<?xml version="1.0" standalone="yes"?>
<Paper uid="C88-2144">
  <Title>Er~glish : and French yields French : Present -4 {post,simul} Past ~-~ {~nte} Future ~ {post} Conditional ~-~ 0 PrEsent ~ {post,simul }</Title>
  <Section position="2" start_page="0" end_page="0" type="metho">
    <SectionTitle>
1. THE PROBLEM
</SectionTitle>
    <Paragraph position="0"> It is a facL of language that the number and the use of the tense and aspect forms are different for every language.</Paragraph>
    <Paragraph position="1"> Even for closely related languages the differences teld to be large. As a consequence, it is not possible to state one-to-one conespondences between the tense and aspect forms of different languages. Some exanlples: EN he has lived in London for 20 years (presem perfect) FR. il vit ~t Londres depuis 20 aus (simple present) EN he has been watching TV for hours (pl~esent perfect progressive) PR il a regard6 la t616 pendant des heures (pl~esent perfect) Differenccg like these pose non-trivial problems for machine translation. In general there are two ways in which they can be handled : either by defining complex mappings from source lant;uage forms to target language forms in transfer SL folan .................. --~ TL form complex mappings or by defining mappings between language specific forms and interlingual meanings in the monoUngual components  meaning .............. ~ meaning 1&amp;quot; identity I mapping I I mapping i $ SL form TL form  Because of Eurotra's policy to keep the bilingual transfer components as small and simple as possible it has been decided tc, pursue the interlingual approach. The resulting system is based on htsighls from  - interval semantics (cf. Bennett, Partee, Dowty, Bruce) - the Reicheubachian analysis of tense and aspect in terms of time of speech, time of reference and time of event (cf. Reichenbach, Johr~von, Smith) - discourse representation theory (of. Kamp, Rohrer, Partee) - descriptive typological studies (cf. Comrie)</Paragraph>
  </Section>
  <Section position="3" start_page="0" end_page="699" type="metho">
    <SectionTitle>
2. THE FORMALISM
</SectionTitle>
    <Paragraph position="0"> As a starting point I take the temporal structure &lt;T,&lt;,ca&gt;, where T is a set of intervals, &lt; is a binary relation that linearly orders time (precedence), and n is a binary operation on intervals (intersection).</Paragraph>
    <Paragraph position="1"> An interval is a continuous subpart of the time line (a). It may consist of one single moment of time (b), but it cannot contain any gaps (c):</Paragraph>
    <Paragraph position="3"> The intersection of two intervals is that subpart of the intervals which they have in common:</Paragraph>
    <Paragraph position="5"> Given the temporal structure &lt;T,&lt;,n&gt;, the number of possible relations between intervals can be determined in a principled way: for any ordered pair of intervals (I and J), it will be the case that either I ca J = O and then &lt;(l,J)</Paragraph>
    <Paragraph position="7"/>
    <Paragraph position="9"> These are the seven logically possible relations between ordered pairs of intervals on a one-dimensional time line.</Paragraph>
    <Paragraph position="10"> For the analysis of single isolated clauses I will nse the Reichenbachian notions of time of speech, time of reference and time of event. The time of event (E) is tile interval for which a basic tenseless proposition is said to be true, and the function of the tense and aspect forms is to define the relation between that interval and the time of speech (S) via the intermediary time of reference (R.).</Paragraph>
    <Paragraph position="11"> For the analysis of clauses in context I make use of a generalised model (cf. 3.4.).</Paragraph>
  </Section>
  <Section position="4" start_page="699" end_page="701" type="metho">
    <SectionTitle>
3. A THEORY OF TENSE
</SectionTitle>
    <Paragraph position="0"> 3.1. file tense meanings Tense meanings will be defined as relations between a time of reference and a time of speech : Rel(R,S).</Paragraph>
    <Paragraph position="1"> The number of possible tense meanings is, hence, equal to the number of possible relations between R and S, which is seven (cf. 2.). However, since the time of speech is generally conceived to be a moment of time rather than an interval of some length, some of these relations carmot hold in principle. The overlap-relations (&lt;&lt; and &gt;&gt;), for instance, can tufty obtain between two intervals of a certain length, and the proper part-of relation (c) cannot hold between R and S either, for if S is one moment of time, R can only be a proper part of S if it is smaller than a moment, which is impossible.</Paragraph>
    <Paragraph position="2"> Furthermore, there seems to be no linguistic evidence lot making a distinction between proper inclusion (w) and identity (=), since &amp;quot;... languages do not have distinct grammatio eal categories of tense indicating location in time at a par-. ticular point vs. location in time surrounding a particular point.&amp;quot; \[Comrie 1985, 123\] As a consequence, the number of possible relations between R and S can be reduced to the following three:</Paragraph>
    <Paragraph position="4"> These COlTespond to the traditional concepts of Past, Future and Present. Notice, however, that the latter is not defined in terms of identity, but in terms of improper inclusion.</Paragraph>
    <Paragraph position="5"> The language specific forms for the expl~ession of these concepts are the tense forms and the time adverbials.</Paragraph>
    <Paragraph position="6"> 3.2. tire deictic time adverbials Typical examples of deictic time adverbials are &amp;quot;now&amp;quot;, &amp;quot;tomorrow&amp;quot;, and &amp;quot;two weeks ago&amp;quot;. Their function is to relate the time of ~eference to the time of speech. Depending on the kind of relation they express they can be charactefised as simultaneous : now ...</Paragraph>
    <Paragraph position="7"> anterior : yesterday, two weeks ago ...</Paragraph>
    <Paragraph position="8"> posterior : tomorrow, next summer ...</Paragraph>
    <Paragraph position="9">  3.3. the tense forms In contrast to the tense meanings which are language independent the tense forms are langnage specilic. Their number, names and distribution differ from language to language.</Paragraph>
    <Paragraph position="10"> As h~r the Eumtra languages there seem to be two type~ of telJse fonn systems: the one of the Romance langllages v~M the one of the Germanic languages and Greek.</Paragraph>
    <Paragraph position="11"> An example of the latter type is English:  There is a bound morpheme \[+/4~:D\] and an optional auxili-ary &amp;quot;will&amp;quot;. The latter can &amp;quot;also have a modal meaning, especially in its past tense lonn, bnt in this context I will only discuss its temporal meaning.</Paragraph>
    <Paragraph position="12"> An examtfle of the former type is French:</Paragraph>
    <Paragraph position="14"> In this system the tense forms are combinations of bound moq)heines; there are no anxiliaries involved.</Paragraph>
    <Paragraph position="15"> As for the meanings of the tense forms they will be defined as elements of the power set of possible tense meanings. This power set contains eight elements : {~, {&lt;), {&gt;), (~}, {&lt;,&gt;1, (&lt;,_~}, (&gt;,~l, 1&lt;,&gt;,~}).</Paragraph>
    <Paragraph position="16"> Not all of these combinations can be assigned to particular lense forms, though, for there are a few general constraints. Bernard Comrie has argued, lbr instance, that &amp;quot;in a tense system, the time reference of each tense is a continuity&amp;quot; \[Comrie 1985, 50\]. This implies that there can be no tense lbm~s which can express posteriority and anteriority without expressing simultaneity as welt. The combination {ante,post} can, hence, be discarded a priori. For the Eurotra languages this restficition appears to hold.</Paragraph>
    <Paragraph position="17"> A second restricition concerns tile combinatiotts {ante,simnl} and {post,simul}. The former is a possible me,'ming in languages which make a basic distinction between Future ({post}) and non-Fut~e ({ante,simul}) ; the latter is a possible meaning in languages which make a basic distinction between Past ({ante}) and non-Past ({post,simul}). Since a h'mguage cam~ot belong to both types at the same time, it follows that for any given langnage either file contbination {post,simul} or the combination {ante,simld} is rnl~ out.</Paragraph>
    <Paragraph position="18"> As far as the Enrotra languages are concerned, they all belong to the latter type.</Paragraph>
    <Paragraph position="19"> In older to find out wtfieh of tile six remaining combina.</Paragraph>
    <Paragraph position="20"> tions can he assigned to file tense forms one can make nse of a grammatie,'dity test : a tense from X can have a meaning Y (where Y is any of {simultaneous, anterior, posterior}), if and only if it can be combined with a deictic adverbial of type Y.</Paragraph>
    <Paragraph position="21"> The application of this test to English the following results :</Paragraph>
    <Paragraph position="23"> The conditional tenses get the value O since they do not have a temporal meaning in single isolated clauses (cf.</Paragraph>
    <Paragraph position="24"> 3.6.).</Paragraph>
    <Paragraph position="25"> 3.4. a discourse model The model presented so far is useftd for the analysis of isolated clauses. For tl~e analysis of texts we need an extension, or rather a generalisation of the original model. The main extensions concern the introduction of another kind of interval, the point of perspective P (the term is bon~wed from \[Rohrer 1985\]), and the addition of indices to the intervals.</Paragraph>
    <Paragraph position="26"> Instead of defining the time of reference with respect to the time of speech I will now define its position with respect to the point of perspective. For any clanse i which is part of a discourse, there will be one peint of perspective Pi and one time of reference Ri. If the clause is the lirst main clause of the discourse, then its point of perspective is derived from the time of speech. In other cases the point of perstx~'ctive will be, derived fl~m tile time of reference of a dominating or preceding clause.</Paragraph>
    <Paragraph position="27">  The notation &amp;quot;V(Pi,Rjy means that Pi is derived from Rj. The interval Rj from wlfich the pesition of Pi is derived will be called the temporal antecedent of the clause with point of perspective Pi. In the example the temporal antecedent of the third clause is &amp;quot;Monday&amp;quot; (R2), the temporal antecedent of the second clause is the time of her promising (R1), and the temporal antecedent of the first clause is the lime of speech (R0=S).</Paragraph>
    <Paragraph position="28"> Notice that file temporal antecedent of a clause i need not always be the time of reference of the immediately preceding clau,~:e (Ri-1). In the sequence (2) she promised that she would come on Monday, but then she changed her mind the temporal antecedent Of the third clause is the thne of her prondsing (R1) rather than &amp;quot;Monday&amp;quot;, and in (3) she promised that she would come on Monday, but now it seems that she cannot the temporal antecedent of the third clause is the time of speech : &amp;quot;now&amp;quot; refers back to the time of speech directly. The differences between the discourse model and the original temporal model are minor: the tense meanings are now relations between Ri and Pi (instead of between R and S) but, since Pi is always a moment of lime (just like S), the number of possible tense meanings remains the same. The expressive power of the tormalism, however, has been enhanced considerably. It now provides a formalism lbr the temporal analysis of all types of clauses - whether enrbedded or not, whether isolated or in context - and for the description of anaphoric temporal expressions. The latter will be discussed in the next paragraphs.</Paragraph>
    <Paragraph position="29"> 3.5. the anaphorie time adverbials There is a class of adverbials which do not refer to the time of speech, as the deictie ones, but rather to the time of relerence of a dominating or preceding clause. They can also be grouped in the three subclasses simultan : the same time, at that moment ...</Paragraph>
    <Paragraph position="30"> anterior : two weeks before, previously ...</Paragraph>
    <Paragraph position="31"> posterior : one week later, then ...</Paragraph>
    <Paragraph position="32"> Together with the deictic adverbials they form the class of relational time adverbials.</Paragraph>
    <Paragraph position="33"> For completeness sake I 'also mention the locational time adverbials, such as &amp;quot;at 2 o'clock&amp;quot;, &amp;quot;on Monday&amp;quot; and &amp;quot;in the summer&amp;quot;. They do not express any relational information and can, therefore, be combined with all possible tcoses.</Paragraph>
    <Paragraph position="34"> Common to both the relational and the locational time adverbials is that they can be used as ,answers to &amp;quot;when&amp;quot;questions. In this respect they differ from the a.';pectual adverbials (cf. 4.3.).</Paragraph>
    <Paragraph position="35"> 3.6. anaphoric tenses The use of the tense forms in texts is somewhat different from their use in single isolated clauses. This is due to the fact that in anterior contexts the present is often replaced by file past and the future by the conditional. This phenomenon, which is called transposition (of. Rohrer 1985), can be seen at work in the following sentences:  (4) he said that he was ill (5) he entered the room and fell on his face  In (4) tbe time of Iris being ill is simultaneous with his saying that he is ill, and in (5) the time of his falling on his face is posterior to the time of his entering the room. In both cases one would expect a present tense in the second clause, bnt since the first clause is in the past, transposition applies and results in the use of the past tense. Tfie discourse diagrams for these sentences look as follows:</Paragraph>
    <Paragraph position="37"> Similar remarks can be made about the use of the conditional in (6) we all hoped that he would soon recover From the point of view of analysis there are at least two ways of dealing witb the phenomenon of transposition: it can be treated as a syntactic transformation or as an irregularity in the relation between form and meaning. In the former case one first maps the past on the present and the conditional on the future and then applies the normal rules for the assignment of meanings. In the latter case one defines extra rules for the assignment of meanings to the past and the conditional tenses.</Paragraph>
    <Paragraph position="38"> The former alternative is more constrained than the latter and, hence, more attractive, but the choice for the one or the other might turn out to be language dependent.</Paragraph>
  </Section>
  <Section position="5" start_page="701" end_page="702" type="metho">
    <SectionTitle>
4. A THEORY OF ASPECT
</SectionTitle>
    <Paragraph position="0"> 4.1. the aspect meahings There is a considerable confusion in the literature about the definition of aspect. This is largely due to the fact that many authors donot make a distinction between grammatical aspect and lexical aspect. The former concerns the syntax and semantics of aspectual auxiliaries and adverbials, whereas the latter concerns the semantics of main verbs and propositions (cf. the event/state/process distinction). In this paper I will use the term aspect for the former only. The latter will be called Aktionsart.</Paragraph>
    <Paragraph position="1"> As a general definition of aspect I will adopt the formulation by Marion Johnson: &amp;quot;What I am proposing concerning the semantiCS of the aspect forms is that they specify the relation between reference time and event time in an utterance.&amp;quot; \[Johnson 1981, 153\] Starting from this definition of aspect meanings as binary relations between intervals and combining it with the observation that the number of possible binary relations between intervals is seven (of. 2.) it is possible to predict that there will be seven aspectual relations. In the following paragraphs I will discuss them in some detail and relate them to the traditional aspectological terminology.</Paragraph>
    <Paragraph position="2"> A well-known aspectual distinction is the one between the perfective and the imperfective. The perfective presents a situation as a single unanalysable whole, whereas the imperfective looks at a situation from the inside and focusses on the beginning, ending or continuation of it (el. Comrie 1976, 3-4).</Paragraph>
    <Paragraph position="3"> As formal counterparts of these definitions I propose the relations =(E,R) and c(E,R) for the perfective :</Paragraph>
    <Paragraph position="5"> These relations express the intuition that ti~ time of event (E) is seen as one unanalysable whole from the point of view of the reference time. The formal definition of the perfective is, hence, ~(E,R).</Paragraph>
    <Paragraph position="6"> For the imperfoctive I will make a distinction between three types. If the focus is on the continuation, the aspect is durative. For its representation I use the relation of proper</Paragraph>
    <Paragraph position="8"> The situation is clearly looked at from the inside : R is in E.</Paragraph>
    <Paragraph position="9"> For the two other types of imperfectivity I will make use of the overlap relations :</Paragraph>
    <Paragraph position="11"> In the ease of right overlap the focus is on the beginning of the situation. This aspect I will call the inchoative. In the case of left overlap the focus is on the end of the situation. This aspect I will call the terminative.</Paragraph>
    <Paragraph position="12"> Another aspect that is often mentioned in the literature is the socalled perfect (# perfective !). In conformity with Reichenbach, Johnson and others I will analyse it in terms of precedence : &lt;(E,R). I will, however, not use the term &amp;quot;perfect&amp;quot; for it, but rather the term &amp;quot;retrospective&amp;quot;. The reason for this is that the perfect aspect form should be distingnished from the retrospective aspect meaning : the former is syntactic, the latter is semantic, and the relation between both is not necessarily one-to-one.</Paragraph>
    <Paragraph position="13"> Finally, there is the inverse of the retrospective, i.e. the prospective : &gt;(E,R). It is one of the meanings of the English &amp;quot;be going to&amp;quot; and of the French auxiliary &amp;quot;aller&amp;quot;. In short, there are six different aspect meanings. Their language specific counterparts are aspeetual auxiliaries and adverbials.</Paragraph>
    <Paragraph position="14"> 4.2. the aspectual auxiliaries As for the Eumtra languages the aspect form systems show a larger diversity than the tense form systems.</Paragraph>
    <Paragraph position="15"> Some typical aspectual distinctions are the ones between  - perfect and non-perfect (have + past participle) - progressive and non-progressive (be + present participle) - go and non-gu (go + to-infinitive)  The first distinction is made in all of the Eumtm languages, but the two other ones are not eumversal. They are present in English, but not ha German and Danish, for instance, and French has the third distinction, but not the second. It may be worth stressing that I will only analyse the aspectaal anxiliaries. Full lexical verbs, such as &amp;quot;stop&amp;quot;, &amp;quot;start&amp;quot; and &amp;quot;continue&amp;quot;, and periphrastic forms, such as &amp;quot;~tre en train de&amp;quot; and &amp;quot;venir de&amp;quot;, will not be discussed here. l?br English, the set of aspectual auxiliaries can be defined as follows:</Paragraph>
    <Paragraph position="17"> the definition looks as follows :</Paragraph>
    <Paragraph position="19"> aller = Futur proche aller pleuvoir As tor the assignment of meanings to the auxiliaries I will follow the same procedure as for the tense meanings. The meaning of an aspect form is an element of the power set of possible aspect meanings. This set contains 64 elements. For the dr;fruition of the mappings one can start from the following euroversal scheme (euroversal = common to the  If one of these tbrms is not present in the language, its meaning may he expressed by another form. In general this will be the form whose basic meaning is the least distant from the meaning to be expressed. For a specification of the notion of distance between meanings I will use the following scheme: retm term perf, dur incho pro</Paragraph>
    <Paragraph position="21"> The distance between two ,aspect meanings is equal to the difference of their numbers.</Paragraph>
    <Paragraph position="22"> it follows frmn the principle of minimal distance that a language without a progressive will express the durative by means of the form which expresses the perfective (13-31=0), i.e. the simple form. This is indeed tree for French, Dutch, German and Danish.</Paragraph>
    <Paragraph position="23"> Tbe principle also guides the choice of a form for the expression of file terminative. Some languages have a special form for this aspect. English, for instance, has the per\[izct progrcssive for this meaning. Most languages, however, donot have such a tbrm and in those cases the minimal distance principle predicts which forms can be used for the expression of terminativity, i.e the (retrospective) perfect lbrm, the (perfective) simple form or the (durative) progressive foim: 12.-11=12-31=1 (see also 4.3.).</Paragraph>
    <Paragraph position="24"> Taking into account the basic scheme and the principle of minimal distance, ,and complementing it with language specific observations, one can derive the following mappings for English and French:  4.3. aspectual adverbials Tile aspectual adverbials include the duration adverbials and the boundary adverbials.</Paragraph>
    <Paragraph position="25"> The duration adverbials specify the length of the time of event. Depending on whether the basic proposition is an event or a state/process they are expressed by an 1Nadverbial or a FOR-adverbial:  (7) she ran the mile in five minutt~ (8) he has been sleeping for ten hours (9) we have been in France for a month  They do not express any relational information.</Paragraph>
    <Paragraph position="26"> The boundary adverbials specify the beginning and/or the end of the time of event. They are prepositional pbrases introduced by &amp;quot;since&amp;quot;, &amp;quot;from&amp;quot;, &amp;quot;until&amp;quot;, &amp;quot;till&amp;quot;, &amp;quot;from .. till&amp;quot;. One of these expresses relational information: the &amp;quot;since&amp;quot;adverbials denote an interval which begins in the past at some specified time, e.g. Christmas in &amp;quot;since Chrismlas&amp;quot;. The end point of such an interval is not specified by the adverbial, but is normally taken to be included in the time of reference. The relation between time of event and time of reference will, hence, be one of overlap:</Paragraph>
    <Paragraph position="28"> It tbllows that &amp;quot;since&amp;quot; adverbials express terminativity and that the compatibility of these adverbials with the aspect forms can be used as a test for deciding whether a given aspect form can be terminative.</Paragraph>
    <Paragraph position="29"> What the aspectual adverbials have in common is that they can be used as answers to &amp;quot;how long&amp;quot;-questions. This distinguishes them from the time adverbials.</Paragraph>
  </Section>
  <Section position="6" start_page="702" end_page="703" type="metho">
    <SectionTitle>
5. DEGREES OF COMPOSITIONALITY
</SectionTitle>
    <Paragraph position="0"> Tense and aspect forms do not occur in isolation: finite verbs have both a tense form and an aspect form. The meaning of their combination is the relational product of the meanings of the tense form and the meanings of the aspect form. An example: the meaning of the English present perfect progressive is the relational product of the meanings of the present tense with the meanings of the perfect progressive aspect. In other words, the meaning of the present perfect progressive is compositional.</Paragraph>
    <Paragraph position="1"> Not all combinations of tense and aspect are compositional, though. In some cases a form can have a meaning which cannot be derived eompositionally, in other cases a form may lack a meaning which can be derived compositionally.</Paragraph>
    <Paragraph position="2">  An 'example of the former type is the present perfect in languages like French, Dutch and German. Apart from or even instead of its compositional meanings this form has an &lt;anterior, peffective&gt; meaning, i.e. the meaning of the English simple past. This appears a.o. from their compatibility with anterior time adverbials: FR je l'ai vu bier DU ik beb hem gisteren gezien GE ich habe ihn gestem gesehen EN * I have seen him yesterday I saw him yesterday An example of the latter type is the French passd simple. The simple aspect can have three different meanings in French, but in combination with the passg, it can only have the perfective interpretation: the durative and the terminative meaning are expressed by the imparfait.</Paragraph>
    <Paragraph position="3"> Depending on how many exceptions there are, the tense and aspect system of a given language will be more or less compositional. In ease of a low degree of compositionality one could decide to assign meanings to combinations of tense and aspect forms, rather than to tense and aspect forms separately.</Paragraph>
  </Section>
  <Section position="7" start_page="703" end_page="703" type="metho">
    <SectionTitle>
6. THE SYSTEM IN USE
</SectionTitle>
    <Paragraph position="0"> Eurotra is a transfer based system. The integration of the given analyses in the Eurotra framework has been achieved as follows.</Paragraph>
    <Paragraph position="1"> In analysis the tense and aspect forms are mapped onto their meanings. This mapping is many-to-many and will, hence, result in the assignment of many meanings to one and the same form. Disambiguation is done on the basis of the context. Factors to be taken into account are the temporal adverbials and the Aktionsart of the basic proposition. null In transfer the tense and aspect meanings are simply copied: their representations are interlingual.</Paragraph>
    <Paragraph position="2"> In generation the meanings are mapped onto forms. Unlike the mapping in analysis, this mapping is a function.</Paragraph>
    <Paragraph position="3"> The system presented in this paper has been applied to the nine Eurotra languages and has been implemented in terms of the unification based Eurotra formalism. Still lacking at this moment are the treamaent of the transposed uses of the tenses and the rules for determining the Aktionsart of basic propositions.</Paragraph>
  </Section>
class="xml-element"></Paper>