is a ypical imperfecive form: a siuaion is considered from he inside. Perfecive, on he oher hand, considers a siuaion as a whole. Perfec is relaed o he consideraion of a given even ogeher wih a new sae which is a consequence of his even happening, for example, I have prepared some coffee. In our sysem, perfec is represened by an even wih a resuling sae. Aspec plays a dual role: i deermines he ype of siuaion and i allows for choosing a specific frame of reference. Lexical meaning combined wih he aspec of he verb, he ypes of he subjecs and complemens, and he emporal adverbials yield he ype of siuaion. For example, he verb dcrire (o wrie) has an inrinsic lexical meaning : aelic aciviy. However, i can change : Paul dcrivi un roman (John wroe a novel) : elic process; Paul dcri des romans (John wries novels) : aelic aciviy; Paul dcrivai an roman (John was wriing a book) : ongoing process How o choose he righ ense.9 The following able shows how ense could be compued on he basis of an iconic represenaion of a siuaion. The sysem deermines he posiion of he main par of he siuaion relaive o he poin of speech (PS). The qualificaion is necessary only in he case of an even wih a resuling sae, where he even is he main par of he siuaion. If he main par of he siuaion is poserior o he poin of speech, a fuure ense is chosen by defaul, hough presen ense remains a possible candidae. Using eiher or amouns o communicaing a special modal nuance. ACT,S DE COLING-92, NArcr~s, AOUT PROC OF COLING-92 NANTES, AUG , 1992

7 F fdlo. PS umla~n ms'm~l 9 amy -1 r{... } ~mr {~n)} / / L ',.~ ~ ~ ~ ~ [.. ",*.,,~ um~ p,mcu~a *~m e~n w($ I~mlfinll m. ~ ~ li f pu~ ~l~o~,~ pr ~m p*~ v= --i (;:,~,im#*) l~.~,.iz Table 1 If he main siuaion is prior o he ime of speech, he conclusion depends on he presence of a poin of reference (PR) prior o he ime of speech. If no such poin is ~],:i "... "" - [::iiiil::i '"... " " l!!iil.,,..... u,,.d,...., r ~1 ~.~.'~'=d ~,,-~a~o,. li~i ~%!::i~i::::~i~::~i~::~!~:: lh~.,~... d ~. ~. [::~:.~ Figure 10 specified, and if he ype of siuaion is an even, one can use he pas perfec. In he case of an even or a puncual even, he pas perfec is a sylisic alernaive (polished syle). Temporal adverbials are classified as (a) ime-posiion adverbs (las nigh, on Monday); (b) as span and duraion adverbs (for a monh, up o 1979); or (c) as frequency adverbs (usually, ofen). Temporal adverbs consrain ense generaion. Duraional adverbs may enforce a habiual meaning on verbs having an inrinsic puncual meaning : <<He has been diving for several monhs now.~, Temporal subordinae clauses: In he case of simple senences, he ask of deermining he ense from an icon is sraighforward. More suble sraegies are needed in senences wih emporal subordinae clauses 9. Suppose he main clause concerns <<sudens working>~ (PAST), whereas he subordinae clause is <<he headmaser enering he classroom ~ (PAST). Le he emporal conjuncion be quand (when). Then we have wo possibiliies: 9 We leave aside he problem of generaing subordinae clauses in he subjuncive mood; many emporal conjuncions (e.g. avan que : before) require he use of he subjuncive mood. (1) Quand le direceur es enrd dars la classe, les dudians ravaill. aien (When he headmaser enered he classroora, he sudens were working ); (2) Quand le direceur es enr~ dam la classe, les ~udians on ravaiim (When he headmaser enered he classroom, he sudens began o work ). See figure 10 The firs case represens an ongoing process or aciviy, wih an even occuring somewhere in he middle of i, while he second is a sequence of wo adjacen evens. So far we handle he case of quand (when), using icons and sraegies similar o he ones described by Moons & Seedman [16]. The way in which emporal subordinae clauses reflec on he ypes and order of he siuaions has been examined in he lieraure ([24, 15, 16]). 7 Conclusion We have presened in his paper a mehod of visualizing ime, ense and aspec. By using icons we have provided he user wih an inerface ha allows him o ranslae absrac hough ino concree hough. The advanage being ha his houghs are expressed in erms of concree acions raher han in erms of a (mere)language, which is ofen oo absrac o be meaningful o he nadve user. 8 References [1] Allen J. F., Mainaining Knowledge abou Temporal Inervals, Communicaions of he ACM 26 (11), ] Besougeff H., Ligoza G., Ouils logiques pour le raiemen du emps, Masson. l)aris, [3] CL88, Special lssue on Tense and Aspec. Compuaional Linguisics 14 (2), [3a] Borillo, M., Vieu, L.. Eldmens pour la formalisaion du raisonnemen spaio-emporel naurcl, Congr/~s RFIA, AFCETI', (1989) [4] Comrie B., Aspec, Cambridge Universiy Press, [5] Dowry D. R., The Effecs of Aspeeual Class on he Temporal Srucure of Discourse: Semanics or Pragmaics? ; Linguisics and Philosophy 9 N 1 (1986) 3%61. [6] Ehrich V., The Generaion of Tense, in Naural Language generaion, Gerard Kempcn (ed.), , 1987, Marinus Nijhoff Publishers, Dordrech. [7] Fum D., P. Giagrandi & C. Tasso, ET: an Inelligen Tuor for Foreign Language Teaching. In : Proceedings of ITS-88, Monreal, Canada, 1988 AClEs DE COLING-92. NAMES, AOfr ' PROC. OF COLING-92, NANTES, AUO , 1992

8 [8] Fum, D.; P. Giagrandi & C. Tasso, Tense generaion in an Inelligen Tuor for Foreign Language Teaching: some Issues in he Design of he Verb Exper. In Proceedings of he 4h Conference of he European Chaper of he Associaion for Compuaional Linguisics, Mancheser, UK, 1989 [9] Grasso E., Losmo L., Lombardo V., Maccario P.M., Saiao R., Terenziani P., Semanic Inerpreaion of Tense, Acionaliy and Aspec. Proceedings of he 9h European Conference on Arificial Inelligence (ECAI-90). Sockholm, Augus 1990, [ 10] Grvisse M., Prcis de Grammaire Franeaise. Duculo. I979 [11] Hinrichs E., Temporal anaphura in discourses of English; Linguisics and Philosophy 9 N 1 (86) [12] Hinrichs E., A Composiional Semanics of Temporal Expressions in English. in:proceedings of he25h Annual Meeing of he Associaion for Compuaional Linguisics. 6-9 July 1987, Sanford, [12a] Hornsein, N., Towards a Theory of Tense, Linguisic Inquiry 8, , (1977) [12b] Hornsein, N., As Time Goes By. Tense and Universal grammar, MIT Press (1990) [13] Ligoza, G., Weak Represenaions of Inerval Algebras, Proceedings of he 8h Conference on Arificial Inelligence, (AAAI-90). 1990, Boson, [13a] Ligoza, G., On generalized inerval calculi, Proceedings of he 9h Conference on Arificial Inelligence, (AAAI-91), 1991, Anaheim, CA, [13b] Ligoza, G. & Besougeff, H., Reasoning abou emporal knowledge, in Klein, E., Pouradier, F & Wagner, K.H. (eds.) Beriebslinguisik und Linguisikberieb, Niemeyer, Tllbingeo, 1991, pp [14] McDermo D., A Temporal Logic for Reasoning abou Processes and Plans, Cogniive Science 6, , [15] Moens M., Seodman M., Temporal Onology in Naural Language. in: Proceedings of he25h Annual Meeing of he Associaion for Compuaional Linguisics, 6-9 JMy 1987, Sanford., 1-7. [16] Moens M., Seedman M.. Temporal Onology and Temporal Reference, in [CL88], [17] Nakhimovsky A., Aspec, Aspecual Class, and he Temporal Srucure of Narraive, in [CL88], [18] Nerbonne J.. Reference Time and Time in Narraion; Linguisics and PhUosophy 9 N 1 (1986) [19] Passonneau R. J., Siuaions and inervals, in:proceedings of he25h Annual Meeing of he Associaion for Compuaional Linguisics, 6-9 July 1987, Sanford, [20] Passonneau R.J., A Compuaional Model of he Semanics of Tense and Aspec,/n [CL88], [21] Quirk R., Greanbaum S., Leech J.. Svarvik J., A Comprehensive Grammar of he English Language, Longman, London and New York, [22] Reichenbech H.. Elemens of Symbolic Logic, The Free Press, New York, [23] Rohrer C. (ed), Time. Tense. and Quanifiers, Proceedings of he Sugar Conference on he logic of ense and quanificaion, Niemeyer, Tibingen, [24] Smih C., A Speaker-based Approach o Aspec; Linguisics and Philosophy 9 N 1 (1986) [24a] Smih, C., The synax and inerpreaion of emporal expressions in English, Linguisics & Philosophy 2, 1, (1978). [24b] Smih, C., A Theory of Aspecual Choice, Language 53, 3, 1983). [25] Sows L, Concepual srucures : Informaion Processing in Mind and Machine, Addison-Wesley, Reading, Massachuses, 1984 [25a] van Eynde, F., The Semanics of Tense and Aspec, in: Proceedings of EAIA-90, 2rid Advanced School in Arificial Inelligence. Naural Language Processing in Arificial Inelligence, Filgueiras M., Dumas L., Moreira N.. Tomas A.P. (eds.) (1990). [26] Vendler Z., Verbs and Time, in: Vendler Z., Linguisics in philosophy, , Cornell Universiy Press. Ihaca. New York, 1967 (Revised version of Veodler Z., Verbs and imes, The Philosophical Review 66, 1957, ). [27] Webber B.L., The Inerpreaion of Tense in Discourse, in:proceedings of he25h Annual Meeing of he Associaion for Compuaional Linguisics, 6-9 July 1987, Sanford [28] Webber B.L., Tense as Discourse Anaphor, in [CL881, [28a] Yip, K. M. K., Tense. Aspec and he Cogniive represenaion of Time. in: Proceedings of ljcal-85, (1985). [29] Zock, M., If you can' open he Black Box, Open a Window [ or, how o Visualize Dependency Relaionships when Mapping Meaning ono Form 7, In: COGNITIVA-90, Madrid [30] Zock, M., How o Transform a Black Box ino a Glass Box? or, Why use Inermediae Srucures in Compuer Assised Language Learning, In: Proceedings of he 7h Annual Apple European Universiy Consorium, Paris [31] Zock, M. & Laroui, A., Visualising resuls of choices in language generaion : he use of inermediae srucures. In: Proceedings of he Workshop on Naural Language Learning. IJCAI-91, Sidney [32] Zock, M., SWIM or Sink : he Problem of Communicaing Though. In M. Swarz & M. Yazdani (Eds.) The Bridge o Inernaional Communicaion : Inelligen Tuoring Sysems for Foreign Language Learning, Springer Verlag, Berlin, New York, 1992 [33] Zock, M., Naural Languages are Flexible Tools, ha's wha Makes hem Hard o Learn, o Explain and o Use. In, M. Zock & G. Sabah (Eds.), Advances in Naural Language Generaion. coediion Piner, London - Ablex, N.J. [34] Zock, M. & R. Mikov. How o ask a foreigner quesions wihou knowing his language : proposiions for a concepual inerface o communicae hough. In, Proceedings of he Naural Language Processing Pacific RIM Symposium, Singapore Ac'rEs DE COLING-92, NANTES, AOI~T PROC. OF COLING-92, NANTES, AUG , 1992