<?xml version="1.0" standalone="yes"?>
<Paper uid="P98-1007">
  <Title>Tense and Connective Constraints on the Expression of Causality</Title>
  <Section position="3" start_page="48" end_page="50" type="metho">
    <SectionTitle>
2 Consequence-Cause Configuration
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="48" end_page="49" type="sub_section">
      <SectionTitle>
2.1 Data
</SectionTitle>
      <Paragraph position="0"> Even if a causality (the second sentence introducing the cause of the first one) is pragmatically possible in all these examples, we observe that a sequence PS-PS imposes in French a temporal sequence interpretation: in all the examples (3), the main eventuality of the second sentence is interpreted as temporally located after the one of the first sentence, and this is strictly incompatible with a causality, where cause must precede its effect. Notice that here Ps in French behaves differently from simple past in English. 5  (3) a. Jean tomba. La branche cassa Jean fell. The branch broke b. Jean attrapa une contravention.</Paragraph>
      <Paragraph position="1"> Il roula trop vite Jean got fined. He drove too fast c. Marie cria. Jean lui cassa la figure  Marie cried. Jean hit her d. Jean prit sa vulture. Il la r@para Jean took his car. He repaired it e. Jean se salit. Il r@para sa voiture Jean dirtied himselPS He repaired his car Now, if one chooses, with the same order of presentation, the tense combination PS-IMP, the causality effect is easily achieved. This is the case for the examples (4).</Paragraph>
      <Paragraph position="2">  (4) a. Jean attrapa une contravention.</Paragraph>
      <Paragraph position="3"> I1 roulait trop vite Jean got a fine. He was driving too fast b. Marie cria. Jean lui cassait la figure Marie cried. Jean was hitting her 5The translation of the ambiguous example (2a) (Lascarides and Asher, 1993) is not ambiguous in French where no causal interpretation is available (2b). (2) a. John fell. Max pushed him.</Paragraph>
      <Paragraph position="4"> b. Jean tomba. Max le poussa.</Paragraph>
      <Paragraph position="5">  However, this choice is not always applicable, since it can give rise to unacceptable forms: (5)  are either incorrect, or do not convey causality. (5) a. @ Jean tomba. La branche cassait Jean fell. The branch was breaking b. @ Jean prit sa voiture. Il la r6parait  Jean took the car. He was repairing it The connective &amp;quot;donc&amp;quot; can be used in such configurations, without changing acceptability. The denoted relation in this case concerns both the epistemic level (attitudinal) and the descriptive level (propositional) (Jayez and Rossari, 1998). We consider in this paper only uses of &amp;quot;donc&amp;quot; where the epistemic level does not change fondamentaly the relation. 6</Paragraph>
    </Section>
    <Section position="2" start_page="49" end_page="50" type="sub_section">
      <SectionTitle>
2.2 Discussion
</SectionTitle>
      <Paragraph position="0"> We think that these acceptabilities can be explained if one takes into account two principles: one concerns causality itself in connection with aspectuality, the other concerns the French IMP'S ability to act as an aspectual operator.</Paragraph>
      <Paragraph position="1">  To account for the contrast between (4) and (5), we have to be more specific about the way causality can hold between eventualities. Let us assume el is the cause of e2. We can distinguish  two cases: 1. el has to be completed to be the cause of e2. For instance, the breaking of the branch has to be completed before Jean can fall; Jean's car has to be repaired before he can drive it.</Paragraph>
      <Paragraph position="2"> 2. it is not necessary for el to be completed  to be the cause of e2. For instance, starting to repair the car is enough to be the cause of one's getting dirty; driving too fast is enough to get a fine, independantly of the completion of el.</Paragraph>
      <Paragraph position="3"> We call the first case accomplished causality. Notice that this distinction is independant of the aspectual class of the verb describing the eventuality. It is only a matter of world knowledge. 6In this configuration, &amp;quot;car&amp;quot; (.for) is the non marked connective. Its introduction does not change notably the acceptability jugements, we leave the examination of its specific constraints for another study.</Paragraph>
      <Paragraph position="4">  One of the most important properties of IMP is that it imposes an imperfective (durative, non accomplished) view on the eventuality (Vet, 1980). The way this effect operates can be described the following way, assuming the usual partition of predicates into the four Vendler's (1967) aspectual classes.</Paragraph>
      <Paragraph position="5"> States, activities These eventualities, either homogenious (states) or not (activities), are non terminative, in the sense that they do not have a natural term (end) (e.g., to know the truth--state, to run--activity). Then IMP is entirely compatible, thus have no particular effect.</Paragraph>
      <Paragraph position="6"> Achievements, accomplishments These are characterised by the existence of a natural term. The imperfective point of view brought by IMP imposes a change of point of view on the term of the eventuality.</Paragraph>
      <Paragraph position="7"> As for accomplishments, we can assume that they can be decomposed into several stages, according to (Moens and Steedman, 1988): first a preparatory phase, second a culmination (or achievement) (we are not concerned here with the result state). We can then say that IMP refers only to the preparatory phase, so that the term of the eventuality loses all relevance. This explains the so-called imperfective paradox: it is possible to use IMP even though the  eventuality never reaches its term: (6) a. I1 traversait la rue quand la voiture  l'a 6cras6 He was crossing the street when the car hit him b. * I1 traversa la rue quand la voiture l'a 6cras6 He crossed the street when the car hit him As for achievements, we can assume that they are reduced to a culmination. Then IMP can only be interpreted by stretching this culmination, transforming a fundamentaly punctual event into a process or activity. Then there is no more natural term for such a stretched event.</Paragraph>
      <Paragraph position="8">  So, when we have a non accomplished causality, i.e., when it is possible to state the cause rela- null tion as soon as the eventuality has started, then IMP does not impose further constraint, and the sequence PS-IMP is always correct, and conveys the appropriate causality effect. This is the case for the examples (4, 7), where an explanation discourse relation is infered.</Paragraph>
      <Paragraph position="9"> (7) Jean se salit. I1 rfiparait sa voiture Jean got dirty. He was repairing his car On the contrary, if we have an accomplished causality, i.e. if the cause event has to be completed to be a cause for the other event, then IMP is never possible, for even with terminative eventualities (the branch breaking, fixing the car), it has the effect of blocking the terminativity, and a causal interpretation is no longer possible (5). The contrast (8) can thus be easily explained: in (8a), we have a lexically punctual event, made durative by the IMP. But going through a red light has to be completed to risk a fine; in (8b), we have an activity, and it is sufficient to have started it to risk a fine.</Paragraph>
    </Section>
    <Section position="3" start_page="50" end_page="50" type="sub_section">
      <SectionTitle>
2.3 Application
</SectionTitle>
      <Paragraph position="0"> The consequences of the observations and the hypotheses made earlier, when it comes to text generation, are the following: If one wants to present two eventualities related by a cause relation, so that the consequence is presented before the cause, leading to an explanation interpretation of the discourse,  one should obey the following principles: 1. A PS-PS combination is not appropriate.</Paragraph>
      <Paragraph position="1"> 2. A PS-IMP combination conveys causality,  provided that we have a non accomplished causality. Otherwise, the PS-IMP combination is not valid.</Paragraph>
      <Paragraph position="2"> We should note again that these constraints are not lexical, in the sense that they do not rely on aspectual classes, but rather on world knowledge.</Paragraph>
    </Section>
  </Section>
  <Section position="4" start_page="50" end_page="52" type="metho">
    <SectionTitle>
3 Cause-Consequence Configuration
</SectionTitle>
    <Paragraph position="0"> Let us now turn to the other mode of presentation, namely the one where cause is presented before its consequence. We first consider cases without connectives, and see that good acceptabilities go along with higher ambiguity: correct example do not always convey causality (SS 3.1). Then we consider the use of the connective &amp;quot;donc&amp;quot;, and observe that it changes the acceptabilities (SS 3.2).</Paragraph>
    <Section position="1" start_page="50" end_page="51" type="sub_section">
      <SectionTitle>
3.1 Without connective
3.1.1 Data
</SectionTitle>
      <Paragraph position="0"> The first observation is that it is possible to use a PS-PS sequence. In the absence of other discourse clues, such a sequence is interpreted in French as a temporal sequence relation. Such a temporal interpretation is compatible with, but of course does not necessary imply, a cause re- null lation.</Paragraph>
      <Paragraph position="1"> (9) a.</Paragraph>
      <Paragraph position="2"> b.</Paragraph>
      <Paragraph position="3"> C.</Paragraph>
      <Paragraph position="4"> d.</Paragraph>
      <Paragraph position="5">  La branche cassa. I1 tomba dans le vide The branch broke. He fell down Paul vit sa demande rejet~e. IIen r~digea une autre Paul's application was rejected. He wrote an other one I1 rut nomm~ PDG.</Paragraph>
      <Paragraph position="6"> I1 contr61a tout le personnel He was appointed chairman.</Paragraph>
      <Paragraph position="7"> He had control over the whole staff I1 appuya sur la d~tente. Le coup partit. He pressed the trigger. The gun went off Changing the PS-PS sequence into a PS-IMP changes only marginally the acceptabilities, and the same observation as before holds: these discourses do not necessarily imply causality.  For instance, (10b-c) can also be interpreted as background discourses, where the IMP of the second sentence is seen as introducing a background situation holding before and after the event introduced in the first sentence. This interpretation, often given as the default one for IMP-PS sequences (Kamp and Rohrer, 1983), is nevertheless only available when world knowledge does not exclude it (10a). In any case, such an interpretation is incompatible with a causal interpretation.</Paragraph>
      <Paragraph position="8">  So it turns out that PS-IMP sequences can have in general two interpretations: one where the two events follow each other, and this interpretation is thus compatible with a causality interpretation, and another one where the eventuality described by the IMP sentence overlaps with the event given before.</Paragraph>
      <Paragraph position="9"> This can be explained if one assumes the operation of IMP as described in (Molendijk, 1994), in a DRT framework, itself inspired by (Reichenbach, 1947).</Paragraph>
      <Paragraph position="10"> One of the features of IMP is to state the simultaneousness of the eventuality described with some reference point (henceforth Rpt), located in the past of the speech time. This operation can be called anaphoric, since IMP needs some other point given by the context. This is clearly what happens with the background effect. But it has also been shown, in particular by Tasmowski-De Ryck (1985), that there are some uses of IMP (called imparfait de rupture-&amp;quot;breaking IMP&amp;quot;) which are not strictly anaphoric, in the sense that the Rpt cannot be identified with any previously introduced event. Rather, it seems that such uses of IMP strongly entail the existence of an implicit Rpt, distinct from the events already introduced. It is also observed that this ability of IMP to bring with it a Rpt is constrained. In particular, there must be a way to connect this Rpt to the other eventualities of the discourse. Molendijk (1996) shows that this connection can be a causal relation. It has also been observed that an implicit Rpt is always temporally located after the last event introduced. So this is compatible with a causality interpretation.</Paragraph>
    </Section>
    <Section position="2" start_page="51" end_page="51" type="sub_section">
      <SectionTitle>
3.1.3 Application
</SectionTitle>
      <Paragraph position="0"> From a text generation point of view, the observations we have just made cannot be easily exploited: obviously, in a Cause-Consequence configuration, all the tense combinations we have seen are not informative enough, and cannot be used, if one wants to guarantee that the concept of causality is conveyed by the discourse.</Paragraph>
      <Paragraph position="1"> It is thus necessary to be more explicit, for instance by adding a connective. This is what we are concerned with in the next section.</Paragraph>
      <Paragraph position="2"> So, if we leave apart the PS-PS sequence, what we have seen so far in SS 2 is that the tense combination is sufficient to convey a causality relation in Consequence-Cause configurations, and then the connectives do not impose further constraints and do not change what is conveyed.</Paragraph>
      <Paragraph position="3"> The situation in this section (SS 3) is in a way symetrical: in a Cause-Consequence configuration, the tense configuration is not sufficient, so that adding a connective is necessary. But, as we see in the next section, there are further constraints on the connectives.</Paragraph>
    </Section>
    <Section position="3" start_page="51" end_page="52" type="sub_section">
      <SectionTitle>
3.2 With the connective &amp;quot;doric&amp;quot;
3.2.1 Data
</SectionTitle>
      <Paragraph position="0"> One can observe that &amp;quot;donc&amp;quot; is perfectly compatible with PS-PS sequences like the ones in (9).</Paragraph>
      <Paragraph position="1"> What is more surprising is that adding &amp;quot;donc&amp;quot; to the PS-IMP sequence examples we have seen  (10) clearly changes the acceptabilities: (11) a. ?? La branche cassa. I1 tombait donc  dans le vide The branch broke. He was therefore falling down b. Paul vit sa demande rejet~e. IIen r~digeait doric une autre Paul's application was rejected. He was therefore writing another one c. I1 fut nomm~ PDG. I1 contr61ait donc tout le personnel He was appointed chairman. He was therefore having control over the whole staff d. ?? I1 appuya sur la d~tente.</Paragraph>
      <Paragraph position="2"> Le coup partait donc.</Paragraph>
      <Paragraph position="3"> He pressed the trigger. The gun was therefore going off The clearer contrast concerns cases where the second sentence contains an activity verb. In such cases, the introduction of&amp;quot;donc&amp;quot; leads systematically to bad sentences. On the contrary, it seems that &amp;quot;donc&amp;quot; is always compatible with state and accomplishment verbs.</Paragraph>
      <Paragraph position="4"> As for achievements, it seems that the introduction of&amp;quot;donc&amp;quot; also yields bad sentences, but  it is worth noting that the simple sequence PS-IMP without connective is already slightly problematic, as we have seen in (10d). We come back to this point later.</Paragraph>
      <Paragraph position="5">  We are not yet able to provide a completely elaborated explanation for these observations. What we propose here is a list of possible answers, suggested by more fine-grained considerations on data.</Paragraph>
      <Paragraph position="6"> Note however that from the previous observation we can draw the principle that we can generate sentences in a Cause-Consequence configuration, with a PS-IMP sequence, and the connective &amp;quot;doric&amp;quot; but the aspectual class of the verb has to be taken into account. It leads to acceptable sentences only with accomplishments and states.</Paragraph>
      <Paragraph position="7"> It is clear that aspectual classes play a role, which is not surprising, and this is the reason why all our example lists comprise each time one verb from each aspectual class.</Paragraph>
      <Paragraph position="8"> The most problematic contrast concerns the difference between activities and accomplishments. The connective &amp;quot;donc&amp;quot; seems to work very well with accomplishments and very bad with activities, even though accomplishments can be seen as composed of an activity followed by a culmination. One possible explanation could rely on the observation that the result relation brought by &amp;quot;donc&amp;quot; holds not at the propositionnal level, not even at the aspectual (i.e., point of view on events), but rather at an attitudinal level (Rossari and Jayez, 1997). Besides, one can observe that what distinguishes activities and accomplishments is not the nature itself of the eventuality, but rather the fact that one expects/considers the culmination of it in one case and not in the other. So this can be seen as a difference of (propositional) attitude over the eventualities. We are presently working on the elaboration of a proposal based on this viewpoint. It is also worth observing that the temporal interval that lies between a cause and its consequence might play a role, as suggested by (Jayez, 1998), especially for this contrast between activities and accomplishments.</Paragraph>
      <Paragraph position="9"> As for achievements, we have already noted that their compatibility with IMP is not entirely established, for reasons coming from the punctual nature of achievements. It is also worth noting that there is an affinity between achievements and &amp;quot;imparfait de rupture&amp;quot; (Tasmowski-De Ryck, 1985). Of course, as suggested by its name, such use of IMP introduces a sort of break in the discourse, which is of course compatible with causality, but might not be compatible with the way &amp;quot;donc&amp;quot; operates, requiring a strong connection between two utterances.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>