HOW TO VISUALIZE TIME, TENSE AND ASPECT? 
COMMENT VISUALISER LE TEMPS ET L'ASPECT ? 
GErard LIGOZAT - Michael ZOCK 
LIMSI, Langage & Cognition 
B.P. 133, 91403 Orsay Cexiex / France 
courtier 61eetronique : ligozat @ limsi.fr, ou, zock @ limsi.fr 
Nous allons dEcrire ci-dessous une extension 
d'un syst~me interactif de gEnEration de phrases. 
Apr~s une description de SWIM,--acronyme de 
See What I Mean ?-- nous allons discuter la 
fa~on dont on communique au syst~me des 
notions relatives au temps et ~t l'aspect. 
Avec SWIM nous poursuivons trois objectifs : 
(a) assister un Etudiant dans rapprentissage 
d'une langue, plus prdcisEment aider l'apprenant 
h passer des idEes aux phrases (production du 
langage); (b) visualiser la relation sens-forme.; 
(c) fournir un environnement voisin des condi- 
tions dans lesquelles on apprend naturellement 
une langue. 
Savoir parler une langue n'est pas innt, c'est le 
rEsultat d'un apprentissage. Selon le cadre on 
parle d'apprentissage naturel (on apprend dans 
le pays mtme 05 est utilisEe eette langue) ou 
d' apprentissage institutionnel (enseignement). 
Une difference fondamentale entre les deux 
types d'apprentissage se situe au niveau du 
contr61e. Dans un cas rapprenant est act ± f 
(apprentissage naturel), - il utilise la parole pour 
rtsoudre un problSme le concernant (il cherche h 
obtenir des informations, il veut verifier quelque 
chose, etc.) - tandis que darts rautre cas il est 
essentiellement react i f, c'est-h-dire qu'il ne 
parle que lorsqu'011 le lui demande. 
L'inconvEnient majeur de 1 'apprentissage 
naturel est lid au fait que l'apprenant se trouve 
dans des situations off tout change constam- 
ment : la situation discursive, les buts ou les in- 
tentions de communication, les contenus et les 
formes linguistiques correspondantes. Cela est 
normal, car l'objeetif est la communication -- 
faire passer un message en vue d'un but 
donne ~ et non la fixation, rexplication ou 
rillustration systEmatique d'lm point precis de la 
langue. 
Quanth l'apprentissage dans un contexte 
scolaire (enseignement), les inconvEnients en 
sont les suivants: (a) la situation est artificielle 
(on s'exerce pour apprendre la langue et non 
pour communiquer un message en rue d'un but 
donn6); (b) le professeur ignore gtndralement la 
nature des besoins de l'E16ve. Si l'E16ve sait ce 
qu' 11 veut dire, en revanche, il ignore sou- 
vent coranent le dire. Quant au professeur, il 
sait conmwnt le dire, mais il ignore ce que veut 
dire rEtudiant. 
Pour rtsoudre ce probl~me de communication 
entre le professeur et rEtudiant nous avons 
con~u un systtme oh l'ElSve peut poser des 
questions et off le professeur rEpond. On notera, 
que c'est rEtudiant qui a rinitiative du dialogue. 
C'est lui qui pose des questions, et il les pose en 
fonction de ses besoins. 
A l'beure actuelle le systtme a plusieurs fai- 
blesses et un des grands probltmes est celui de 
l'interface conceptuelle : comment lui comnmni- 
quer le contenu du message, puisqu'on ne parle 
pas encore cette lmlgue ? 
Nous montrons dans la deuxitme partie de ce 
papier la fa~on dont l'dlSve communiqne des in- 
formations relatives au temps et h l'aspect. Ces 
infonnations sont d61icates h transmettre pour au 
moins trois raisons : (a) il n'y a pas 
d'isomorphie entre les temps conceptuels et les 
temps grammaticaux (le "prEsent" peut aussi 
bien exprimer des 6vEnements qui ont eu lieu 
que des 6vtnements qui vont avoir lieu); (b) il 
n'y a pas de correspondance stricte entre les 
temps des diffErentes langues (c) la tenninologie 
utilisEe dans les manuels est trop absu'aite. Elle 
pose des probltmes d'une maniSre gEnErale, et h 
plus forte raison lorsque la catEgorie 
mdtalinguistique utilisEe n'a pas de 
correspondant dans la langue cible (le passd 
simple, par exemple, n'existe pas dans toutes les 
langues). 
Partant des travaux de Reichenbach et de 
Vendler nous avons conqu un langage ic6nique 
pour communiquer les informations relatives au 
temps. Pour ce faire, nous avons complEtE leurs 
iddes tout en les rendant plus accessibles ~t 
l'utilisateur (mEta) linguisfiquement innocent. 
Mots clefs : entree conceptueile, gEnEration de 
phrases, relations fonne-sens, EIAO 
ACRES DE COLING-92. NANTES, 23-28 AO{n' 1992 4 7 $ PRo(:. OF COLING-92. NAI'CrEs. AUG. 23-28, 1992 
1 Introduction 
The major goal of the system described here is to provide 
a natural envirc~'~nent I for learning to produce 
sentences in French. In order to achieve this goal, we 
have integrated the student into the process. He should be 
active rather than reactive, that is, he should talk not 
only when being asked. After all, he knows what he 
wants to say, what he lacks are the means of translating 
his thoughts into language. 
SWIM 2 is an exploratory environment which has been 
built to simulate natural learning. The student asks the 
questions and the system answers them. Any of the 
following questions could occur in a natural setting : 3 
1) How does one say ±c~a .9 
2) Can one say linguistic fo~m ? 
3) How should one say idea .9 
4) Why does one say fen'n-1 and not form-2 ? 
5) What would happen if conceptual rectification ? 
6) What would happen if syntactic modification ? 
7) What would happen if word y instead of word x ? 
8) What is the difference between form xand fozaxa y ? 
leaning-Representafio: 
t t p~l lia~Uhtl decitmtlv¢ 
t t 
t 
Ix gargon .'a pas mgardte~ ies filles. 
t., g=¢o. n'a pa* ~ les filles. 
t 
1) Lt&~¢~r¢S~dclaftOe. 
\[21 IX' ilalPm t~"t PU i~8~d4 It fi\[~ \] 
J ) IL ~6~ "'a pas regard# I~ fi/lo. 
Figure 1 
1 A computerized learning environment is said to be 
natural if the transfer of knowledge is achieved essentially 
in rite same way as could be the case between two human 
agents, where a learner (child. student) asks questions and 
the teacher (native speaker, parent/teacher) answers them. 
2 SWIM is the acronym of See What I Mean ? It has been 
conceived by Michael Zock and implemented by A. Laroui 
on a Macintosh. 
3 For the time being only the first and last four types of 
questions are implemented for a small subset of French 
(simple sentences). 
The first communication-mode (how does one say 
<idea>?) raises an interesting problem : how to tell the 
computer what one wants to say (message). The 
fundamental question here is in what terms to code 
thought and access it. Obviously, there are several ways. 
One could ask in another natural language (translation), 
one could use images, or one could use some form of 
metalanguage. 
In this paper we discuss some of the problems that arise 
by using a metalauguage. We then show how a special 
category of thought, tense, can be communicated by 
nsing a hybrid form of representation. But, before doing 
so let us give a short description of the system. For 
more details see \[ 29, 30, 31 \]. 
2 Description of the system 
The dialogue is initiated by specifying the 
communication mode. Let us suppose that the user had 
started with the first question: How does one say <idea> 
? In that case the system presents a menu (type hierarchy) 
from which the user has to choose: 
Speech act : statement, question, command 
Actions - processes: taxonomy of verbs 
tense : present, past, future, etc. 
Discourse objects : taxonomy of nouns 
number : singular, plural 
communic, status : definite, indefinite, 
Attributes: taxonomy of attributes 
By choosing specific values from a set of al- 
ternatives, the user tells the system what he 
wants to say. 4 As the dialogue develops, the 
system builds the underlying meaning in the 
form of a semantic network. It then invites 
the user to try to express this meaning, after 
which it outputs its own form. 
As one cat* see from Figure 1, the screen is 
divided into five parts. The large window at 
the top represents the underlying meaning of 
the message a student is trying to convey. 
The next two windows contain respectively 
the user's attempt to express this meaning 
and the system's version. Possible 
mismatches between these two versions are 
highlighted on the screen. This allows the 
system to draw the student's attention to 
errors. Actually, by looking at Figure 1 you 
can see that the student made a mistake in 
the verb agreement "regard6". Finally, the 
two windows at the bottom represent, from 
left to right, a snapshot of the system's 
rcorozy of meaning and the user's mexmry of festa . 
The former is a device to display, hence to recall the 
underlying conceptual representation of a specific 
sentence chosen from the rt~m~ry of fern window, 
whereas the latter is an incrementally built database 
(trace) of all the sentences encountered so far. 
The idea behind this separation is to allow the user to 
make a contrastive analysis of meaning and form between 
4 For more details concerning the interface controlling the 
conceptual input (What to say-component), see Zoek (1991). 
ACRES DE COLING-92, NANTES, 23-28 AO~t' 1992 4 7 6 PREC. OV COL1NG-92, NANTES, AUO. 23-28, 1992 
two senteuces. 5 Choosiug a seateucc ill thC lt-t6~tlOrr y Of 
foran window gives a representation of the seutence's 
tueaniug in the mc~'ory of n~..anin9" wintlow. By 
comparing the surface forut and the underlying meaning 
of two sentences, file user can appreciate the relationship 
between meauing aud rural. The critical feature, dig one 
that is respousible for the difference of form, is 
highlighted by the system. In our exantple it is the value 
"singular". 
Past this point, the user has various options : eithel he 
continues in the basic communication m<rde (How doe~ 
one nay<idea> ?), or he changes the kind of question he 
wants to ask. Let us suppose that he wants to build a 
completely different senlence. In this case he. could either 
go through the whole routine, which i,~ quite 
cumbersome, or he could perfomt the conceptual changes 
direetely on the graph. Obviously, this latter metltod is 
much faster. 6 
Actually, everything you see on the graph, except deep- 
case relations, is considered by the system as a vaiiable 
whose value may be changed directely by the user. ht 
order to do so, he clicks on any of the showu attributes 
(verb, tense, mode, etc.) and chooses a new value. For 
example, if he clicks on the verb "regarder'" (to watch) the 
system answers by displaying a list of cmtdidates from 
which we must choose.. By clicking on the feature 
"present", rite system shows different values ( present, 
past, future, etc.) of the variable "tense", etc. Everytiute 
the system is given some piece of information it will 
change the meaning representation accordingly and, il 
asked, output rite corresponding form. For example, if 
one started from sentence (at asking the system to chm~gc 
the number of the direct object l¥om "singular" to 
"plural" it would produce (b). If one asked to change the 
tense from "simple past" to "present perfect", it would 
produce (e), etc. 
a) Le garcon regarda In lille (the boy watched tile J;jr!) 
b) Le garcon regarda!ees fill e~s (file boy watched the gir~. ) 
c) Le garcon a re rrrrrrrr~d.~ los filles (the Ir~,y has watched the 
girls) 
AS one can see, this method allows h)r local as well as 
for global changes. Actually, the dialogue described 
corresponds to the commuttication mode 5. If one 
changes only one value each time, asking the system to 
tell right away how this meaning change is reflected in 
form, one can very quickly build attd explore a large 
search space. It should be noted, that this kind of 
dialogue between man and machine is much faster attd 
5 See communication mode 8 : What is the difference 
between <form x> and <form y> ? 
6 Actually, that is one of the major advantages of abstract 
representations compared to, let us say, icons. Another 
advantage is that this kind of representatimt shows on-line 
how conceptual or pragmatic choices affect the intermediate 
structure, hence more or less directly tile final form. In 
consequence, by displaying on-line the way how the 
intermediate structure changes after each choice, we have a 
means of moving from a black box to a glass box. The 
former only shows how changes of the input '(conceptual 
choices) are reflected in the output, whereas the latter also 
shows how discourse choices (topicalisation, word choice, 
etc.) affect the intermediate levels. 
less tedious than it would be betweeu a studeut anti a 
teacher. The change of trieauiug attd rite system's 
generation of file corresponding forrn take about a ,second, 
cvea lor ml unffaiued u~r. 
Auother contmunicalitm mode the system allows for is 
callod: wtuat wot~q d t~,~ i~ <syntactic 
modificatkm>? This ntode allows rite user to perforut 
certain trausformations such as passive voice or 
prnnominalizatiou. In the latter case, the system will ask 
the user to specify which element he wants 
pronominalized (the boy, the girl, or both). A&sume tlutt, 
starting from the conceptual structure underlying sentence 
(at, the user wauts to proaontiualize respectively the 
agent, the object, and finally both arguntents. In these 
case,s he would get the following outputx : 
II regarde la fille. (he watches the gkl) 
I.e gargon la regartle. (the I× W watches her) 
11 ht regarde. (lie watches her) 
By exploring and contrasting the diflerent possibilities, 
the user is nteant to leant how variations of nteani,tg are 
reflectt~.l in hirer. 
3 Discussion 
We have described a computerized learning euviromnent 
that has several iuterestittg features : (at it allows a 
potential user to learn French on his own. (b) it fosters 
learning through exploratkm (hypothesis-testing, 
inductive learning); (c) it allows the user to ask 
questions rather I\]lan simply answer them. As he knows 
best what his needs are, he is untst likely to ask the 
right questions at the rigid moment; 
On the other hand, the system has also various 
shortcomings: (at the linguistic coverage is extremely 
small : only very simple structures c~tn be used; (b) the 
conceptual interface is a surface- rather than a deep- 
structure compouent (it already contains the lexieal 
items, i.e., the progrant thinks ill terms of the target 
language); (c) the size of the discourse universe : one can 
duly talk about a very limited range of facts; (d) the 
rigidity of the conceptual coutponeut : the order in which 
lhe meaning is conveyed is defined a priori. The choice of 
verbs always precedes the choice of the uouns; (e) the 
terminology used (metalanguage) may be ilmccessible to 
the linguistically innocent user. 
In the remainder of this paper we will only be eoncenteal 
with the last point. More precisely, we will focus on the 
problem ol conveying the notious of time, tense attd 
aspect. For a proposal of how to increase the power and 
flexibility of the conceptual compoueat see \[33, 34\]. 
4. Sentence generation and tense specification 
While much atteutiou has been devoted to rite problem of 
representing and interpreting tense \[2, 13, 13a, 13b\], 
little, if any atteution has been devoted to the problem of 
generating tense ill the context of natural language 
generalimt. A notable exception is the work of V. Ehrich 
\[6\] who describes the properties of a situation (a 
convenient neutral term for making indiscriminately 
reference to eveals, actious, processes, states) in terms of 
category, aspoct, relation, and position. 
Sentence geaeration consists of translating a conceptual 
swucture into a surface form. Obviously, information 
concerning time, tense aud aspect is part of the 
ACrE.S DE COLING-92, NANTES, 23-28 ao(rr 1992 4 7 7 Picot. ol: COI.ING-92, NANTES, AUG. 23-28, 1992 
conceptual input. The problem is in what terms a naive 
user could communicate these abstract notions. 
4.1 The problem of choosing the right form 
Tense can be communicated by choosing among a set of 
metulinguistic terms (past, present, future, etc.). These 
terms are actually used in schools, in text books and 
even in intelligent tutoring systems \[7, 8\]. Yet this 
approach has several shortcomings : 
(A) Students are not linguists. While notions such as 
"past, present, future" seem to be meaningful when we 
deal with simple tenses, things change rapidly as soon as 
we move to more complex tenses like present 
continuous, past perfect, etc. The student is certainly 
short of intuition when he comes across tenses for which 
there is no equivalent in his mother tongue. 
(B) There is no one-to-one correspondance between 
grammatical tenses and semantic values. The "present 
forms" are quite often used in order to convey the 
meaning of the past, the present, or the future (1-4). In a 
similar vein, past forms can be used with a present value 
(3) 
1) J'arrive fi l'instant. (I have just arrived.) pl~, 
2) Je r~fl~chis. (I'm thinking.) pI'g~,, 
3) J'arrive duns un instant. (I'm coming right now.) 
4) J'arrive. (l just arrived.) PKST, - (I ant coming.)eXYlx~: 
5) Je venais vous demander une augmemntion 
(I just wanted to ask for a salary increase) Pr~.qEh~;, 
(C) The possibility of using a certain form depends on 
the context of the sentence (presence of a temporal 
adverbial, nature of the subject and complements). 7 
Consider the following examples : 
6) Je suis content (I am very glad.); 
7) * J'dtais content (I was glad.); 
8) * J'ai ~t~ content.(I have been glad.); 
9) Hier, j'~tais content. (Yesterday, I was very glad.); 
l O) * Hier, j'ai $t$ content. (Yesterday, I have been glad.); 
ll) Hier, j'ai ~t~ content d'apprendre ton retour (Yester- 
day, I wa.v glad to learn that you were coming back). 
While (6) is quite acceptable, its translations into past -- 
be it the simple past (7), or past perfect (8),-- need an 
explicit reference. Moreover, the form of the past perfect 
in French can only be used if one refers to an event, -- 
the ingression of a state,-- as in (11). 
5. How to visualize tense ? 
Icons provide a very natural way of representing tense. 
Several grammars for English \[21\] and for French \[10\] 
make use of them. We draw on these sources and adapt 
them for SWIM. In order to do so we provide the user 
with a basic set of elements for translating tense into 
icons : (a) a time axis; (b) a representation of the point 
of speech (NOW) on the time axis; (c) a set of types of 
7 When using a specific tense one should also be sensitive 
to the kind of inferences the addressee might draw. For 
instance, the following two sentences have completely 
different implications. : (a) I've been thinking about you. (b) 
I thought about you. While (a) suggests that one cares for 
the person one is talking to, (b) implies that one doesn't care 
any more. (We owe this example to Nicolay Vazov from the 
university of Sofia) 
situations (state, event, etc.); (d) a frame of reference (a 
window); and/or (e) a point of reference. 
The time axis represents the intuitive notion of time as 
an oriented linear structure. The time of vertical arc 
locates the time of utterance: on the left hand side is the 
past, on the right the future. The situation to be referred 
to can be communicated by choosing an icon from a set 
of alternatives (see figure 7). The last two components, 
the frame and the point of reference represent the way 
(point of view) the speaker looks at the situation. 
We start by describing these basic elements and illustrate 
their use through examples. We then explain how these 
elements are related to form a coherent metulinguistic 
vocabulary. 
5.1 The basic elements 
The minimal machinery necessary for generating tense is 
composed of a time axis with a point of speech (now) 
and a means of locating an action/event with respect to 
this point of speech (Figure 5). 
point of speech o 
o 
time axis O 
Figure 5 
Reichenbach used the following concepts to characterize 
tense: a point of speech (S), a point of reference (R) and a 
point of event (E). 8 Furthermore he used two kinds of 
connectors in order to express the fact that two events 
coincide with one another C,"), or that an event preceded 
another C_"). The following formulas shall illustrate his 
classification. 
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 
S,R.E E R S 
I Idl I I I 
R.E S,R , 
,., , 
E S,R S E R 
I I O) ' ' ' 
Figure 6 
present tense I see Paul 
simple past I saw Paul 
present perfect I have seen Paul 
past perfect I had seen Paul 
simple future I will see Paul 
future perfect I will have seen Paul 
Since then the system has been extended or re-interpreted 
by a number of authors \[3a, 12a, 12b, 24, 24a, 25a,28a\]. 
(A) Types of situation and corresponding icons: A 
sentence refers to some point in time (temporal 
situation). A situation will be represented by any of the 
following icons 
8 Voint of speech refers to the very moment in which the 
utterance takes place (now). The point of reference is the 
perspective from which the event is considered. The r~int 
of event is the very moment at which the event took place 
(yesterday). 
AcrEs DE COLING-92, NANTnS, 23-28 AO~r 1992 4 7 8 PROC. OV COLING-92, NArcrl~s, AUG. 23-28, 1992 
Ptul likes ltMimn cotling. 
Ptul ut.eM to pUy ttmnt~. 
Ptul wong the ntce. 
Ptul doav¢ to l,cmdml. 
It hu been r~ininll. 
Psul it pitying ~nm~ 
Paul is dnvbtg to l~tdmt, 
Figure 7 : Types of situation and corresponding icons 
A state is considered as covering a stretch of time with 
no definite boundaries. The icon reperesenting this 
suggests this open-endedness by having fuzzy delimiting 
zones. Both ongoing processes and activities have 
a well-defined starting point but no determined point of 
termination. Note however, that activities can be 
interrupted at any time and be reported as events. Let us 
suppose that "Paul was swimming", and then stopped. 
This can be reported by ~Paul swam,. Telie 
processes do not have this property, because they have 
some natural point of completion. If <<Paul was 
swimming across the Channel~, and then stopped 
swimming, we could describe this situation as Paul has 
been swimming, but we could not say ~Paul swam 
across the Channel.~> In that case, the point of 
completion can be referred to, even if it is never reached: 
~Paul was swimming across the channel, but he never 
made it>>. The situation can also be considered as 
indeteffninate: ,Paul was swimming across the Channel 
when I last heard about him>~. 
An event has both a starting point and a point of 
termination. It can be punctual or not. According to 
Reichenbach it is represented as a point or au interval, 
whereas we allow for the additional use of a frame. An 
event can also be considered with regards to the result of 
its occurrence. There are events with a resulting state. In 
the corresponding icon, we represent this state as adjacent 
to the event which has brought it about. Usually, the 
precise nature of this resulting state needs to be inferred 
(cf. \[10\]): It has been raining; in consequence the 
pavement is wet, it got comer, there are puddles in the 
street, my hairdress is spoiled, etc. 
Itabitual events are an undetermined ||umber of events. 
The icon of habitual events is meant to suggest their 
resemblance to states. It should "also be noted that the 
number of events may be conceptnally indefinite. 
Compare: (a)l played tennis many times last year.(b) 1 
used to play tennis when I was a young man 
In gust case we are dealing with a sequence of individual 
events, not with a habitual event. Hence it is quite 
possible to continue the dialogue by asking the 
following question "how many times did you play?", 
while this question .seems quite odd in the second case. 
(B) Frame of reference, point of reference: Optional 
elements are a frame of reference and a point of reference, 
both of which can be manipulated by the user. The 
former consists of a window which represents the period 
of time considered by the speaker. It is usually associated 
with a temporal adverbial (Yesterday Paul visited the 
Eiffel Tower). The point of reference can be used in order 
to represent some instant which plays the role of a 
translated point of speech (ht 1980 Paul had already 
visited Paris twice.). 
5.2 Using these elements tn produce a time 
scenario 
The procedure is as follows: (a) Choose a suitable type 
of situation (icon) to represent the temporal structure of 
the situation; (b) Position the icon relative to the point 
of speech; (c) If necessary, introduce and position a frame 
and/or point of relerence and position it.(d)In the overall 
representation, some part of the situation should be 
adjacent to, or included in an element of reference 
(point of speech, point of reference, frame of reference). 
If not, reexamine step (c). 
Suppose we have an untensed representation of : ~Paul 
aime la cuisine italienne, (Paul likes Italian cooking). In 
order to generate the tense, we choose a state and the 
position of the point of speech on the time axis (Figure 
8a). From this representation the corresponding time 
(present tense) can be generated. 
~ml --- --imlml 
(a) Pmtul mhne ik cuisine i~licnnc 
_ (Paul likc~t ltaltim cooking) 
, ........... / 
L'mm~ pas~, Patul aimait ht cuisine italiem~ 
n 
(c) t hat. j'fic~ivmi~ mort tuUcle (I was writing my paper ye*terday) \] 
........... ill ..................... ! 
Z 
(d) llier, j'6ctivtis mma irtick 
(1 w~ writing my ~ ye6t~'dJy) 
Figure 8 
Figure 8b conveys the same mes~ge as 8a in the past. 
Since no direct reference to the point of speech is 
available, a frame of reference is necessary. Here it plays 
the same role as the point of speech in figure 8a. If the 
user adds the adverbial ~<l'an pass6,~ (last year) as a marker 
of reference, we get ~<l'an pass6, Paul aimait la cuisine 
ACTES DE COLlNG-92, NANTES. 23-28 ho~r 1992 4 7 9 PROC. OF COLING-92, NANTES, AUG. 23-28, 1992 
italienne~, (Last year Paul liked italian cooking.) The 
interpretation of maimer la cuisine italienne,(to like 
italian cooking) in terms of a state should be 
distinguished from its interpretation as an event. The 
latter would be the instantaneous change from one state 
to another. That's why we need further specifications: 
~Quand Paul a rencontrd Gina, il a aimg la cuisine 
italienne.~ (When Paul met Gina, he began to like Italian 
cooking). Figures 8c and 8d show two possible 
representations of similar, yet different situations.The 
completion of the work is not precisely determined in 
this case. 
Among the difficulties faced by a learner of French is the 
dual value of "pass6 compos6". Consider, (tlier), j'ai dcrit 
mon article - \[ (yesterday) 1 wrote my paper\] In one 
interpretation (figure 9a), the writing of the paper is 
considered as a past event. Locating the action within a 
reference frame requires the use of an adverbial like ~hier>> 
(yesterday). 
9t)11i¢~ jqu &:tit non trtiele (I wrote my paper yesterday) 
....................................................................................................................................................... i ii I::::::~ 
........ ili 
T" I ~ii 
............ •ili 
i \[ 
Figure 9 
This case is to be differentiated from the situation 
depicted in Figure 9b. Here the implicit reference is now. 
The past event is considered with respect to the state 
resulting from it (present relevance). In fact, the same 
tense is used in this ease in French, whereas <<Yesterdayl 
wrote my paper~ would be used in English in case (9a), 
and e.l have written my paper~ in case (9b). 
Another typical case of ambiguity is the use of the 
"present tense" in French. The Figures 9c and 9d, while 
different in English, yield the same form in French. The 
first one (9e) signals habitual or repetitive use of the 
present tense, corresponding to <~I play tennis~. The 
second situation corresponds to "1 am playing tennis", 
implying an ongoing activity at the time of speech. 
Please note that the first situation does not imply any 
actual activity at the time of speech, but only a general 
validity of a kind of property. As the icon suggests, it 
looks very much like a state. 
6. The underlying temporal ontology and its 
use for generation 
6.1. Types of situation and corresponding 
icons 
Our classification of types of situation (cf. also \[2\]) is 
based on conceiving the temporal structure of a situation 
as a result of (a) the lexical meaning, the tense and the 
aspect of a given verb, (b) the nature of the subject, the 
complements and the time adverb'rals. 
Lexical meaning : We basically subscribe to Vendler's 
typology \[26\] who classifies situations in terms of 
states, activities, accomplishments and achievements. 
Nakhimovsky, in a more recent classification \[17\], 
introduces additional features (generic vs. specific 
resources for atclic processes) to Vendler's punctuality, 
stativity, and telieity. Our classification is based on four 
classes of lexical meanings, corresponding to 
combinations of the features: punctuality, stativity, 
telicity. 
• punctual events (punctual, non stative, atelic); 
• states (non punctual, stative, atelic); 
• telic processes. 
• atelic activities; 
Aspect is related to the perspective from which a given 
situation is viewed. Following Comrie \[4\], we 
distinguish between perfect, perfective and imperfeetive 
aspect. In French, the imparfait (continuous paso is a 
typical imperfective form: a situation is considered from 
the inside. Perfective, on the other hand, considers a 
situation as a whole. Perfect is related to the 
consideration of a given event together with a new state 
which is a consequence of this event happening, for 
example, I have prepared some coffee. 
In our system, perfect is represented by an event with a 
resulting state. Aspect plays a dual role: it determines the 
type of situation and it allows for choosing a specific 
frame of reference. 
Lexical meaning combined with the aspect of the verb, 
the types of the subjects and complements, and the 
temporal adverbials yield the type of situation. For 
example, the verb dcrire (to write) has an intrinsic lexical 
meaning : atelic activity. However, it can change : 
Paul dcrivit un roman (John wrote a novel) : telic process; 
Paul dcrit des romans (John writes novels) : atelic activity; 
Paul dcrivait an roman (John was writing a book) : ongoing 
process. 
6.2. How to choose the right tense .9 
The following table shows how tense could be computed 
on the basis of an iconic representation of a situation. 
The system determines the position of the main part of 
the situation relative to the point of speech (PS). The 
qualification is necessary only in the case of an event 
with a resulting state, where the event is the main part of 
the situation. 
If the main part of the situation is posterior to the point 
of speech, a future tense is chosen by default, though 
present tense remains a possible candidate. Using either 
or amounts to communicating a special modal nuance. 
ACT,S DE COLING-92, NArcr~s, 23-28 AOUT 1992 4 8 0 PROC OF COLING-92 NANTES, AUG. 23-28, 1992 
~mr tum~ fdlo. PS {~nt)} 
F 
-1 r{ .... } umla~n ms'm~l 9 amty 
/ / L' ,.~t ~ ~ ~¢~t 
\[.. "t,*t.,,~ 
p,mctu~a *~m 
e~nt w($ I~mlfinll m ~ lit f pu~ ~l~o~ ,~ pr¢~m p*~ t . ~ ®v= --i (;:t,~,im#*)t l~.~,.iz 
Table 1 
(1) Quand le directeur est entrd darts 
la classe, les dtudiants travaill. 
aient (When the headmaster en- 
tered the classroora, the students 
were working ); 
(2) Quand le directeur est entr~ dam 
la classe, les ~tudiants ont tra- 
vaiIM (When the headmaster en- 
tered the classroom, the students 
began to work ). 
See figure 10 
The first case represents an 
ongoing process or activity, with 
an event occuring somewhere in 
the middle of it, while the second 
is a sequence of two adjacent 
events. So far we handle the case 
If the main situation is prior to the time of speech, the 
conclusion depends on the presence of a point of reference 
(PR) prior to the time of speech. If no such point is 
~\],:i " ......... "" - \[::iiiil::i '" .......... " " l!!iil t.,, ..... u, ,.d,. ........ , r 
~1 ~.~.'~'=d ~,,-~a~o,. li~i ~%!::i~i::::~i~::~i~::~!~:: lh~.,~ ...... d ~. ~. \[::~:.~ 
Figure 10 
specified, and if the type of situation is an event, one can 
use the past perfect. In the case of an event or a punctual 
event, the past perfect is a stylistic alternative (polished 
style). 
Temporal adverbials are classified as (a) time-position 
adverbs (last night, on Monday); (b) as span and duration 
adverbs (for a month, up to 1979); or (c) as frequency 
adverbs (usually, often). Temporal adverbs constrain 
tense generation. Durational adverbs may enforce a 
habitual meaning on verbs having an intrinsic punctual 
meaning : <<He has been diving for several months now.~, 
Temporal subordinate clauses: In the case of simple 
sentences, the task of determining the tense from an icon 
is straightforward. More subtle strategies are needed in 
sentences with temporal subordinate clauses 9. Suppose 
the main clause concerns <<students working>~ (PAST), 
whereas the subordinate clause is <<the headmaster 
entering the classroom ~ (PAST). Let the temporal 
conjunction be quand (when). Then we have two 
possibilities: 
9 We leave aside the problem of generating subordinate 
clauses in the subjunctive mood; many temporal 
conjunctions (e.g. avant que : before) require the use of the 
subjunctive mood. 
of quand (when), using icons and strategies similar to the 
ones described by Moons & Steedman \[16\]. 
The way in which temporal subordinate clauses reflect on 
the types and order of the 
situations has been examined in 
the literature (\[24, 15, 16\]). 
7 Conclusion 
We have presented in this paper a 
method of visualizing time, 
tense and aspect. By using icons 
we have provided the user with 
an interface that allows him to 
translate abstract thought into 
concrete thought. The advantage 
being that his thoughts are 
expressed in terms of concrete 
actions rather than in terms of a 
(mere)language, which is often 
too abstract to be meaningful to the nadve user. 
Ac'rEs DE COLING-92, NANTES, 23-28 AOI~T 1992 4 8 2 PROC. OF COLING-92, NANTES, AUG. 23-28, 1992 

References 

\[1\] Allen J. F., Maintaining Knowledge about Temporal 
Intervals, Communications of the ACM 26 (11), 
1983. 832-843. 

2\] Bestougeff H., Ligozat G., Outils logiques pour le 
traitement du temps, Masson. l)aris, 1989. 

\[3\] CL88 , Special lssue on Tense and Aspect. 
Computational Linguistics 14 (2), 1988. 

\[3a\] Borillo, M., Vieu, L.. Eldments pour la formalisation 
du raisonnement spatio-temporel naturcl, Congr/~s 
RFIA, AFCETI', (1989) 

\[4\] Comrie B., Aspect, Cambridge University Press, 
1976. 

\[5\] Dowry D. R., The Effects of Aspeetual Class on the 
Temporal Structure of Discourse: Semantics or 
Pragmatics? ; Linguistics and Philosophy 9 N 1 
(1986) 3%61. 

\[6\] Ehrich V., The Generation of Tense, in Natural 
Language generation, Gerard Kempcn (ed.), 424-440, 
1987, Martinus Nijhoff Publishers, Dordrecht. 

\[7\] Fum D., P. Giagrandi & C. Tasso, ET: an Intelligent 
Tutor for Foreign Language Teaching. In : 
Proceedings of ITS-88, Montreal, Canada, 1988 


\[8\] Fum, D.; P. Giagrandi & C. Tasso, Tense generation 
in an Intelligent Tutor for Foreign Language 
Teaching: some Issues in the Design of the Verb 
Expert. In Proceedings of the 4th Conference of the 
European Chapter of the Association for 
Computational Linguistics, Manchester, UK, 1989 

\[9\] Grasso E., Losmo L., Lombardo V., Maccario P.M., 
Saiato R., Terenziani P., Semantic Interpretation of 
Tense, Actionality and Aspect. Proceedings of the 
9th European Conference on Artificial Intelligence 
(ECAI-90). Stockholm, August 1990, 320-325. 

\[ 10\] Grtvisse M., Prtcis de Grammaire Franeaise. Duculot. 
I979 

\[11\] Hinrichs E., Temporal anaphura in discourses of 
English; Linguistics and Philosophy 9 N 1 (86) 63- 
82. 

\[12\] Hinrichs E., A Compositional Semantics of 
Temporal Expressions in English. in:Proceedings of 
the25th Annual Meeting of the Association for 
Computational Linguistics. 6-9 July 1987, Stanford, 
8-15. 

\[12a\] Hornstein, N., Towards a Theory of Tense, Linguistic 
Inquiry 8, 521-557, (1977) 

\[12b\] Hornstein, N., As Time Goes By. Tense and 
Universal grammar, MIT Press (1990) 

\[13\] Ligozat, G., Weak Representations of Interval 
Algebras, Proceedings of the 8th Conference on 
Artificial Intelligence, (AAAI-90). 1990, Boston, 
715-720. 

\[13a\] Ligozat, G., On generalized interval calculi, 
Proceedings of the 9th Conference on Artificial 
Intelligence, (AAAI-91), 1991, Anaheim, CA, 234- 
240 

\[13b\] Ligozat, G. & Bestougeff, H., Reasoning about 
temporal knowledge, in Klein, E., Pouradier, F & 
Wagner, K.H. (eds.) Betriebslinguistik und 
Linguistikbetrieb, Niemeyer, Tllbingeo, 1991, pp. 
83 -94 

\[14\] McDermott D., A Temporal Logic for Reasoning 
about Processes and Plans, Cognitive Science 6, 
101-155, 1982. 

\[15\] Moens M., Steodman M., Temporal Ontology in 
Natural Language. in: Proceedings of the25th Annual 
Meeting of the Association for Computational 
Linguistics, 6-9 JMy 1987, Stanford., 1-7. 

\[16\] Moens M., Steedman M.. Temporal Ontology and 
Temporal Reference, in \[CL88\], 15-28. 

\[17\] Nakhimovsky A., Aspect, Aspectual Class, and the 
Temporal Structure of Narrative, in \[CL88\], 2943. 

\[18\] Nerbonne J.. Reference Time and Time in Narration; 
Linguistics and PhUosophy 9 N 1 (1986) 63-82. 

\[19\] Passonneau R. J., Situations and intervals, 
in:Proceedings of the25th Annual Meeting of the 
Association for Computational Linguistics, 6-9 July 
1987, Stanford, 16-24. 

\[20\] Passonneau R.J., A Computational Model of the 
Semantics of Tense and Aspect,/n \[CL88\], 44-60. 

\[21\] Quirk R., Greanbaum S., Leech J.. Svartvik J., A 
Comprehensive Grammar of the English Language, 
Longman, London and New York, 1983. 

\[22\] Reichenbech H.. Elements of Symbolic Logic, The 
Free Press, New York, 1966. 

\[23\] Rohrer C. (ed), Time. Tense. and Quantifiers, 
Proceedings of the Stuttgart Conference on the logic 
of tense and quantification, Niemeyer, Ttibingen, 
1980. 

\[24\] Smith C., A Speaker-based Approach to Aspect; 
Linguistics and Philosophy 9 N 1 (1986) 97-115. 

\[24a\] Smith, C., The syntax and interpretation of temporal 
expressions in English, Linguistics & Philosophy 
2, 1, 43-99 (1978). 

\[24b\] Smith, C., A Theory of Aspectual Choice, Language 
53, 3, 1983). 

\[25\] Sows L, Conceptual structures : Information 
Processing in Mind and Machine, Addison-Wesley, 
Reading, Massachusetts, 1984 

\[25a\] van Eynde, F., The Semantics of Tense and Aspect, 
in: Proceedings of EAIA-90, 2rid Advanced School in 
Artificial Intelligence. Natural Language Processing 
in Artificial Intelligence, Filgueiras M., Dumas L., 
Moreira N.. Tomas A.P. (eds.) (1990). 

\[26\] Vendler Z., Verbs and Time, in: Vendler Z., 
Linguistics in philosophy, 97-121, Cornell 
University Press. Ithaca. New York, 1967 (Revised 
version of Veodler Z., Verbs and times, The 
Philosophical Review 66 , 1957, 143-60). 

\[27\] Webber B.L., The Interpretation of Tense in 
Discourse, in:Proceedings of the25th Annual 
Meeting of the Association for Computational 
Linguistics, 6-9 July 1987, Stanford. 147-154. 

\[28\] Webber B.L., Tense as Discourse Anaphor, in 
\[CL881, 61-73. 

\[28a\] Yip, K. M. K., Tense. Aspect and the Cognitive 
representation of Time. in: Proceedings of lJCAl-85, 
806-814 (1985). 

\[29\] Zock, M., If you can't open the Black Box, Open a 
Window \[ or, how to Visualize Dependency 
Relationships when Mapping Meaning onto Form 7, 
In: COGNITIVA-90, Madrid 

\[30\] Zock, M., How to Transform a Black Box into a 
Glass Box ? or, Why use Intermediate Structures in 
Computer Assisted Language Learning, In: 
Proceedings of the 7th Annual Apple European 
University Consortium, Paris 

\[31\] Zock, M. & Laroui, A., Visualising results of choices 
in language generation : the use of intermediate 
structures. In: Proceedings of the Workshop on 
Natural Language Learning. IJCAI-91, Sidney 

\[32\] Zock, M., SWIM or Sink : the Problem of 
Communicating Thought. In M. Swartz & M. Yazdani 
(Eds.) The Bridge to International Communication : 
Intelligent Tutoring Systems for Foreign Language 
Learning, Springer Verlag, Berlin, New York, 1992 

\[33\] Zock, M., Natural Languages are Flexible Tools, 
that's what Makes them Hard to Learn, to Explain and 
to Use. In, M. Zock & G. Sabah (Eds.), Advances in 
Natural Language Generation. coedition Pinter, 
London - Ablex, N.J. 

\[34\] Zock, M. & R. Mitkov. How to ask a foreigner ques- 
tions without knowing his language : propositions 
for a conceptual interface to communicate thought. 
In, Proceedings of the Natural Language Processing 
Pacific RIM Symposium, Singapore 
