Scaling causal relations and connectives
in terms of speaker involvement
HENK PANDER MAAT and LIESBETH DEGAND*
Abstract
This article presents a scalar approach to the semantics and distribution of
causal connectives in Dutch and French. The proposed scale is called the
speaker-involvement scale, and is de®ned in terms of the degree of implicit
involvement of the speaker in the construction of the causal relation. First,
a number of causal coherence relations identi®ed in the literature are
reanalyzed in terms of increasing speaker involvement, yielding the following
order: nonvolitional, volitional, causal epistemic, noncausal epistemic,
speech-act relations.
Subsequently, a number of frequent causal connectives in Dutch and French
are investigated in terms of speaker involvement. Initially, we examine the
distribution of these connectives in terms of the ®ve relational interpretations.
These distributions are taken as ®rst indications of the speaker-involvement
levels encoded by the di€erent connectives. The speaker-involvement pro®les
thus developed are further supported by observations regarding the semantic
e€ects of substituting a connective with a di€erent speaker-involvement
level. It is found that connectives are not necessarily tied to the expression
of a particular coherence relation. The relative independence pertaining
between connective meanings and categories of relational interpretation
yields expressive possibilities for speakers who want to introduce assumptions
to, or remove them from, the interpretation of a certain relation.
Finally, more ®ne-grained di€erences in speaker involvement were
revealed by analyzing the utterance containing the connective for linguistic
features having to do with subjectivity. One of the results of this kind of
analysis is that, in all relations, high speaker-involvement connectives more
often accompany segments with a ®rst-person protagonist than do other
connectives.
Keywords:
coherence relations; connectives; categorization; causality;
subjectivity; speaker involvement.
Cognitive Linguistics 12±3 (2001), 211±245
0936±5907/01/0012±0211
# Walter de Gruyter
212 H. Pander Maat and L. Degand
1. Introduction
It has often been noted that causal connectives come in (at least) two kinds.
On the one hand, we have connectives like French de ce fait, English as a
result, and Dutch daardoor, which seem to specialize in relations between
states of a€airs:
(1)
a. Le soleil se leva. De ce fait la tempeÂrature grimpa.
b. The sun came up. As a result, the temperature went up.
c. De zon kwam op. Daardoor steeg de temperatuur.
On the other hand, connectives like French donc, English so, and Dutch
dus may occur in so-called nonassertive speech acts (see example [2]) and
speaker conclusions (see example [3]), while de ce fait, as a result, and
daardoor are inappropriate in such environments.
(2)
a. Tu as eÂte treÁs impoli. Donc /*de ce fait sors immeÂdiatement de
cette pieÁce!
b. You have been very impolite. So /*as a result leave the room
immediately!
c. Je bent erg onbeleefd geweest. Dus /*daardoor verlaat de kamer
onmiddelijk!
(3) a. La neige fond, donc /*de ce fait la tempeÂrature est au-dessus de
zeÂro.
b. The snow is melting, so /*as a result the temperature is above
zero.
c. De sneeuw smelt, dus /*daardoor de temperatuur ligt boven nul.
The di€erence between the two sets of connectives has mainly been
accounted for in terms of dichotomies (such as external vs. internal
[Halliday and Hasan 1976; Martin 1992], or semantic vs. pragmatic [van
Dijk 1977; Moeschler 1989]). Roughly, the di€erence is seen as one
between relating the contents of utterances and relating the utterances
themselves. A similar dichotomy has been proposed to classify coherence
relations as such, that is, apart from their linguistic realization (e.g.,
subject matter vs. presentational [Mann and Thompson 1988], semantic
vs. pragmatic [Sanders et al. 1992; Sanders 1997], or propositional vs.
intentional relation [Knott 2001]). A somewhat di€erent account is
provided by Sweetser (1990), who has proposed distinguishing not two,
but three domains of use of use for connectives: the content domain,
relevant for (1), the speech-act domain, exempli®ed in (2), and the
epistemic domain, illustrated in (3).
In this article, our objective is to propose an alternative account
of the distribution of connectives. Going beyond dichotomous and
Scaling causal relations and connectives 213
trichotomous classi®cations, we want to propose a scale of speaker
involvement, on which the inherent expressive power of connectives can be
represented. Two observations lay at the basis of this proposal.
First, most connectives that allows epistemic and speech-act uses may
also express so-called ``volitional'' causal relations, while as a result and
daardoor may not, and de ce fait is at least marked here:
(4)
a. J'eÂtais fatigueÂ, donc /? de ce fait je suis parti.
b. I felt tired, so /*as a result I left.
c. Ik was moe, dus /*daardoor ik ging weg.
However, all the publications mentioned thus far unanimously state that
volitional causality falls within the content domain, since it is concerned
with states of a€airs in the real world, external to the discourse. But note
that dus, donc, and so cannot express non-volitional real-world relations
(see example 4').
(4')
a. Il y a eu beaucoup de vent, #donc trois tuiles sont tombeÂes du toit.
b. There was a strong wind, #so three tiles fell o€ the roof.
c. Er stond een harde wind, #dus er zijn drie pannen van het dak
gevallen.
In these examples, the connectives do not express a nonvolitional content
relation; rather, they suggest that the consequence is entirely predictable
from the cause. This is what makes these examples decidedly odd. In other
words these epistemic and speech-act connectives encroach upon the
content domain, but only on the volitional part of it.
The second observation concerns the behavior of some connectives that
seem to speci®cally express volitional causality, like c'est pourquoi, that's
why, and daarom. While these connectives are not acceptable in epistemic
and speech-act contexts (examples [5] and [6]), they become sxxx as soon as
speech-act or modal operators are inserted1 (see [5' ] and [6' ]).1 At the same
time, nonvolitional connectives are excluded, with the possible exception
of de ce fait:
(5)
a. La neige fond. *C'est pourquoi /*De ce fait, la tempeÂrature est
au-dessus de zeÂro.
b. The snow is melting. *That's why /*As a result, the temperature is
above zero.
c. De sneeuw smelt. *Daarom /*Daardoor ligt de temperatuur
boven nul.
(5') a. La neige fond. C'est pourquoi /?De ce fait, je pense que la
tempeÂrature est au-dessus de zeÂro.
b. The snow is melting. That's why /*As a result, I think the
temperature is above zero.
214 H. Pander Maat and L. Degand
c.
De sneeuw smelt. Daarom /*Daardoor denk ik dat de
temperatuur boven nul ligt.
(6) a. Tu as eÂte treÁs impoli. *C'est pourquoi / ? De ce fait, sors de cette
pieÁce immeÂdiatement!
b. You have been very impolite. *That's why /*As a result, leave the
room immediately!
c. Je bent erg onbeleefd geweest. *Daarom /*Daardoor verlaat de
kamer onmiddelijk!
(6') a. Tu as eÂte treÁs impoli. C'est pourquoi / De ce fait, je te demande
de sortir de cettte pieÁce immeÂdiatement!
b. You have been very impolite. That's why /*As a result, I demand
that you leave the room immediately!
c. Je bent erg onbeleefd geweest. Daarom /*Daardoor vraag ik je
de kamer onmiddelijk te verlaten!
Hence, these volitional connectives seem to take an intermediate position
between the content and epistemic relations. On the one hand, they may
express a certain type of relation in the content domain, on the other hand,
they lend themselves to use in epistemic and speech-act contexts as long
as the speech acts and conclusions are an explicit part of the proposition.
In our opinion, this situation is an indication of the need to apply
a scalar perspective to the spectrum reaching from nonvolitional causality
in the content domain to epistemic and speech-act causality. Our
hypothesis is that the di€erent relational interpretations can be ordered
along a scale from minimal to maximal speaker involvement. Speaker
involvement (SI) refers to the degree to which the present speaker is
implicitly involved in the construal of the causal relation. More speci®cally,
speaker involvement increases with the degree to which both the causal
relation and the related units are constituted by the assumptions and
actions of the present speaker. That is, the maximal degree of speaker
involvement represents the situation in which the causal event is identical
to the speech event.
Four aspects of coherence relations may enhance the prominence of
speaker assumptions in the relation and hence enhance its level of speaker
involvement. First, causal coherence relations may be characterized in
terms of the degree to which they necessarily imply the subjective
involvement of a conscious participant, which may, but need not be
the speaker (see also Pit 1997). For instance, a volitional relation
(see example [4]) involves such a participant, while a nonvolitional
relation (as in example [1]) does not. Normally, both nonvolitional
and volitional relations are seen as relations in the content domain.
However, since conscious participants may entertain assumptions
Scaling causal relations and connectives 215
which the speaker may share, volitional relations carry a higher degree of
speaker involvement than nonvolitional ones. See section 2.2 for details.
Second, a given causal relation may be more or less isomorphic with a
causal relation in the real world, that is, a causal relation in which the
present speaker has no role. For instance, the volitional relation in (7)
represents a real-world causal relation, while the relation in (8) does not; it
is made up of speaker assumptions. That is, in (8) the reason±consequence
relation in the real world is transposed to the mental domain of speaker
knowledge and inference. Hence the epistemic relation in (8) carries a
higher degree of speaker involvement. In example (9) this process is taken
one step further. Here the causal relation in the real world is turned around
to yield a pattern of abductive inference. In this situation the epistemic
relation is detached from the real-world one.
(7)
(8)
(9)
John loved her, so he came back.
John loved her, so he probably came back.
John came back, so he probably loved her.
Third, a given relation may be placed a di€erent distances from the present
speaker, and at di€erent distances from the moment of speaking. The
closer a given relation is to the present speaker, the more it constitutes a
vehicle for the expression of assumptions made by the speaker. After all,
the speaker is more likely to accept the general assumptions underlying
his own decisions than those underlying the decisions of others. Hence, the
®rst-person relation in (10) exhibits a higher degree of speaker involvement
than the third-person relation in (11), and (12) is even higher in speaker
involvement because it is in the present tense.
(10)
(11)
(12)
I felt tired. I left.
He felt tired. He left.
I feel tired. I'm going home.
Fourth, the involvement of a conscious participant may vary in
explicitness. The more implicit it is, the more subjectively this participant
is construed, to use the term coined by Langacker (1990); that is, it
functions as an implicit reference point for the conceptualization of the
causal relation.
We will illustrate this dimension with reference to the presence of the
speaker in an epistemic relation (examples [13] to [15]). Taken in isolation,
(13) expresses a factual statement, originating outside the domain of the
speaker's conceptualizing activity. That is, it presents itself as carrying
a minimal degree of speaker involvement, because the speaker is only
minimally involved in the production of the reported cognition. In
contrast, (14) and (15) contain some explicit elements referring to the
216 H. Pander Maat and L. Degand
speaker's perspective (comparable to Langacker's [1990] notion of
ground ). In example (14), the modal adverb probably invokes the present
speaker as the source of the probability judgement. This is made even more
explicit in (15), where the speaker's perspective (I think) is more or less
``objecti®ed'', that is, is made part of the situation referred to in the
utterance. This can be seen by the fact that it may be referred to by
demonstrative pronouns in subsequent utterances (following [15], that is
not true could refer to the I think clause, not just to its complement). In
terms of speaker involvement, (14) ranks higher than (15), while (15) is still
higher than (13).
(13)
(14)
(15)
He is Hungarian.
He is probably Hungarian.
I think he is Hungarian.
However, such factual statements as (13) can occur in a discourse
context that heavily suggests they be interpreted as a speaker conclusion. This would be the case, for instance, if (13) were preceded by
He can only be Czech or Hungarian and I am sure he is not Czech. In this
case, the speaker's inferential activity is part of the interpretation, but
at the same time maximally implicit. Hence, on this occasion, (13), and
the epistemic relation it entertains with the preceding utterances, exhibit
a maximal degree of speaker involvement higher than in (14) and
consequently also higher than (15).
In the remainder of this article, we ®rst analyze ®ve prototypical
relational interpretations exemplifying increasing degrees of speaker
involvement, thus elucidating the speaker-involvement scale. In the
second part of article, we study the interaction between a number of
Dutch and French causal connectives and these ®ve relational environments. We will report on corpus studies empirically supporting a number
of claims concerning the speaker-involvement scale.
2. From nonvolitional to speech-act relations
2.1. Nonvolitional causal relations
Instances of nonvolitional causality appear as entirely objective phenomena (in the real world), that is, as phenomena in which no observer has
any part. The causality is presented as concerning states of a€airs, which
are simply taken as fact; the same goes for the causal relation, which is also
presented as a fact in which the speaker is not involved. Of course, we
might also say that reporting something as a fact constitutes a minimal
degree of subjective involvement, in which the speaker assumes no role
other than that of being the reporter.
Scaling causal relations and connectives 217
Apart from the speaker (who reports on the relation), no conscious
participant is involved in the nonvolitional relation. In example (16), a
potentially conscious participant is involved in the segments but not in the
relation, since the cause±e€ect relation falls outside the domain of his
intentions. The situation referred to in the ®rst segment need not be known
by the protagonist. Actually, this ignorance may be the prime cause of the
event taking place.
(16)
Monday his train to Amsterdam arrived at another platform. He
got on the train to The Hague. (Stukker et al. 1999: 72)
Nonvolitional causal relations obey certain temporal constraints. Given
that the state of a€airs p is valid at time point or interval t1 and the state of
a€airs q is valid at t2, t2 cannot precede or start earlier than t1; and, in the
case that t1 and t2 are not identical, the causal event itself takes place
somewhere between t1 and t2.
Oversteegen (1997) and Lagerwerf (1998) have suggested that connectives like because presuppose a general underlying causal relation,
de®ned in terms of a defeasible implication. More speci®cally, they claim
that because p, q presupposes p'wq', where p and q are clauses and p' and q'
are propositions associated with p and q respectively, andwis a defeasible
implication.2 It is interesting to explore whether this idea of underlying
general implications can also be applied to causal coherence relations as
such, that is, aside from their being either marked or not by connectives. In
our opinion, this is indeed the case for volitional and epistemic causal
relations. But for nonvolitional relations, a (defeasible) implication cannot
be assumed: in interpreting (17) we do not assume that landslides normally
cause the loss of electricity supplies in four neighborhoods. In order to
understand this relation, the only assumption one needs to make is that
causes like the one presented may have e€ects like the one presented, that
is, natural disasters may cause a loss of electricity.
(17)
There were landslides in Malibu last week. Four neighborhoods
lost their electricity. (Mann and Thompson 1986: 65)
The degree of generalization to which the underlying relation has been
subjected can only be determined by invoking knowledge of the world.
In some cases it will be dicult to generalize both p and q. However, in
the absence of any general relation the causal relation will remain
incomprehensible or even undetected. For instance, in (18a) a causal
relation can be identi®ed on the basis of a general relation between
noise and the inability to concentrate; such a relation cannot be
constructed for (18b).
218 H. Pander Maat and L. Degand
(18)
a. The neighbor is playing the trumpet. I cannot concentrate on
my book.
b. The neighbor is working in the garden. I cannot concentrate
on my book.
We will now summarize this analysis in more formal terms.3
Nonvolitional causality
relational segments: pt1; qt2 ; p and q stand for states of a€airs
causal relation:
[p ? q]t1±t2
time constraint:
t1¦t2
Additional assumption
General possibility of the relation: ^(P ? Q)
2.2. Volitional causal relations
In contrast to nonvolitional causality, volitional causality (in the real
world) involves decision making by some protagonist, in this case an
actor; and decision making implies reasoning. Reasoning in turn implies
awareness of premises: what is causally e€ective in example (19) is not the
state of a€airs p (the fact that it was 12 p.m.) but its representation by the
protagonist. In example (20), the situation is even more complex. It is hard
to consider the guests' being boring as a simple state of a€airs. Rather, it is
an evaluative predication, which the actor believes to be valid, i.e., the
protagonist judges the guests to be boring and this is a reason for her to
leave the party.
(19)
(20)
It was 12 p.m. She/I went home.
The guests were boring. She/I left the party early.
The fact that p falls within the mental domain of the actor (compare
Fauconnier's [1994] idea of mental spaces), is the ®rst di€erence between
volitional and nonvolitional relations. This may be made explicit in the
®rst segment, as in She thought the guests were boring. She left, but most
often it is not. In the latter case, it is part of the assumptions inherent to
the volitional interpretation. This assumption can only be conveyed by a
speaker who, at least temporarily, adopts the viewpoint of the actor.
In other words, in reporting a volitional causal relation, the speaker
implicitly becomes involved in the construal of the causal relation. Hence
we claim that volitional relations presuppose a certain degree of speaker
involvement.
The temporal constraints in volitional causal relations are more
complex than those in nonvolitional relations. For instance, the segment
Scaling causal relations and connectives 219
containing the reason for a certain action can refer to a state of a€airs
in the future:
(21)
Her/my plane takes o€ tomorrow at 6 a.m. She/I left the party early.
In this example, the state of a€airs designated by p takes place at a time
point succeeding that of the state of a€airs designated by q. This again
indicates that volitional relations do not related the states of a€airs as
such; instead, they relate the representation of a state of a€airs to the
performance of a certain action.
So far we have shown that the cause segment of the volitional relation is
not a state of a€airs in the real world. How about the consequence segment
of this relation? Can we consider a volitional action as merely a simple
event in the external world? Partly, because such actions are, in principle,
observable for other participants. On the other hand, every (volitional)
action involves the particular mental state of an actor, which we could call
actor awareness: one cannot perform an action without being aware of
performing precisely this action. In Fauconnier's (1994) mental-space
terminology, we might say that action predicates automatically set up
actor spaces. In reporting a volitional action, the speaker also conveys the
actor awareness assumption. Again, most of the time this assumption
tends to be unnoticed, presumably because the speaker normally shares the
assumption that this action is indeed the one performed. A divergence
between the actors' and the speakers' perspective need to be marked. For
instance, (22) is acceptable whilst (23) is not.
(22)
(23)
She thought she took the train to The Hague, but she took the train
to Amsterdam.
*She took the train to The Hague, but she took the train to
Amsterdam.
Thus, both the cause and the consequence segment of a volitional causal
relation are (at least partially) situated in the mental domain of the actor,
which is implicitly shared by the speaker when conveying the situation.
The same goes for the causal relation itself: ®rst, causation is a mental
phenomenon since it constitutes a process of reasoning; second, the
protagonist herself is aware of this process, in which believing p leads to
knowingly performing action q. Again, in unmarked cases the speaker
shares this assumption with the actor. When the speaker disagrees with this
assumption, that is, when the speaker thinks the action was not performed
for this reason, he must indicate this explicitly, as in (24b).
(24)
a. *It rained all afternoon. That's why she didn't go walking.
However, the real reason was that she had a paper to ®nish.
220 H. Pander Maat and L. Degand
b.
It rained all afternoon. That's why she said she didn't go
walking.
However, the real reason was that she had a paper to ®nish.
Thus far, we have said that believing p is what led to the action in the
second segment. However, this account is still insucient. Intentional
participants are endowed with a certain minimum of consistency. Hence,
it is natural to infer a general pattern of cause and e€ect underlying the
speci®c sequence that is reported. That is, the protagonist of (20) is
presented as a person who generally considers boring guests a reason for
leaving a party early, though boring guests will not always be sucient to
actually make her leave. The same goes for the protagonist in example (19)
who generally accepts midnight to be an adequate time to go home.
Moreover, in volitional relations the protagonist is always aware of this
general motivational relation. Further, she also accepts this relation in
a normative sense. An essential feature of volitional causation is that
the acceptance of the normative principle is part of the reason for
acting: the protagonist does not go home merely because it is 12 p.m.,
but because she thinks it is 12 p.m. and she accepts this to be a good
reason for leaving a party. In other words, from the protagonist's point
of view a reason is always a good reason.
As with nonvolitional relations, the level of generality of the relation
must be determined contextually. In this example, it is doubtful whether
party may be generalized to social gatherings, since the protagonist may
not consider the time (midnight) or the level of entertainment (boring
guests) relevant at all when it comes to leaving or not leaving birthday
parties of family members. In the same way, the situation of Cinderella
leaving at midnight cannot be generalized to a situation in which her
leaving a party at midnight is always conditioned by the nearing end of an
enchantment; rather, the general underlying principle could be something
like obedience, or her wish not to be ridiculed. In other words, the principle
underlying a volitional relation cannot always be derived immediately
from the segments. However, some general principle must be at work.
(See also Itkonen [1983] for a conceptual analysis of actions and their
rationality.)
This underlying principle need not always be accepted by the speaker.
One may describe actions that follow a ``logic'' of which one does not
approve. For instance, the speaker himself may regard it as bizarre to leave
a party at 12 o'clock. Nevertheless, in reporting the relation, the speaker at
least temporarily adopts the perspective of the protagonist, especially
when, as is often the case, the ®rst segment is not marked for (speaker)
perspective.
Scaling causal relations and connectives 221
Moreover, in some cases the relation may apply to actors in general, not
just to the protagonist in question. As well will see later, the choice of a
connective may a€ect this aspect of the volitional interpretation.
In sum, volitional relations can only be understood by constructing a
number of assumptions concerning the knowledge and attitudes of the
protagonist, assumptions that are usually left unstated, and are hence
temporarily adopted by the speaker. Our analysis is summarized as
follows:
Volitional relations
relational segments:
causal relation:
time constraint:
Bt1(a, ps/pe), possibly realized in unmarked form as
ps/pe
PFt2(a, qa), possibly realized in unmarked form as qa
[Bt1(a, ps/pe) ? PFt2(a, qa)]t1±t2
t1¦t2
Additional assumptions
1. Awareness of the causal relation: Bt1±t2(a, [Bt1(a, ps/pe)?PFt2 (a, qa)]t1±t2)
2. Generality of the relation: B(a, P) ? PF(a, Q)
Generally, actor a performs PF an action like q when he believes
something like p.
3. Actor acceptability of the relation: ACCt1±t2 (a, n2o)
At the time of causal event, a considers the general assumption as an
acceptable way of doing things.
2.3. Causality-based epistemic relations
In causality-based epistemic relations, a segment describing a real-world
cause constitutes a reason for drawing a certain conclusion regarding its
(real-world) consequence. Like reasons for actions, premises in arguments
are known or accepted by the concluder. It is the propositional attitude
that is causally e€ective, not its content. Unlike volitional relations,
however, in epistemic relations the consequence is not some state of a€airs
in the spatiotemporal world, but a mental state of the protagonist, i.e., the
concluder (c). By default, this concluder is the speaker him or herself. So
when an epistemic relation does not refer to the protagonist, the speaker
generally conveys a conclusion of his own. When this is not the case, this
must be explicitly mentioned in the second segment as in (25), or in earlier
utterances as in (26).4
(25)
It had rained continuously for two days. Peter thought the tennis
court would be unplayable.
222 H. Pander Maat and L. Degand
(26)
Peter looked out of the window. It had rained continuously for two
days now. The tennis court would probably be unplayable.
Since epistemic relations have no spatiotemporal real-world segments,
they are not observable. Hence a speaker presenting such a relation must
entirely adopt the internal perspective of the protagonist, even when he or
she is not the protagonist. Therefore, speaker involvement in epistemic
relations is higher than in volitional ones, where the second segment is at
least partially a spatiotemporal phenomenon.
In causality-based epistemic relations, the reasoning is based on an
assumption concerning a real-world causal relation. Concluding q from p
is based here on the belief that p generally causes q in the real world, or that
knowing p will generally lead an actor to opt for a certain action. That is,
the causality upon which the inference is based may be of a nonvolitional
(see example [27]) or a volitional (example [28]) kind. In both cases, the
causation takes place in a (real-world or mental) domain di€erent from the
mental domain of reasoning. With regards to time constraints, this means
that one has to consider two layers: the reasoning process and the
underlying causal process. The epistemic relation as a reasoning process
relates two mental states, the second which cannot precede the ®rst. The
two states may be temporally coextensive or the ®rst state may have an
earlier onset than the second. As regards the real-world causal event, the
time constraint is identical to the ones discussed earlier with regard to
nonvolitional and volitional relations.
(27)
(28)
It has rained continuously for two days. The tennis court will
probably be unplayable.
It has rained continuously for two days. Peter will probably not go
to the tennis court today.
Like the protagonists in volitional relations, epistemic protagonists are
also thought to be consistent. As a result, epistemic relations are understood against the background of assumptions relating premises and
conclusions. In this case, the generality assumption not only concerns the
relation between two beliefs, but rather the relation between knowing
about a certain real-world causal pattern and a pattern of reasoning
modeled on this real-world pattern. In contrast to volitional relations, this
assumption is taken to apply to persons in general, not solely to the
protagonist.
This di€erence between volitional and epistemic relations with regard to
the general validity of the underlying assumption is also manifest in the
way we talk about conclusions and actions, respectively. Conclusions are
often described as (in)correct, strong, weak, faulty, and the like; actions
Scaling causal relations and connectives 223
are not. According to the reasoning protagonist, there is a community
of concluders who will accept it has rained for two days as an argument
for the conclusion that the tennis court will be unplayable, since continuous rainfall often causes it to be unplayable. There is a smaller group
of concluders that also accepts the argument as a decisive one, as the
speaker/concluder does. However, the speaker does not necessarily
assume that the present hearer is a member of this smaller group.
He assumes only that the hearer belongs to the larger group for whom the
acceptance of his argument increases the probability of his conclusion
being correct.
In sum, epistemic causality is not necessarily tied to the protagonist
speci®ed in the propositions themselves; typically it relates to a
``contextually salient'' protagonist: the speaker c and a group of
concluders C to which he belongs.
Causality-based epistemic relations
relational segments: Bt1(c, p), possibly realized as p; Bt2(c, q), possibly
realized as q
causal relation:
[Bt1(c, p) ? Bt2(c, q)]t1±t2
time constraint:
t1¦t2
Additional assumptions
1. Awareness of the epistemic causal relation: Bt1±t2(c, [Bt1(c, p) ?
Bt2 (c, q)]t1±t2)
2. Underlying ``external'' causal relation: Bt1±t2[c, ( p ? q)]
3. ``External'' relation leads to the epistemic relation:
n2o ? [Bt1(c, p) ? Bt2(c, q)]t1±t2
4. Generality of the external relation: {B[C, (P ? Q)]}
Generally, a concluder C believes that P causes Q.
5. Generality of the epistemic relation: {B[C, (P ? Q)]} ? {[B(C, P)?
DB(C, Q)]}
Generally, when a concluder C believes that P causes Q, believing P
leads him to increased belief in conclusion Q.
6. Concluder acceptability of the relation: ACCt1±t2(c, n5o)
At the time of the causal event, c considers it acceptable for every
concluder C to reason in this very same way.
2.4. Noncausal epistemic relations
In causality-based epistemic relations, the real-world cause is taken as the
argument supporting a conclusion concerning the real-world e€ect. That
is, the causal relations in the real world and the epistemic domain have the
224 H. Pander Maat and L. Degand
same direction. In noncausal reasoning, either the real-world causality has
a di€erent direction to the epistemic one, or real-world causality is not
relevant at all.
The ®rst case concerns abductive reasoning: the real-world e€ect is
taken as an argument supporting a conclusion concerning the real-world
cause. In (29), the observation of melting snow is used to infer its cause,
namely a temperature above zero degrees. These kinds of examples have
become prototypical of epistemic coherence relations (Sweetser 1990;
Sanders et al. 1992). This is unfortunate for two reasons. First, the
majority of epistemic relations encountered in discourse are causality
based (see the following); second, besides abduction there are other
kinds of reasoning not modeled on real-world causality. What underlie
the arguments in (30) and (31) are not any kind of causal relation between
the situations referred to in the relational segments, but rather other kinds
of assumptions: most people leave for work well before 10 a.m. (the fact
that it is 10 a.m. does not cause them to have left), and twin brothers
are usually the same height (one of them being tall does not cause the other
to be tall).
(29)
(30)
(31)
The snow is melting. The temperature must be above zero.
It's 10 o'clock. Everyone has probably left for work.
His twin brother is very tall. He will probably be tall too.
These are only some examples taken from a larger set of inferences, some
of which have been analyzed and categorized, while others have not. Some
of the better known members of this set include reasoning from a dilemma
( p or q; not p; hence, q) and from an analogy (for p, A is the case; q is
comparable to p; hence, for q, A is the case too).
In our view, noncausal inferences exhibit a larger degree of speaker
involvement than causality-based epistemic relations. The di€erence
here lies not in the number or the strength of the assumptions being
conveyed, but in their nature. Causality-based inferences simply transpose
a real-world link into the inferential domain. By contrast, the assumption
behind abductive and other types of noncausal inference in no longer
modeled on real-world causal links, although it may sometimes be based
on real-world regularities. With regard to real-world causality, however,
the assumption is non-iconic, and springs entirely from mental activity by
the speaker.
Noncausal epistemic relations
relational segments: Bt1(c, p); Bt2(c, q)
causal relation:
[Bt1(c, p) ? Bt2(c, q)]t1±t2
time constraint:
t1¦t2
Scaling causal relations and connectives 225
Additional assumptions
1. Awareness of the epistemic causal relation: Bt1±t2(c, [Bt1(c, p) ?
Bt2(c, q)]t1±t2)
2. The acceptance of some inversely causal or noncausal relation Rp, q
between p and q, e.g., a disjunction or similarity: Bt1±t2(c, nRp, qo)
3. The acceptance of Rp, q leads to epistemic relation:
Bt1±t2 (c, nRp, qo)?[Bt1(c, p)?DBt2(c, q)]t1±t2
4. Generality of the relation:
{B(C, nRp, qo)}?{[B(C, P)?DB(C, Q)]}
Generally, when a concluder C believes that Rp, q, believing P leads
him to increased acceptance of conclusion Q.
5. Concluder acceptability of the relation: ACCt1±t2(c, n4o)
At the time of the causal event, c considers it acceptable for every
concluder C to reason in this very same way.
2.5. Speech-act relations type 1: Motivate a speech act
So far we discussed relations which concern the way a speaker
conceptualizes a reality that is external to the speech situation. This
includes causal and noncausal reasoning, since reasoning patterns exist
independently of their expression in discourse (thinking is not saying).
By contrast, speech-act relations concern the structure of the present
discourseÐand nothing else. They appear in discourse in response to the
interactional needs of a speci®c/potential interlocutor, not to present
facts or draw conclusions concerning the real world. In these kinds of
relations, the speaker is not involved as a thinking being, but solely in
his role as a speaker. In (32) and (33), two speech-act relations are
illustrated in which one of the segments prepares the ground for an
upcoming speech act.
(32)
(33)
I want to pay you for your work. How long have you been cleaning?
There is a good movie on. Did you already have plans for tonight?
Like epistemic relations, speech-act relations are based on assumptions
that need to be shared between speaker and hearer. In example (32), the
speaker wants the hearer to consider his question to be an appropriate
speech act, he appeals to an assumption that is acceptable to the hearer:
when you want to pay somebody for (simple) work, it is appropriate to ask
how much time he or she spent on it (assuming you do not already know).
Likewise, in (33) the hearer needs to consider the request for information
on his plans for tonight as an appropriate speech act, given the fact that
there is a good movie on. Of course, not just any hearer can be approached
226 H. Pander Maat and L. Degand
with such a request and the motivation behind it; the complex speech act
sa presupposes a certain view of the social relationship between speaker s
and hearer h. In this regard, the speech-act relation is more speci®cally
hearer-oriented than the epistemic relation, in which the hearer is only
addressed indirectly as a member of the community sharing certain
assumptions. The epistemic relation in turn is more hearer-oriented than
the volitional relation, which does not assume anything with regard to the
hearer's views concerning the appropriateness of the ensuing action.
Other characteristics of the speech-act relation are that the actor is
always identical to the speaker and that both segments take place at the
time of speaking. Finally, the fact that the second segment of the relation
consists of the performance of some speech act is necessarily implicit.
Speech-act relations type 1
relational segments: Bts(s, ps/pe), possibly realized as ps/pe; PFts(s, h, saq),
realized as q
causal relation:
[Bts(s, ps/pe) ? PFts(s, h, q)]t1±ts
time constraint:
t1¦ts
Additional assumptions
1. Awareness of the action performed: Bts(s, PFts[s, h, saq])
2. Awareness of the causal relation:
Btl±ts(s, [Bt1(s, ps/pe) ? PFts(s, h, saq)]t1±ts)
3. Generality of the relation: B(s, P) ? PF (s, h, SAQ)
Generally, when he believes something like p, speaker s performs an
action like saq with regard to h.
4. Speaker acceptability of the relation: ACCt1±ts(s, n3o)
At the time of the causal event, s considers the general assumption as
an acceptable way of doing things.
5. Hearer acceptability of the relation: ACCts(h, n3o)
At the time of the speech act, h considers the general assumption as an
acceptable way of doing things.
2.6. Speech act relations type 2: Paraphrasing and summarizing
(34)
To win the elections an absolute majority (more than half of the
votes) is required.
In example (34), a certain discourse segment, sap an absolute majority, is
re-expressed in segment saq more than half of the votes. The assumption
is that in this context the interpretation of speech act sap licences an
inference identical to the interpretation of saq. This identity is unknown to
Scaling causal relations and connectives 227
the hearer at the point of processing sap, otherwise saq would not be
relevant. However, after processing saq the hearer is intended to be aware
of the licensing relation. A second assumption is that saq will present less
comprehension problems for the hearer than sap does. Again, the present
hearer needs to share this assumption. When he does not, he may feel
underestimated or he might still be puzzled by the second speech act.
As in the ®rst kind of speech-act relations, the speaker and the
hearer are heavily involved in the relational segments without being
mentioned as such. A di€erence, however, lies in the generality of the
relation. The interpretative relation between the two speech acts need
not be general, since interpretations of utterances are by their nature
context dependent (as opposed to meanings of linguistic items). For
instance, in (35) the ®rst segment receives an entirely context-speci®c
interpretation.
(35)
Par contre, 38% des sondeÂs ... preÂfeÁrent que la situation actuelle reste
inchangeÂe, donc que le droit de vote reste reÂserve aux Belges.
On the contrary, 38% of the respondents ... prefer the present
situation to remain unchanged, that is, the right to vote remains
reserved to Belgians.
At present, we do not postulate a di€erence in speaker involvement
between speech-act relations of the ®rst and second type. Our characterization of these two relations shows that both of them embody a speci®c
hearer-directness, which distinguishes them from epistemic relations in
our hierarchy.
Speech-act relations type 2
relational segments: It1(s, h, sap), realized as p; Its(s, h, saq), realized as q
causal relation:
It1(s, h, sap) ? Its(s, h, saq)
time constraint:
t1¦ts
Additional assumptions
1. Awareness of the interpretations involved: Bts[s, (It1[s, h, sap])] and
Bts[s, (Its[s, h, saq])]
2. Awareness of the causal relation: Bts[s, (It1[s, h, sap] ? Its[s, h, saq])]
3. Speaker acceptability of the relation: ACCt1±ts(s, n2o)
At the time of the causal event, s considers the causal relation an
acceptable interpretation process.
4. Hearer acceptability of the relation: ACCts(h, n2o)
At the time of the speech act, h considers the causal relation an
acceptable interpretation process.
228 H. Pander Maat and L. Degand
2.7. Summary
The discussion so far may be summarized in terms of the four aspects of
speaker involvement mentioned in the introduction (see Table 1):
i. Causal coherence relations may be characterized in terms of the
degree to which they necessarily imply the active involvement of a
subject of consciousness, which may be the speaker himself. When
segments change from states of a€airs into actions, beliefs, or speech
acts, and when the role of the protagonist changes from none at all
into actor, concluder, and speaker, respectively, speaker involvement
clearly increases. Our discussion has also revealed di€erent degrees of
hearer involvement, which increases from no hearer involvement at all
to indirect involvement in the case of epistemics to direct involvement
in the case of speech-act relations.
ii. The second aspect concerns the connection between the discourse
relation on the one hand and real-world causality on the other. In
nonvolitional and volitional relations, a type of real-world causality
is represented. In epistemic causal relations a real-world causal
relation is transposed to the mental domain of inference making. In
this kind of relation, real-world causality is not represented, but it
continues to impose constraints on the propositional content of the
related segments. This is no longer the case for noncausal epistemic
relations and speech-act relations, which show maximal detachment
from real-world causality.
iii. Third, a given relation may be placed at di€erent distances from the
present speaker and at di€erent distances from the moment of
speaking. The ®rst kind of distance progressively decreases as the
identity of the protagonist changes from none at all into ``any person''
and then into the speaker. In epistemic relations, the speaker is the
unmarked protagonist, and in speech-act relations he or she is
the only conceivable protagonist. The time distance to the moment
of speaking decreases from no temporal constraint at all for
nonvolitional and volitional relations to an obligatory coincidence
of t2 and ts for speech-act relations.
iv. The involvement of a protagonist may be more or less implicitly
realized in the discourse. As we proceed on the scale, explicit
realization gradually changes into obligatory implicit realization.
Table 1 recapitulates the dimensions that characterize the di€erent
causal relations. As already mentioned, some of these dimensions are
interdependent. For instance, the dimension ``Identity of the protagonist''
constrains the dimension ``Realization of the protagonist''. The direction
Segments
states of a€airs
belief; action
beliefs
beliefs
belief; speech act
interpretations of speech acts
Type of
relation
Nonvolitional
Volitional
Causal epistemic
Noncausal epistemic
Speech act type 1
Speech act type 2
none
actor
concluder
concluder
speaker
Role of
protagonist
none
none
indirect
indirect
direct
Hearer
involvement
Table 1. Summary of speaker-involvement dimensions of causal relations
representation
representation
transposition
detachment
detachment
Connection
to real-world
causality
no protagonist
any person
speaker
speaker
speaker
Identity of
unmarked
protagonist
none
none
often t2~ts
often t2~ts
t2~ts
Time
constraint
on t2
±
rarely implicit
often implicit
often implicit
implicit
Realization
of the
protagonist
Scaling causal relations and connectives 229
230 H. Pander Maat and L. Degand
of development in each dimension is clear, and most upward transitions on
the scale are marked by more than one corresponding transition in one of
the dimensions. Hence Table 1 shows how the dimensions work together in
making a relation into more or less a vehicle for the expression of speakers'
assumptions and activities.
3. The speaker-involvement scale
Speaker involvement (SI) is a characteristic of relational interpretations.
Hence it can be used for analyzing
i. the interpretational options in a stretch of discourse;
ii. the semantic pro®le of a connective (at least a central component
of it); and
iii. the interaction between these two.
Up until now we have been concentrating on the ®rst topic; the rest of
the article will be devoted to the other two issues. A ®rst impression of the
interaction between connectives and their discourse environments can be
gained by considering inappropriate combinations of connective and
discourse contexts, i.e., ``starred'' fragments. In our proposal, these are due
to discrepancies in speaker-involvement pro®les. That is, the connective
encodes a higher or lower degree of speaker involvement than is allowed
for by the discourse context. The ®rst case was illustrated by example (4'),
in which the connective dus/so introduces rather strong assumptions (see
assumptions 5 and 6 in section 2.3) that appear implausible in view of the
discourse segments. The second case was illustrated by examples (5) and
(6), in which the relational interpretations suggested by the segments
carry a degree of speaker involvement which apparently cannot be
expressed by the connectives daarom, c'est pourquoi, and that's why. When
the protagonist's mental activity is made explicit in the second segment,
however, the speaker-involvement level of the discourse segments
decreases, with the result that these connectives can be used.
More generally, we want to make the following claims concerning the
relationship between connectives and speaker involvement:
1.
2.
A connective encodes a certain speaker-involvement level, which it
contributes to the interpretation of its discourse environment. When
this level is too low or too high to be combined with the level allowed
for by the discourse environment, the use of the connective is
inappropriate.
The set of relational environments of a connective can be represented
as an area on the speaker-involvement scale. Our expectation is that
the distribution of each connective occupies a contiguous area on the
Scaling causal relations and connectives 231
scale. That is, we do not expect to ®nd a single connective that may
express, e.g., volitional and noncausal epistemic relations but not
causal epistemic relations. In order to see why this should be so we
need to realize that a connective contributes a certain level of speaker
involvement to the interpretation of its relational environment.
Hence, a connective that may be used in volitional and noncausal
epistemic relations apparently carries a level of speaker involvement
compatible with these relations. However, these two relations
themselves also cover a certain area on the speaker-involvement
scale. This area happens to include the speaker-involvement location
of causal epistemic relations. Therefore the speaker-involvement level
encoded by the connective in question must also be compatible with
the speaker-involvement levels allowed for by causal epistemic
relations.
3. The most frequent causal connectives in a given language should di€er
signi®cantly from each other on the scale. That is, the prime reason for
a language to have more than one causal connective is to be able to
express several levels of speaker involvement.
4. These claims are cross-linguistically valid (starting with Dutch and
French).
In our view, analyses of the speaker-involvement potential inherent to
connectives cannot do without systematic corpus analyses. Hence, we
carried out corpus analyses of forward causal connectives in (written)
Dutch and French.
4. Empirical support for the speaker-involvement scale
4.1. A contrastive speaker-involvement analysis of causal connectives
We selected the three, for French four, most frequent connectives
expressing forward causation in the two languages: dus, daarom, and
daardoor for Dutch; donc, deÁs lors, c'est pourquoi, and de ce fait for French.
For each of these connectives, we assembled 50 occurrences from a
newspaper corpus.5
First, we identi®ed the coherence relation of each fragment by means of
a paraphrase test.6 The results for Dutch and French are given in Tables 2
and 3, respectively. In both languages, the contiguity hypothesis (the
second claim in section 3) is con®rmed.
For Dutch, dus-fragments are clearly highest on the speaker-involvement
scale in terms of relational interpretations, daardoor-fragments clearly
lowest, and daarom-fragments occupy an intermediate position. All
connectives signi®cantly di€er from one another (dus±daarom: w2~25.90,
232 H. Pander Maat and L. Degand
Table 2. Relational interpretations co-occurring with forward causal connectives in a Dutch
newspaper corpus
Relation
Connective
dus
(%)
daarom
(%)
daardoor
(%)
Nonvolitional causal
Volitional causal
Causal epistemic
Noncausal epistemic
Speech act, type 2
±
6
21
19
4
±
12
42
36
8
1
26
18
5
±
2
52
36
10
±
50
±
±
±
±
100
±
±
±
±
Total
50
50
50
Table 3. Relational interpretations co-occurring with forward causal connectives in a French
newspaper corpus
Relation
Connective
donc
(%)
deÁs lors
(%)
c'est pourquoi
(%)
de ce fait
(%)
Nonvolitional causal
Volitional causal
Causal epistemic
Noncausal epistemic
Speech act
±
8
26
12
4
±
16
52
24
8
1
8
32
7
2
2
16
64
14
4
2
33
15
±
±
4
66
30
±
±
33
7
10
±
±
66
14
20
±
±
Total
50
50
50
50
df~4, and p<.0001; dus±daardoor: w2~100.00, df~4, and p<.0001;
daarom±daardoor: w2~96.08, df~3, and p<.0001).
We may also regard Table 2 from a di€erent point of view. We could
ask which relational distinctions are lexicalized in Dutch. A distinction
between two relational interpretations is lexicalized when lexical items
exist which may express one but not the other relation. Taken in this sense,
the distinction between nonvolitional and volitional causality is lexicalized
by dus and daardoor, and the distinction between noncausal epistemic
relations and speech-act relations by daarom.
For French, donc and deÁs lors-fragments are highest on the speakerinvolvement scale in terms of relational interpretations, followed by c'est
pourquoi-fragments, which in turn are more subjective than de ce faitfragments (donc±c'est pourquoi: w2~36.20, df~4, and p<.0001; deÁs lors±
c'est pourquoi: w2~30.73, df~4, and p<.0001; donc±de ce fait: w2~56.18,
df~4, and p<.0001; deÁs lors±de ce fait: w2~50.71, df~4, and p<.0001;
c'est pourquoi±de ce fait: w2~45.36, df~2, and p<.0001). Our third claim
Scaling causal relations and connectives 233
is supported for French with the exception of the pair donc and deÁs lors.
We will return to this issue later.
In French, the distinction between nonvolitional and volitional causality
is lexicalized by donc, and the distinction between causal and noncausal
epistemics is lexicalized by c'est pourquoi and de ce fait. A special remark
regarding the di€erence between de ce fait and c'est pourquoi is in order
here. In the introduction we saw that these connectives were dicult to
distinguish on the basis of intuitions, though we felt that de ce fait was the
least subjective of the two. Table 3 shows that they occur in the same subset
of relations, ranging from nonvolitional causality to causal epistemics.
However, their distribution di€ers signi®cantly in the expected direction:
de ce fait is predominantly used to express nonvolitional causality while
c'est pourquoi only rarely expresses the latter.
So far, we have only discussed relational interpretations of fragments
containing a certain connective. These interpretations are the result of an
interaction between the connective and the connected discourse segments.
Here, what we are interested in is the contribution of the connective to this
interaction. In section 3, we characterized this contribution in terms of
``a certain speaker-involvement level that is added to the interpretation
of its discourse environment''. How do we determine the speakerinvolvement level encoded by a connective? We could postulate that
the relational interpretations of the fragments containing a connective
also constitute the area on the speaker-involvement scale covered by the
connective. However, this line of reasoning only leads to the unsatisfactory
conclusion that the speaker-involvement areas of the di€erent connectives show considerable overlap. For instance, the distinction between
volitional causality and epistemic causality is not lexicalized, neither in
Dutch nor in French. Of course, the distributions show ``peaks'' at
di€erent points. But how do we get from these di€erent distribution
patterns to the semantic pro®le of the connectives?
We could, of course, make claims based on our own intuitions regarding
the meanings of the connectives or based on intuitions expressed in the
linguistic literature. However, we prefer a somewhat more cautious
approach, which combines distributional data and semantic intuitions. Let
us ®rst consider dus, donc, and deÁs lors. These connectives may occur in
speech-act interpretations. Hence, we may assume that these connectives
encode a maximum level of speaker involvement, i.e., the level at which
the present speaker and the present hearer are deemed to accept the
assumptions underlying the causal relation. This is because speech-act
relations cannot be realized with any lower level of speaker involvement.
The question is why these connectives also occur in volitional contexts.
In answering this question we need to realize that the interpretation of
234 H. Pander Maat and L. Degand
a volitional relation may vary in its speaker-involvement level without
changing the volitional nature of the relation. The minimum speakerinvolvement level required has been discussed in section 2.2Ðthe
volitional relation is based on a principle acceptable to a speci®c
protagonist. However, it is perfectly possible to present an action as
acceptable for actors in general, including the current speaker and hearer.
Our claim is that this is exactly what is being done by using connectives of a
maximal speaker-involvement level in volitional environments. This can
be shown by substituting connectives with a lower speaker-involvement
level for such connectives in fragments describing actions that cannot be
taken as acceptable for both the speaker and the hearer.
Imagine the following utterance addressed to a trac controller who
pulls you over because you ignored a ``do not enter'' sign. (this example is
inspired by Hybertie 1996)
(36)
a. Ik had haast, ?dus/daarom/?daardoor hield ik me niet aan het
inrijverbod.
b. J'eÂtais presseÂ, ?donc/c'est pourquoi/de ce fait j'ai pris le passage
interdit
c. I was in a hurry, so/that's why/as a result I ignored the no
entry sign.
Because dus and donc carry an assumption of general acceptability, they
seem inappropriate when the utterance is meant as an excuse; by contrast,
daarom and c'est pourquoi suggest no more than that to the speaker the
action seemed the best thing to do at the time. De ce fait further reduces the
responsibility of the speaker; daardoor, however, sounds rather silly since it
suggests that the speaker did not act intentionally at all.
On the other hand, when the fragment describes an action following
a generally acceptable rule, high speaker-involvement connectives are
perfectly possible (examples [37a] and [37b]).
(37)
a. Ik had haast, dus/daarom/?daardoor nam ik een taxi.
b. J'eÂtais presseÂ, donc/c'est pourquoi/de ce fait j'ai pris un taxi.
c. I was in a hurry, so/that's why/as a result I took a taxi.
It is not only volitional relations that allow di€erent degrees of Speakerinvolvement, epistemic relations do so too. Variation in epistemic relations
concerns the degree to which the hearer is supposed to accept the
relationship between argument and conclusion. In this sense, epistemic
relations may be weakly or strongly constraining with respect to hearer
acceptability. In the ®rst case, the hearer is addressed indirectly as a
potential member of the community sharing the assumption that the
argument increases the acceptability of the claim. This is the minimum
Scaling causal relations and connectives 235
level needed for epistemic relations, which implies that a certain reason
may be sucient for the speaker, but not necessarily for the hearer. This
level of speaker involvement seems to be expressed by daarom in Dutch,
and by c'est pourquoi and de ce fait in French.
If one wants to introduce stronger assumptions, i.e., force the hearer
to accept the protagonist's/speaker's reasoning, then dus, or respectively
donc and deÁs lors, should be used. If this suggestion is correct, substituting
donc/dus for c'est pourquoi/daarom in epistemic relations should increase
the speaker-involvement level. Compare examples (38a) and (38b).
(38)
a. Ik vind hem helemaal niet geschikt voor de functie. Daarom/dus
zou hij ontslagen moeten worden.
b. Je trouve qu'il ne convient pas du tout pour la fonction. C'est
pourquoi/donc il faudrait le renvoyer.
c. I think he does not ®t the job. So/`that's why' he should be ®red.
With c'est pourquoi and daarom, (38) re¯ects the opinion of the
speaker which one is invited but not compelled to follow.7 With donc
and dus, however, the speaker assumes that the hearer agrees with his
conclusion (the argument is decisive and hence there is no alternative to
®ring him).
The variability in speaker-involvement level does not apply to the
interpretation of nonvolitional and speech-act relations, that is to the two
extremes on the scale. Nonvolitional relations turn into epistemic relations
when high speaker-involvement connectives are inserted, as was already
shown in the example (4'), where this change resulted in inappropriateness.
At the other end of the scale, speech-act relations only exist by virtue of
a high speaker-involvement level, at which speci®c assumptions regarding
the speaker and hearer are made.
In the central area of the scale, however, we have shown that a
connective yields an increase in the degree of speaker involvement in the
relational interpretation when its position on the speaker-involvement
scale is higher than the minimum speaker-involvement level of the interpretation required by the environment; that is, the environment constitutes
the lower bound on the speaker-involvement level interpretations for the
fragment. Likewise, the semantic e€ects of substitution of one connective
for another (if the substitution is acceptable at all) can be explained by
reference to the positions of the connectives on the speaker-involvement
scale. Thus, the relative independence of connective meanings from
categories of relational interpretations yields expressive possibilities for
speakers who want to introduce assumptions to, or remove them from, the
interpretation of a certain relation.
236 H. Pander Maat and L. Degand
4.2. Subjectivity features as pointers to additional speaker-involvement
di€erences
4.2.1. Donc and deÁs lors
In French, all connectives except donc and deÁs lors di€er signi®cantly from
each other in terms of relations, though donc intuitively appears to be more
subjective. If this intuition is correct, we should be able to ®nd di€erences
in the degree of subjectivity of utterances containing the two connectives.
Hence, we will analyze the second segment for three speaker-involvement
features derived from two of the characteristics of speaker involvement
introduced in section 1: the identity of the protagonist, and the explicit
or implicit realization of the protagonist. With regard to the second
characteristic, two features will be investigated: the presence of subjective
quali®cations in the second segment, and the explicitness of the modal
protagonist in this segment.
First, we determine the identity of the causal protagonist in the ®rst
segment (S1). To this end, we ®rst select the fragments which actually have
a causal protagonist and which show some variation with respect to the
identity of this protagonist. Hence we leave out the factual segments,
which lack a protagonist, and we also exclude speech-act fragments, for
which the identity of causal protagonist is always the speaker. This leaves
us with volitional and epistemic relations. Possible participants include
the author, a group including the author and others, a cited speaker, a
generic third person, a pronominal third person, and a nominal third
person. The principal distinction in this domain is the one between ®rstperson participants and third-person participants (we failed to ®nd
second-person protagonists in our newspaper corpora). The hypothesis
is that a connective with a higher speaker-involvement level occurs more
often with ®rst-person participants, since the actions and conclusions of
speakers will generally be formulated with a higher speaker-involvement
level than the actions and conclusions of third persons. After all, the
speaker is more likely to accept the general assumptions underlying his
own decisions than those underlying the decisions of others.
The second step is to determine how this responsible participant is
realized in the consequence segment (S2), i.e., whether or not he or she
is linguistically expressed, and if so, whether this protagonist is explicitly
(I think/he thinks) or implicitly present in the segment, that is, invoked
as a reference point (e.g., probably). In nonvolitional relations there is
no conscious causal participant. In volitional relations, this participant
is nearly always explicitly mentioned (with the possible exception of
some passive and impersonal constructions). Hence, the expression of the
protagonist role only varies in epistemic relations. For these relations, we
Scaling causal relations and connectives 237
®rst compare the number of factually presented consequence segments
(no participant involved) with the number of subjective consequence
segments. Our hypothesis is that connectives with a high speakerinvolvement level more often co-occur with factual second segments,
since they are capable of introducing subjectivity into the interpretation
of the relation all by themselves.8
Within these subjectively formulated second segments we distinguish
implicit and explicit cases. Of course, this distinction is not entirely
independent of the distinction between ®rst and third-person protagonists,
since in general ®rst persons are implicitly and third persons explicitly
realized. However, since ®rst persons are occasionally explicit and third
persons occasionally implicit, it makes sense to formulate a separate
hypothesis regarding the implicit and explicit realization of the participant
responsible for the epistemic causal relation. Since we have de®ned speaker
involvement as the degree of implicit involvement of the present speaker
in the relation, our hypothesis must be that a more subjective connective
more often occurs in ®rst segments with implicit participants. When no
participant is involved in the segment (factual conclusions in epistemic
relations), the speaker-involvement hypothesis predicts a higher occurrence of low-level speaker-involvement connectives.
The results regarding these three hypotheses on the di€erence between
donc and deÁs lors are presented in Tables 4, 5, and 6.
Table 4. The identity of the participant responsible for the relation in fragments with donc
and deÁs lorsa
donc
First person
Third person
36
10
Total
46
(%)
deÁs lors
78.3
21.7
28
19
100
47
(%)
59.6
40.4
100
Total
64
29
93
(%)
68.8
31.2
100
a
Nonvolitional and speech-act relations are excluded. w2~3.78, df~1, p~.026,
one-tailed test.
Table 5. The formulation of the second segment in epistemic relations with donc and
deÁs lorsa
donc
Factual
Subjective
14
24
Total
38
a 2
(%)
36.8
63.2
100
deÁs lors
6
33
39
w ~4.61, df~1, p~.016, one-tailed test.
(%)
15.4
84.6
100
Total
20
57
77
(%)
26.0
74.0
100
238 H. Pander Maat and L. Degand
Table 6. The realization of the participant responsible for the epistemic relation in fragments
with donc and deÁs lors with subjectively formulated second segmentsa
donc
Implicit
Explicit
21
3
Total
24
(%)
87.5
12.5
100
deÁs lors
18
15
33
(%)
54.5
45.5
100
Total
39
18
57
(%)
68.4
31.6
100
a 2
w ~6.98, df~1, p~.004, one-tailed test.
The hypothesis that donc encodes a higher speaker-involvement level
than deÁs lors is supported by all three subjectively features: donc more
often accompanies segments with ®rst-person protagonists (see Table 4),
more often accompanies factual segments (see Table 5), and more often
has an implicit participant (always the speaker) in the second segment
(Table 6).
The di€erence between donc and deÁs lors regarding the modality of the
second segment is especially striking in noncausal epistemic fragments.
Here, donc regularly accompanies factually formulated conclusions, which
do not occur with deÁs lors. That is, a fragment like (39) was not found
for deÁs lors.
(39)
L'un d'entre eux portait un anorak vert. Les bandits n'eÂtait donc pas,
comme on l'avait cru un instant, deÂguiseÂs en agents de la Stib.
One of them wore a green anorak. So the bandits were not, as was
thought for a moment disguised as agents from the Stib (a public
transport agency from Brussels).
In our view, this is due to the di€erence in meaning between donc and
deÁs lors. If our reasoning is correct, substituting donc for deÁs lors in
fragments in which a third person, not the speaker, is responsible for the
second segment, should introduce a sense of increased speaker involvement. As a matter of fact, in fragment (40) the substitution of donc leads
to a more prominent role for the speaker at the expense of the explicit
third person (the Antwerp authorities).
(40)
(Car) pour acceÂder aÁ la mer, le port d'Anvers deÂpend bien eÂvidemment
du ¯euve. Les autoriteÂs anversoises se reÂjouissent deÁs lors/donc
grandement du deÂmarrage d'importants travaux de draguage pour
l'approfondissement de l'Escaut.
To have access to the sea, the Antwerp port depends on the river. So
the Antwerp authorities are very content with the beginning of the
dredging works of the Scheldt. [With donc, the second segment is not
Scaling causal relations and connectives 239
read to be a view of the Antwerp authorities but rather as a
conclusion of the speaker regarding this view.]
4.2.2. Dus and daarom
Interestingly, Dutch dus and daarom show di€erences parallel to the
ones between donc and deÁs lors. Of course we need to investigate those
di€erences while keeping the relational interpretation constant. In
Tables 7 and 8 we report on the identity of the responsible participant
for dus and daarom with volitional and epistemic relations, respectively.9
Since daarom occurs only sporadically in noncausal epistemic fragments,
we compared dus and daarom for causal epistemic relations alone. As
Tables 7 and 8 show, dus more often accompanies ®rst-person participants
in both relational contexts. Tables 9 and 10 further show that in epistemic
contexts dus less often accompanies subjective segments, and, within
these segments, those with implicit participants. The di€erence between
dus and daarom in terms of speaker involvement appears to be larger
than that between donc and deÁs lors. This is no surprise, since it was
already manifest in the area of relational interpretations covered by both
connectives (see Table 2), which was not the case of donc and deÁs lors.
We may conclude that a more detailed analysis of subjectivity features in
the utterances containing connectives not only reveals more ®ne-grained
Table 7. The identity of the participant responsible for the relation in volitional fragments
with dus and daaroma
dus
(%)
daarom
(%)
Total
(%)
First person
Third person
14
11
56
44
5
20
20
80
19
31
38
62
Total
25
100
25
100
50
100
a 2
w ~6.88, df~1, p~.005, one-tailed test.
Table 8. The identity of the participant responsible for the relation in causal epistemic
fragments with dus and daaroma
dus
(%)
First person
Third person
24
1
96
4
17
8
68
32
41
9
82
18
Total
25
100
25
100
50
100
a 2
w ~4.10, df~1, p~.022, one-tailed test.
daarom
(%)
Total
(%)
240 H. Pander Maat and L. Degand
Table 9. The formulation of the second segment in causal epistemic relations with dus and
daaroma
dus
(%)
daarom
(%)
Total
Factual
Subjective
8
17
Total
25
(%)
32
68
±
25
±
100
8
42
16
84
100
25
100
50
100
a 2
w ~9.52, df~1, p~.001, one-tailed test.
Table 10. The realization of the participant responsible for the relation in causal epistemic
fragments with dus and daaroma
dus
Implicit
Explicit
16
1
Total
17
(%)
94.1
5.9
100
daarom
(%)
Total
15
10
60
40
31
11
25
100
42
(%)
73.8
26.2
100
a 2
w ~6.09, df~1, p~.007, one-tailed test.
di€erences in speaker-involvement level between connectives (as was the
case for donc and deÁs lors) but also provides further support for speakerinvolvement pro®les for connectives that were already established on the
basis of relational environments (as in the case of dus and daarom).
Apparently, relational environments are only one of the manifestations of
the speaker-involvement pro®les characterizing the connectives.
5. Conclusions
Our point of departure in this article was that the dichotomies and
trichotomies concerning domains of coherence relations o€er no explanation for the distribution of causal connectives in several languages. Hence,
we have tried to develop an alternative approach to the semantic analysis
of these connectives, one which is of a scalar nature.
First, a number of relational distinctions found in the literature (nonvolitional vs. volitional, content vs. epistemic vs. speech-act domain) were
reanalysed in terms of a scale of increasing speaker involvement in the
construction of the causal relation, from non-volitional to volitional to
causal epistemic to noncausal epistemic to speech-act relations.
Subsequently, a number of frequent causal connectives in Dutch and
French were investigated in terms of speaker involvement. Initially, we
Scaling causal relations and connectives 241
examined the distribution of these connectives in terms of the ®ve
relational interpretations. These distributions were taken as indications of
the speaker-involvement levels encoded by the di€erent connectives.
The speaker-involvement pro®les thus developed were further supported by observations regarding the semantic e€ects of substituting a
connective with a di€erent speaker-involvement level. It was found that
the interaction between the interpretation allowed for by the relational
segments on the one hand and the contribution of the connective on the
other can be elegantly accounted for in terms of varying speakerinvolvement levels. The relative independence pertaining between connective meanings and categories of relational interpretations yields
expressive possibilities for speakers who want to introduce assumptions
to, or remove them from, the interpretation of a certain relation.
Finally, more ®ne-grained di€erences in speaker involvement were
revealed by analyzing the utterance containing the connective for linguistic features having to do with subjectivity. For instance, in this kind
of analysis it was found that high speaker-involvement connectives in
epistemic contexts more often accompany factual segments and, over all
relations, more often accompany segments with a ®rst-person protagonist
than other connectives.
In concluding this article, we want to point out that the speakerinvolvement approach may also have another advantage: it points to the
commonality between connectives and other linguistic phenomena that
have been treated in terms of concepts similar to speaker involvement. For
instance, it has long been recognized that deictic reference is essentially
relational in nature, in that it does not primarily characterize the referent,
but rather the relation between the referent and the contextual framework
of the present utterance (Hanks 1992). The relational dimensions at issue
include the distance between the referent and the speaker, both spatially
(e.g., this vs. that, here vs. there) and temporally (now vs. then), and the
degree of active involvement in utterance event (I vs. you vs. he/she). If
the causal relation can be considered as a kind of referent, the di€erence between dus and daarom is conceptually parallel to the one between,
say, I and he/she.
Another phenomenon that has been treated in terms of speaker
involvement is modality (Lyons 1977: 797±809; Coates 1983: 32±37;
Nuyts 1992). In a corpus study comparing Dutch modal adverbs like
waarschijnlijk `probably' with constructions involving the corresponding modal adjectives (in English, probable), Nuyts (1994: 45±52) found
that adverbs can only be used to express modal quali®cations ``performatively'', that is, as expressing an epistemic commitment of the
present speaker, whilst adjectives may be used both performatively and
242 H. Pander Maat and L. Degand
``descriptively'', that is, as reporting objective probability assessments or
epistemic evaluations by other people. From a diachronic perspective,
Langacker (1990: 25±29) has noted that the meaning of English modal
verbs (e.g., will, must) has moved toward increased ``subjectivity'', a notion
related to our concept of speaker involvement.
In other words, notions similar to speaker involvement are not only
lexicalized in the ®eld of connectives but also in other semantic domains
like deixis and modality. In all three ®elds, speakers have to express their
position vis-aÁ-vis, or their involvement with, the entities and propositions
presented in their discourse and the relations holding between them.
Language seems to o€er a choice of devices that is sensitive to this
communicative need.
Received 1 November 2000
Revision received 15 April 2001
University of Utrecht
University of Louvain-la-Neuve
Appendix
Legend to analysis summaries (in order of ®rst appearance)
p t1
qt2
pt1±t2
t1
t1¦t2
P, Q
?
^
B(a, p)
Bt1
Bt1±t2
PF(a, q)
ACC(a, p)
ps
pe
q
ts
a
s
c
n2o
DB
state of a€airs (state, event) valid at t1
state of a€airs (state, event) valid at t2
state of a€airs valid somewhere between t1 and t2
time point or interval
t2 may not precede or start earlier than t1
generalizations of p and q
causal relation
possibility operator
a believes that p is the case
belief entertained at t1
belief entertained somewhere between t1 and t2
a performs action q
a accepts p
statement concerning a state of a€airs
evaluative statement
intentional action
time of speaking
actor
speaker
concluder
assumption 2
an increase in the degree of belief in some proposition
Scaling causal relations and connectives 243
(s, h, saq)
h
I(s, h, sap)
speech act with propositional content q, performed by s with
regard to h
hearer
a certain interpretation of sap, by the speaker and the hearer
Notes
* Liesbeth Degand is a postdoctoral research associate at the Belgian Fund for Scienti®c
Research (F.N.R.S.). She is also supported by a grant from the Belgian Federal
Government (I.U.A.P. 4/19). Author's addresses: [email protected];
Ldegand@cassandre.¯tr.ucl.ac.be.
1. An anonymous reviewer has noted that even de ce fait can be used in epistemic contexts
when the reasoning follows the direction of real-world causality and when the modal verb
devoir is used:
(i) La tempeÂrature est nettement au-dessus de zeÂro. De ce fait la neige doit eÃtre en train de
fondre de la montagne.
2.
3.
4.
5.
6.
7.
8.
However, the intuitions of our informants are mixed here. But even, if we accept this use
of de ce fait, the direction of the epistemic relation and the modal verb are evidence that
de ce fait encodes a lower degree of subjectivity, as we will see in the following.
A defeasible implication between p and q can informally be described as ``if p then
normally q''.
Instead of the defeasible implication, indicated by w, we will use another concept of
causality, namely the INUS condition proposed by Mackie (1965), indicated by ?.
For a complete list of the symbols used in theses summaries, please see the appendix.
Later on, we show that cases such as (25) are infrequent.
These corpora were the electronic versions of De Volkskrant, a Dutch national daily
newspaper, and Le Soir, a Belgian francophone daily newspaper. The two corpora are
available on CD-ROM.
The nonvolitional paraphrase was this has/had the following consequence, the volitional
paraphrase was this is/was the reason to perform the following act, and the epistemic
paraphrase was this leads to the following conclusion. For speech-act relations, two kinds
of paraphrases were used: this can be paraphrased/summarized as follows; and this is the
reason to carry out the following speech act. Finally, noncausal epistemic relations were
distinguished from causally based epistemic relations by checking whether substituting a
``real-world'' causal paraphrase (nonvolitional or volitional) resulted in a coherent
sequence (even though the meanings di€er from the original fragment). If yes, the relation
was causally based, if not it was noncausal.
Of course, this does not rule out the possibility of combining these connectives with a
stronger statement in the second utterance, for instance C'est pourquoi la seule solution est
de le renvoyer, as one of our reviewers has noted. However, the point is that the strength
of the conclusion here cannot be provided by c'est pourquoi on its own, whilst donc,
automatically encodes a strong commitment to the conclusion.
At this point it might be asked why a high speaker-involvement discourse context still
needs a high speaker-involvement connective. This is primarily because high speakerinvolvement contexts run the risk of being relationally unclear. For instance, example (i)
may be read as expressing both a nonvolitional and an epistemic relation, and (ii) may be
read both as a forward noncausal epistemic relation and as a backward nonvolitional
(consequence±cause) relation.
244 H. Pander Maat and L. Degand
(i) The temperature is above zero. The snow is melting.
(ii) The snow is melting. The temperature is above zero.
A second question that might arise is whether marking a relation via a connective
decreases its degree of speaker involvement by making the relations more explicit. In one
sense this is true: a high speaker-involvement connective explicitly indicates speaker
responsibility for the relation. However, implicitness is not the essential characteristic of
speaker involvement. What is more important is that the speaker assumptions contained
in the relational interpretation do not change as a result of inserting the connective; in fact
they are presented even more prominently as the speaker's commitments.
9. The date in Table 7 were earlier reported in Pander Maat and Sanders (2000).
References
Coates, Jennifer
1983
The Semantics of Modal Auxiliaries. Beckenham: Croom Helm.
Fauconnier, Gilles
1994
Mental Spaces. Cambridge University Press.
Halliday, Michael A. K. and RuqaiyÂa Hasan
1976
Cohesion in English. London: Longman.
Hanks, William F.
1992
The indexical ground of deictic reference. In Duranti, Alessandro and Charles
Goodwin (eds.), Rethinking Context: Language as an Interactive Phenomenon.
Cambridge, Cambridge University Press, 43±76.
Hybertie, Charlotte
1996
La ConseÂquence en Francais. Paris: Ophrys.
Itkonen, Esa
1983
Causality in Linguistic Theory. London/Canberra: Croom Helm.
Knott, Alistair
2001
Semantic and pragmatic relations and their intended e€ects. In Sanders, Ted,
Joost Schilperoord, and Wilbert Spooren (eds.), Text Representation:
Linguistic and Psycholinguistic Aspects. Amsterdam: Benjamins, 127±152.
Lagerwerf, L.
1998
Causal Connectives Have Prepositions, E€ects on Coherence and Discourse
Structure. The Hague: Holland Academic Graphics.
Langacker, Ronald W.
1990
Subjecti®cation. Cognitive Linguistics 1, 5±38.
Lyons, John
1977
Semantics. vol. 2. Cambridge: Cambridge University Press.
Mackie J.
1965
Causes and conditions. American Philosophical Quarterly 2, 245±255,
261±264.
Mann, William C. and Sandra A. Thompson.
1986
Relational propositions in discourse. Discourse Processes 9, 57±90.
1988
Rhetorical Structure Theory: toward a functional theory of text organization.
Text 8, 243±281.
Martin, John R.
1992
English Texts: System and Structure. Amsterdam: John Benjamins.
Scaling causal relations and connectives 245
Moeschler, Jacques
1989
Pragmatic connectives, argumentative coherence and relevance.
Argumentation 3, 321±339.
Nuyts, Jan
1992
Subjective vs. objective modality: What's the di€erence? In Fortescue,
Michael D., Peter Harder, and Lars Kristo€ersen, Layered Structure and
Reference in a Functional Perspective. Amsterdam: Benjamins, 73±99.
1994
Epistemic modal quali®cations: On their linguistic and conceptual structure.
Antwerp Papers in Linguistics 81. Antwerp: Universiteit Antwerpen.
Oversteegen, E.
1997
On the pragmatic nature of causal and contrastive connectives. Discourse
Processes 24, 51±85.
Pander Maat, Henk and Ted Sanders
2000
Domains of use or subjectivity: The distribution of three Dutch causal
connectives explained. In Couper-Kuhlen, Elisabeth and Bernd Kortmann,
Cause, Condition, Concession, and Contrast: Cognitive and Discourse
Perspectives. Berlin/New York: Mouton de Gruyter, 57±82.
Pit, Mirna
1997
Subjective and objective construal in coherence relations. In Don, J. and
T. Sanders (eds.), Yearbook 1997. Utrecht: Utrecht Institute of Linguistics
OTS, 45±70.
Sanders, Ted
1997
Semantic and pragmatic sources of coherence. On the categorization of
coherence relations in context. Discourse Processes 24, 119±147.
Sanders, Ted J. M., Wilbert P. M. Spooren, and Leo G. M. Noordman
1992
Toward a taxonomy of coherence relations. Discourse Processes 15, 1±35.
Stukker, Ninke, Ted Sanders, and Arie Verhagen
1999
Waar een wil is, is geen wet. De categorisering van causale relaties binnen en
tussen zinnen. [When there is a volitionality, there is no law. Categorizing
causal relations within and between sentences]. Gramma/TTT 7, 66 ±88.
Sweetser, Eve E.
1990
From Etymology to Pragmatics: Metaphorical and Cultural Aspects of
Semantic Structure. Cambridge: Cambridge University Press.
van Dijk, Teun A.
1979
Pragmatic connectives. Journal of Pragmatics 3, 447± 456.