In April
1989 Eugene Loos and Ivan Lowe conducted a linguistic workshop designed to
tackle a number of important questions in the area of logical connectives in
discourse, such as the following:
"What are
the essential typological syntactic and semantic features that must/should
characterize each connective in any language for a relationship between
propositions to be acceptably defined?
“What are the
pragmatic characteristics of a designated connective?”
“What kinds of
data should field workers seek in order to present convincing evidence that
what they claim to be a particular logical connective, is in fact that?"[1]
In the
first two weeks of the seminar, the participants were introduced to the Rhetorical Structure Theory developed by
Mann and Thompson (1986,1987), and also - in one session - to relevant aspects
of the communication theory currently being developed by Kathleen Callow (1998)
in her book Man and Message.
Rhetorical
Structure Theory looks at texts primarily from the analyst's point of view, and
is intended to provide for him "a linguistically useful method for
describing natural text" (Mann and Thompson 1987, p. 1). It does not make
explicit claims as to whether this analysis reflects what goes on in people's
mind when they communicate.
Kathleen
Callow's presentation differed from this in that she was interested in what
goes on in the speaker's mind rather than in possible ways of describing a
text. Consequently the relationships her theory proposes to hold between the
different parts of a text are meant to reflect a psychological reality in the
speaker's mind.
I
believe that Callow's position is preferable because by changing the question
from "What is the best (most appropriate, most elegant) way of analysing
the relationships in a text?" to "How does the communicator relate
the propositions in his mind?", the issue becomes a psychological, hence
an empirical one, rather than one of finding the "best" description
for a set of phenomena - in whatever terms that is to be evaluated, such as
elegance, completeness, simplicity and so forth.
What
both Rhetorical Structure Theory and Callow's approach have in common is the
reliance on a set of relations, thought to hold either between text parts or
between propositions and chunks of propositions. This brings us to the central
topic of this seminar: how do you know in any given case that there is such an interpropositional or
rhetorical relation, and that it is of type X rather than Y? The current
understanding is that both approaches need to define sets of relationships that
are wide enough to cover all possible
relations that speakers can use in texts. At the same time, the relations need
to be defined in sufficiently explicit
terms to differentiate them clearly from one another.
The fact
that even with the more explicit definitions given in Mann and Thompson (1987)
it often proved difficult in the workshop to determine why a particular example
was analysed as exhibiting relation X rather than Y seems to suggest that the
task is not an easy one. This difficulty is further evidenced by the existence
of a number of different schemes of relations that have been proposed in the
past, such as in Ballard et al (1971), Beekman and Callow (1974) and its
further developments reflected in the "Semantic Structure Analysis"
approach (e.g. Callow (1982®LBSSA of 2nd Thessalonians¯)), Pike and Pike
(1977), Longacre (1983) and others.
It is at
this point, I believe, that the relevance theory of communication developed by
Sperber and Wilson (1986) can be of considerable value because it offers the
possibility of solving the problem by going one step further: there is good
evidence that relevance theory can account for these various relationships without necessarily having to refer to them
as theoretical constructs in their own right. In other words, relevance
theory does not need for defining a set of such relationships in order to
account for our intuitions about them. The main point of this paper is to demonstrate
this possibility with examples taken from the Silt'i language, an Ethio-Semitic
language spoken in Ethiopia. As we shall see, this approach also allows a
clearer understanding of how logical connectives work.[2]
Furthermore, we shall find that relevance theory helps us to gain a much richer
understanding of the speaker-intended interpretation than relation-based
approaches.
However,
before we can turn to examples, we will need to look at some basic notions of
relevance theory.
In this
section I can only give the briefest sketch of relevance theory; a few more
notions will be introduced as we go along. For more detailed information the
reader is referred to Sperber and Wilson (1986).[3]
2.1.
Ostensive-inferential
communication and the principle of relevance
The
starting point of relevance theory is a familiar one - it is the question: how
do people manage to communicate? How is it possible for me as communicator to
induce you to entertain thoughts that I have?
At a
very general level, the answer is that a communicator produces a stimulus that
makes evident a) that he intends to convey some thoughts, and b) what those
thoughts are. In other words, the communicator behaves in a way that provides evidence of his intention to communicate
certain thoughts, and it is the audience's task to infer from the evidence provided what these thoughts are. This
means that communication is viewed essentially as an inferential process. Communication succeeds when the audience's
inferences lead it to the thoughts which the communicator intended to convey.
For
example, suppose you are at a party with a friend. You notice that it is time
to go home. So when your friend is looking at you, you catch her eye and direct
her attention to the clock on the wall by conspicuously looking there yourself.
The conspicuous nature of behaviour will induce your friend to infer that you
want to communicate something, and she will try to infer from it what it is
that you want to convey. Thus she will notice that you are drawing her
attention to the clock, and will try to infer what you are intending to
communicate by that. One possible inference would be that you intend her to
take note of what time it is, which may lead to the further inference that you
might think it time to go home and so forth.
Note
that we are not normally conscious of this inference process - it works very
fast and is something that our mind takes care of subconsciously. We will
usually have some awareness of the results
of this process, but the process itself goes on below the level of
consciousness. There are, of course, instances when our interpretation efforts
rise to consciousness, but that is usually only when there are problems.
Of
course, you could also have communicated by using some verbal stimulus, e.g. by
saying, "Look at the clock" or "Look what time it is".
Again your friend would then have to infer what thoughts you wanted to
communicate from your utterance. In either case, the basic process would be the
same: one infers from people's behaviour, both non-verbal and verbal, what they
intend to communicate.[4]
Communication
thus involves inference, and inference involves information processing, but
information processing is something we do
and hence costs some mental effort:
the audience needs to carry out inference, it needs to access information from
memory, i.e. carry out memory searches; and in the case of verbal communication
it also needs to carry out linguistic decoding, which is not for free either.
However,
life is not all work: there are also benefits,
and these are defined in terms of contextual
effects. Put simply, contextual effects can be thought of as gratifying
changes in one's knowledge. I shall say more about these changes in section
3.2.2 below, but crucially contextual effects are achieved not by the utterance
alone, but by the inferential combination of the utterance with the contextual
knowledge of the audience. This captures two basic intuitions: firstly, it
accounts for the fact that one does not usually consider it very rewarding to
be told something one was well aware of already - the information presented has
to make some difference. Secondly it captures the further intuition that in
communication it is not enough for the information to be new - it must rather link up with knowledge already present
in some way.[5]
The main
points so far,then, are these: communication is rewarding, it brings contextual
effects, but there is also a cost attached to it in that it requires mental
processing effort. This relation between contextual effects and the processing
effort needed to derive them is defined as relevance:
the relevance of an utterance will be the greater the more contextual effects
it has, and the less processing effort it requires. It will be the smaller the
fewer contextual effects it has and the more processing effort it requires to
achieve those effects.
And now
we are ready to come to the crucial point: relevance theory proposes that by
beginning an act of communication, a communicator automatically communicates
the following presumption: firstly, that there will be adequate contextual
effects, and secondly, that these effects will be derivable with minimal
processing effort on the audience's part. Put in less technical language, by
claiming the audience's attention, the communicator claims that what he intends
to convey will be rewarding, and that he has made it as easy as possible for
the audience to receive this reward. This presumption is assumed to follow from
our human nature, and relevance theory refers to it as the principle of relevance.[6]
This
principle is so important because it explains how communication overcomes
certain inherent problems. For example, as Kathleen Callow also pointed out in
her presentation, for successful communication it is necessary for the hearer
to have just the right context - communication cannot be successful unless the
hearer actually uses the contextual information envisaged by the communicator.
In relevance theory this follows from the fact that communication necessarily
requires the inferential combination of what is expressed with contextual
information: and if the audience does not use the contextual information
envisaged by the speaker, it is likely to come up with the wrong
interpretation. Therefore since it is in the speaker's interest that the
audience will get it right, he will naturally try to clue them in about the
right context, if he is in doubt that they can access it.
Thus
Kathleen Callow gave the example that if she wanted to talk to her husband
about an individual called 'Phil' who was not a very close friend of theirs,
she would probably introduce her actual remark by saying, 'You know Phil at
church? Now he ...'. She also noted that in the case of a very good friend this
would not be necessary.
In
relevance theory this behaviour is explained by the principle of relevance:
when hearing the name 'Phil', the hearer will have to find a referent in memory
to assign to this name. The speaker can, of course, leave that task completely
to the hearer without giving him further clues, and if the referent is highly
accessible in memory at the time of speaking, then this might well be the best
strategy. Thus information about people very close to us is probably highly
accessible at most times, and it would be wasting the audience's processing
effort to give further information about an individual that the audience would
have thought of right away anyhow.[7]
At other
times, however, the speaker will assist the audience in finding the right
context in order to ensure that they arrive at the intended interpretation. He
can do so not only by prefacing his utterance with preliminary remarks like the
one just considered, but in many other ways. As we shall see, logical and other
connectives are valuable devices for this purpose.
From the
perspective of the communicator this may be clear enough - but given that we
can never know any other person's thoughts directly, how can the audience ever
be confident that they have got things right, both with regard to the context
envisaged by the speaker and the interpretation as a whole? Even if they arrive
at some interpretation what reason do they have to believe that it is the
interpretation the speaker had in mind?
Thus if
you draw my attention to the clock on the wall - how can I tell whether you
intend to convey to me that it is a very expensive clock or that it is time to
go?
The
answer lies in the principle of relevance: there is never more than one
interpretation that is consistent with the principle of relevance. There can be
only one such interpretation, because
as soon as the audience has arrived at an interpretation that the communicator
could have expected to have adequate contextual effects without requiring
unnecessary processing effort, the interpretation process will stop. There may be
other possible interpretations that might have adequate contextual effects -
but they would no longer be optimally relevant because of the additional
processing effort they require. Of course there may be no interpretation that is consistent with the principle of
relevance; in this case one will need further clarification of what the
communicator meant or else it will remain unclear what he had in mind.
Therefore
the criterion of consistency with the principle of relevance entitles the
audience to assume that the first interpretation that they find to be
consistent with the principle of relevance is the one that the communicator
intended to communicate. Thus as soon as the audience has arrived at an
interpretation that seems to have adequate contextual effects for the amount of
processing effort invested, the audience can stop processing and assume that it
has arrived at the intended interpretation. It need not look any further.
With
regard to context selection this also means that the audience will use the
contextual information most highly accessible to them first before spending
more processing effort on going further afield. In other words, if someone
starts talking to you about 'Phil', and what is said about this individual has
adequate contextual effects on the assumption that it is your brother, then you
will simply assume that this is indeed the Phil he means. Occasionally it
happens that the communicator was actually thinking of another person; in that
case what he says about that individual will - sooner or later - cease to yield
adequate contextual effects with the information you have about that
individual. This lack of adequate contextual effects makes itself felt as 'not
making sense': things don't seem to link up in the right way. In this way
relevance theory offers an account of the notion of 'making sense' and of its
importance in communication.
2.2.
The role of
logical connectives
Since
the nature of communication is seen to involve the inferential combination of
what is said with certain contextual pieces of information, it is not difficult
to see that connectives that establish logical relationships can be of great
value in communication.[8]
Let us illustrate this point briefly with an English example, and let us take
the famous example from Grice (1975): “He is an Englishman; he is, therefore,
brave.” (p. 161)
This
example is discussed from a relevance theoretic point of view in Blakemore
(1987). She proposes that English 'therefore' is used to "introduce a
proposition that is derived as output to a synthetic inference rule"
(1987, p. 82). This means that it instructs the audience to treat the
proposition marked by 'therefore' as the conclusion to an argument in which a
synthetic inference rule has applied. What is a synthetic inference rule? It is
simply an inference rule that "... takes two separate assumptions as
input" (Sperber and Wilson 1986, p. 104).
This
means, the speaker expects the hearer to construe an argument of the following
kind that has at least two separate premises and yields the assumption 'he is
brave' as a conclusion.
(1)
Premise
1: ?
Premise
2: ?
-----------------
Conclusion: He is brave.
The task
of the hearer is to find highly accessible contextual assumptions that will
complete this argument.
One of
the most highly accessible contextual assumptions is, of course, the one just
processed: 'He is an Englishman'; so it would be an obvious candidate for
serving as one of the premises:
(2)
Premise
1: He is an Englishman.
Premise
2: ?
-----------------
Conclusion: He is brave.
But what
about premise 2? This is not difficult to supply either: a very simple
assumption that yields 'He is brave' as a valid conclusion is: All Englishmen
are brave.
So, the
completed argument would be:
(3)
Premise
1: He is an Englishman.
Premise
2: All Englishmen are brave.
-----------------
Conclusion: He is brave.
What is
the point of using 'therefore' here? By explicitly instructing the audience to
construe 'he is brave' as the conclusion to an argument in the context of the
proposition 'he is an Englishman', the speaker narrows down or constrains the various ways in which the utterance
could have been related to contextual assumptions; in other words, it imposes a
constraint on the relevance of the
utterance.[9]
Thus by using 'therefore', the speaker strongly suggests to the audience to
supply premise 2 'All Englishmen are brave' as a contextual assumption, because
there are hardly any other, highly accessible assumptions that could fill that
part of the argument. This means that the speaker is strongly implicating the
truth of premise 2, in addition to that of the premise 1 and the conclusion.
Such particular assumptions which the hearer is expected to supply in order to
arrive at the speaker-intended interpretation are called strong implicatures. (In
this case we are dealing with an 'implicated premise'.)
Consider
by contrast what would have happened if the speaker had omitted 'therefore',
saying simply: He is an Englishman; he is brave.
Here the
audience would have no clear guidance as to how these two propositions are to
be treated. It can, of course, treat them as premise and conclusion of an
argument; however, this is not a necessary interpretation of the evidence; if,
for example, it did not seem likely to the audience that the speaker actually
believes 'All Englishmen are brave', then the audience would be entitled to
look for a different interpretation. Thus it could simply be part of a listing
of characteristics, without any intended implication that the bravery is seen
as a consequence of being English. Perhaps that interpretation might still come
to the mind of the audience; however, given that the speaker has given no
indication that the utterance should be interpreted in this particular way, and
given that other contextual assumptions lead to adequate contextual effects,
the evidence that the communicator intended the audience to use this particular
assumption is very weak. Such assumptions are therefore referred to as weak implicatures.
This is
another important feature of relevance theory: it does not necessarily expect
that propositions are either communicated or not communicated, but it assumes
that they can be communicated with varying
degrees of strength: the more essential a particular assumption is to
establishing the optimal relevance of an utterance, the more strongly that
particular assumption is communicated or implicated. The less essential it is
to the interpretation process, the less strongly it is implicated. In other
words: implicatures can differ in degree of strength. Thus we can see that the
use of logical connectives is well-motivated for guiding the audience to the
speaker-intended interpretation: not only do they help the audience to access
the speaker-intended context, but they also enable the communicator to strongly
communicate particular contextual assumptions without having to spell them out.
3.1.
Preliminaries
As
mentioned above, the examples I want to look at are taken from the Silt'i language.
Some of the examples are dealt with in more detail in the paper "Toward an
analysis of pragmatic connectives in Silt'i" (Gutt 1988) presented at the
Eighth International Conference of Ethiopian Studies in 1985. That study was
exploratory rather than exhaustive, and I have not been able to do much more in
this area since then. So the results presented here should be taken as
preliminary rather than final. I think the analysis proposed is a reasonable
one, but I am more interested in demonstrating the approach than arguing
particular, language-specific points.
3.2.
The problem
In
Silt'i there is an affix, -m, whose
meaning seems to be rather hard to pin down. In many instances it seems to make
little difference to the meaning, and seems to be translated best by
"and". In other instances, its influence can be felt more strongly,
and seems to correspond to something like "either" or
"neither" in English. Sometimes it seems to convey the meaning of
"also" or "too." At other times again it seems much
stronger, amounting to something like "even", sometimes acquiring
almost a concessive sense, especially in conditional clauses.
Since
the purpose of this presentation is not to argue for a particular analysis, but
to make some basic points, I will not go into all these different uses here,
but will deal only with the first two. So for now I will just briefly
illustrate the first two uses - those that seem to correspond to English
"and" on the one hand, and something like "either" on the
other.
The
first passage is part of a fable where a mouse has just given birth to nine
little mice, and the cat comes to convey her good wishes on the delivery. The
cat opens the conversation.
(4)
Illustration 1
5.
"mayraam tashiillima
taawt'aash"
baateet.
Mary
you-being-well may-she-bring-you-out she-said-to-her.
6. ufrimte
"aaw, ashnam maayraam
taanbirash"
the-mouse-m yes, you-m Mary may-she-make-you-live
baateet. 7. adanimteenga abbi
luufrite
she-said-to-her.
7. the-cat-m-again then to-the-mouse
"bay, inde, yesh, innate ayb sici;
come-on,
o.k., here, this milk drink;
waldamii ihe liinzinsh" baateet.
the-children-m
I let-me-hold-for-you she-said-to-her.
Free
rendering:
5. "May
Mary make you well," she (=the cat) said to her. 6. And the mouse said,
"Yes, and may Mary make you live/prolong your life." 7. And then the
cat said again to the mouse, "Okay, come on, here drink this milk; and let
me hold the children for you." (SA, SITA03.TXT)
The
other sample passage is part of one of the various fables where the hyena
(thought of as male) tricks the donkey (thought of as female). In this
particular fable, the hyena challenges the donkey to a fight. Since the donkey
is much stronger she throws the hyena down twice. Each time the donkey gets off
the hyena, letting him get up. However, the hyena cunningly claims that he had
actually let himself drop to the ground voluntarily for the donkey. On these
grounds he demands that the donkey, too, should let herself drop to the ground
for the hyena. The stupid donkey does so - but the hyena doesn't let her get up
again and eats her instead.
The part
we are interested in is the final argument: the donkey is being held down by
the hyena, but argues that the hyena should get off since she herself got off
twice from the hyena. The hyena rejects this demand:
(5)
Illustration 2
"...
tee, jigg bay; aabbeemee
bagadala
you(f) quiet say(f); my-father-m
when-he-threw-down
alnak'aan" baalaane inkitkit
ashaane baleet
he-did-not-get-off
he-said-and tearing up he-ate-her
ileen.
one-says.
".... Shut up, you! my father didn't get up either when he threw someone down!" he (=the hyena) said, tore her (=the donkey) to pieces and ate her, they say. (SB, SITA01.TXT)
3.3.3
Formulating a hypothesis
Just
dealing with these two uses now, the tempting solution would be to say either
that there are two homophonous suffixes -m
or that the suffix -m has two
distinct senses, one marking a conjunctive relationship, and the other
something like an alternative relationship. However, Ivan Lowe pointed out in
his introductory lectures that it is not the most helpful way to begin one's
analysis: by assuming a complex solution from the start one may miss a possible
simpler solution. In fact, my analysis started from the minimal assumption that
there is only one -m which has only one meaning, and this meaning corresponds to the inferential
properties of "^" (logical "and").[10]
In the interpretation process it amounts to an instruction to the audience to
conjoin the proposition marked by -m
with another proposition.[11]
In the
course of the analysis I came to postulate in addition the following syntactic
properties: -m is suffixed to the
leftmost subconstituent of the focused constituent of a sentence or to the
sentence as a whole.[12]
3.2.2
Testing the hypothesis in context
One way
to test such a hypothesis within the framework of relevance theory is to check
whether the hypothesis allows for an interpretation consistent with the
principle of relevance for passages containing -m.
The use
illustrated by illustration 1 is not all that exciting: it seems that the -m in all cases is syntactically
associated with the sentence as a whole, and adds the proposition expressed to
the information processed previously. In logic, the conjunction of two
propositions in logic by "and" entitles one to infer that both of the
conjuncts are true. This may not seem very significant at this point, but it
does have pragmatic value, as we shall see below.
Illustration
2 is more interesting. It seems that here -m
is associated with a focused constituent, i.e. abbee 'my father'. It would go too far to present the
relevance-theoretical account of focus here in detail. What is important here
is that the distinction between focus and background makes a difference as to
how the respective pieces of information contribute to relevance: the focused
constituent pinpoints that piece of information that makes for relevance, i.e.
that is intended to have contextual effects. The background, by contrast, has
no contextual effects of its own but contributes to relevance by making more
accessible contextual assumptions needed for interpreting the utterance.
Technically,
a background entailment of an utterance with focus is obtained by substituting
the focused constituent by a semantic variable. Now the sentence we want to
concentrate on in our example is this:
(6)
aabbeemee
bagadala
alnak'aan
my-father-m when-he-threw-down (a victim)
he-did-not-get-off
If the
focused constituent is aabbeemee,
then the next background entailment is:
[ X ]
bagadala alnak'aan
X
when-throws-down (a victim)
X-not-get-off
where
"X" represents a semantic variable that can range over individuals,
perhaps best represented by English "someone". Note that the
background is not itself a complete proposition: firstly because of the
variable X, and secondly because time reference has not been assigned yet, so
that a better paraphrase in English might be: 'when someone throws (a victim)
down he (does not) get off'.[13]
Now by
definition the background entailment does not achieve relevance in its own
right. But why, then, is it there at all? As will be recalled, relevance
depends not only on the contextual effects achieved, but also on the amount of
processing effort spent. It is this second factor that is operational here: the
background entailment contributes to the overall relevance of the utterance by
making some of the contextual information more accessible, and hence by
reducing the processing effort.
And this
is how it works: though not a complete proposition in itself, the background
entailment does provide a schema or 'blue print' which the audience can
complete to fully propositional form with comparative ease.
As we
said above, essentially two pieces of information are missing from the
entailment: firstly, the variable must be replaced by a referent. One of the
most highly accessible referents is, of course, the hyena, so it becomes the
most obvious candidate. With regard to time reference, since the hyena has just
got the donkey on the ground at the time of speaking, the most accessible time
reference is that of the time of speaking. So the proposition can easily be completed
into:
When the hyena throws (a victim) down he
does not get off.
With the
conjunction in effect, we get the following partial interpretation:
(When
the hyena throws (a victim) down he does not get off) and (when the hyena's
father got (a victim) down he did not get off)
The fact
that aabbee is focused furthermore
means that it is the relevance-establishing part of the utterance, and so the
audience will concentrate on this concept for contextual effects. One important
feature of mental concepts in the relevance-theoretical framework is that they
are assumed to be associated with different 'entries' in memory, one of which
is the so-called 'encyclopaedic entry'. The encyclopaedic entry of a concept
contains all sorts of information a person has come to collect about the
referent to which it refers, and when a concept is being thought about
(mentally represented), the encyclopaedic entry is opened up and the
information contained there becomes available for use. Now it seems that one
assumption associated with the concept 'father' in Silt'i seems to be something
like 'Like father, like son', or more formally: 'If father does X, then son
does X.'
So now
the following set of propositions is available:
(7)
Utterance: When the hyena's father threw (a
victim) down
he did not get off
Contextual
assumption 1: and when the hyena throws
(a
victim) down he does not get off
Contextual
assumption 2: If the father did X, then
the son
does X.
As can
be seen, contextual assumption 2 and the proposition expressed in the utterance
constitute an inferential argument of the following form:
(8)
Premise
1: If father did X, then son does X.
Premise
2: The father did not get off when he threw down (a victim).
The son does not get off when he
throws down (a victim).
Now this
is all very well - but what reason would an audience who derived this
interpretation have to believe that it had found the right, i.e.
speaker-intended, interpretation?
The
answer is that it is entitled to assume this if this interpretation is the
first one that comes to mind and seems consistent with the principle of
relevance.
As far
as coming to mind first is concerned, the interpretation suggested seems to be
a good candidate in that it involves only assumptions that can be assumed to be
highly accessible to a Silt'i audience.
To be
consistent with the principle of relevance, the interpretation has to yield
adequate contextual effects. What contextual effects does this interpretation
achieve? To answer this question, we need to have a closer look at contextual
effects.
Contextual
effects can be of three kinds: they can consist in the addition of contextual
implications, they can consist in the cancellation of contextual assumptions,
and they can consist in the strengthening of contextual assumptions. Contextual
implications are defined as implications that are not derivable from the
utterance alone, nor from the context alone, but only from an inference that
uses premises from both. Put in less
technical terms, the contextual implications of an utterance lead to new
insights that are, however, related to what one already knew.
Cancellation
of contextual assumptions results when there is a contradiction between two
assumptions, then one of them may get erased from memory. Put simply, this
accounts for our intuition that it is relevant to discover where one has been
wrong.
The
effect of strengthening existing contextual assumptions has to do with the fact
that we do not just believe or not believe certain propositions to be true, but
we can hold them true with varying degrees of confidence, or strength. Thus we are fully convinced of
the truth of some assumptions, e.g. that there is a sun, but we may be less
confident about the truth of the statement that early man had a lower
intelligence than modern man. Therefore, one way of context modification lies
in the possibility of altering the strength with which assumptions are held to
be true.
Our
example here seems to involve both the cancellation of one contextual
assumption and the strengthening of another. When the donkey agreed to drop to
the ground for the hyena, it did so no doubt on the assumption that the hyena
would let her get up again. So at that point she entertained a thought, that is
a contextual assumption, like:
"The hyena will get off me."
However,
when the hyena gave no indication that he would get off the donkey, another
thought was likely to enter the donkey's mind:
"The hyena will not get off me."
Thus the
donkey would have come to entertain two mutually contradictory assumptions.
What happens in such cases? Relevance theory proposes that in this case our
mind compares the relative strength of the two contradictory assumptions, and
if one of them is weaker than the other, then it will be erased from the mind,
and so the contradiction will be resolved.
What,
then, about the comparative strength of the two assumptions involved in our example?
As the donkey's request indicates, her belief that the hyena would get up was
based, rather naively, on a notion of fairness: since the donkey got off, so
would or should the hyena.
The
assumption implicated by the hyena, however, seems much stronger because it is
not simply asserted, but presented as the conclusion of an argument the
premises of which seem to be quite strong: there would be little doubt that the
hyena's father never let a victim go of a good will, and the truth of the idea
'like father, like son' would certainly not be doubted either.[14]
So the likely result is the erasure of the contextual assumption: 'The hyena
will get off' - in other words the hyena's reply is designed to remove any
doubt in the donkey's mind about its intentions: it is certainly not going to
get off - the fate of the donkey is sealed.
At the
same time, contextual assumption that the hyena would not get off him -
probably not really fully believed by the donkey at first - would now be
considerably strengthened, again because of the argument structure and the
strength of the premises on which it is built.
Thus the
interpretation we arrived at would at least have the two contextual effects of
cancelling one contextual assumption and strengthening another. In this way it
would at least satisfy the minimum requirement for relevance.
So it
seems that we have arrived at a fairly plausible interpretation of this
utterance, assuming nothing more about -m
than that its meaning corresponds to that of logical 'and', and that
syntactically it can be associated with focus.[15]
3.3.3
Interpretation and rhetorical
relationships
As you
will have noticed, we have proposed an analysis of the speaker-intended
interpretation of this utterance without any reference to
"rhetorical" or "interpropositional" relationships. How
then does relevance theory relate to these notions? After all, we do have
intuitions about them. For example, the way we just analysed the story seems to
suggest a hereditary, almost causal, relationship: sons inherit their fathers'
nature - so the hyena's behaviour is causally conditioned by its natural
traits, it actually couldn't behave otherwise. In other words, the thrust of
the argument might be seen to be, "How can I help eating you up - I am
simply acting in accordance with the nature I inherited from my father."
However,
this might not be the only possible interpretation. Perhaps we might feel that
the hyena is giving some kind of moral justification: it is right to follow
one's father's example, therefore it is morally justifiable for the hyena to
eat the donkey.
Since we
can and do have such intuitions - where do they come from and how can we
account for them in the relevance theory framework?
The
answer seems to be that these intuitions simply reflect what particular
contextual assumptions we use in the interpretation process. Let us begin with
the intuition that there is a 'cause-effect relationship'. One of the
contextual assumptions we assumed to be involved here was the following
(contextual assumption 2 in (7):
If the father did X, then the son does X.
We first
assumed this assumption to simply reflect a popular belief, as expressed in our
proverb 'Like father, like son.' It is, however, possible that more is
involved; it is possible that people believe more specifically that heredity is
involved here, that is, that the son inherits the nature and behaviour of the
father:
If the father did X, then by the laws of
heredity the son will do X.
This
belief assumes that the father's nature is the natural cause of the son's
nature, and it is this assumption that gives rise to the impression that a
"cause-effect relationship" might be involved here.
What
about the impression of a justificatory relationship? This would have arisen
if, in addition perhaps to the assumption of natural cause, the following
assumption was also highly accessible:
If the father did X, then it is right for
the son to do X.
Together
with the other propositions already introduced, this would allow the conclusion
'It was right for the hyena not to get off from the donkey.' Which of these
various assumptions would be implicated and with what strength would be
determined by consistency with the principle of relevance against the context
envisaged by the communicator.
This
account is significant for our discussion in a number of respects. Firstly, it
brings out that the interpropositional 'relationships' that we intuitively felt
to be there are not semantic clues that help us to find the intended
interpretation; rather they are possible abstractions we can make about the
interpretation itself once we have found it. In other words, the
'interpropositional relationships' do not determine the meaning of the
utterance, but can be derived from it.
Secondly,
these 'relationships' were not signalled or encoded by the connective itself.
Rather, they follow from an interpretation that arose from the complex
interaction of the proposition expressed by the utterance, the constraint on
relevance imposed by the connective -m,
and certain contextual assumptions - all interacting under the criterion of
consistency with the principle of relevance.
This is
not to say that languages cannot ever use connectives to encode such specific
relationships. There is no reason why they should not. However, this is a language-specific,
lexical matter rather than one of concern for a general theory of either
semantics or communication.
And this
brings us to the third point: there is no need to assume that there is a closed
set of interpropositional relationships, either for one language or universally
for all languages, that allows us to adequately describe how the various units
of texts relate to each other. As we saw in our example, the link between the
propositions is achieved inferentially by specific contextual assumptions.
This, in fact, explains why relationship typologies find it so difficult to
provide adequate categories: there is no reason why contextual assumptions
should fall into such categories.
Fourthly,
relational typologies fall short of bringing out the intended interpretation;
they focus on general abstractions but do not reconstruct the interpretation
with its specific contextual assumptions. Consider the following typical
relationship-based analysis of the utterance 'The car is clean, so John must have
washed it':
(9)
Grounds:
The car is clean,
Conclusion: (so I conclude that) John must have washed it.
(Larson 1984, p. 306)
This
analysis does not show a valid inference: the proposition 'the car is clean' by
itself does not entail that 'John must have washed it'. Rather what is minimally needed is another premise
like:[16]
If the car is clean, John must have washed
it.
When an
assumption like premise 1 is supplied, the inference becomes valid:
(10)
Premise
1: The car is clean.
Premise
2: If the car is clean, John must have
washed it.
Conclusion:
John must have washed the car.
However,
it seems somewhat implausible that premise 2 would be available as a ready-made
contextual assumption. It is more likely that there is another complex argument
in the place of premise 2, that starts from the general, obvious assumption:
If the car is clean, someone must have
washed it.
On the
basis of other contextual information the speaker inferred that the individual
'someone' was John rather than anybody else. Perhaps the speaker knew that John
was the only person around who could have washed the car, or he saw John's
watch lying besides the implements that had been used for washing the car and
so forth. The use of the modal 'must' indicates that this assumption goes back
to inference rather than direct observation.
The
importance of reconstructing these implicit inferences can be seen when we try
to explain why one cannot simply link up any two propositions by 'so' to create
a 'grounds-conclusion relationship'. Consider the following example:
A meteorite exploded, so my baby brother must
have dirtied his nappies again.
Though
nonsensical, this example could easily be analysed in terms of relationships:
Relationship analysis
(11)
Grounds:
A meteorite exploded,
Conclusion: (so
I conclude that) my baby brother must have dirtied his nappies again.
The
analysis does not show up in any way that this interpretation is implausible.
However,
if we try to approach it inferentially, we see that it minimally implicates
something like the following argument:
(12)
Premise
1: Whenever a meteorite explodes, my
baby
brother dirties
his nappies.
Premise
2: A meteorite exploded.
Conclusion:
My baby brother must have dirtied his
nappies again.
Now
according to relevance theory, by inducing the hearer to construct this
argument, he expects him to supply the implicated premise 1 or a set of
inferentially equivalent assumptions; however, at least with our cultural
background, premise 1 itself is not a plausible assumption, nor does there seem
to be a plausible argument that would be inferentially equivalent to it. Thus
the reason why we find it implausible, is that we are unable to supply
plausible assumptions that would be necessary to interpret the utterance along
the lines indicated by the speaker. Thus relevance theory can help us to
distinguish between plausible and implausible interpretations: implausible
interpretations involve implausible assumptions.
If all
this is correct, then what the analysis of connectives requires is not the
development of a set of relationships; rather we need to discover what
semantic, and perhaps also syntactic, properties these connectives have and how
they contribute to an interpretation consistent with the principle of
relevance, against the contextual knowledge of the audience.[17]
In the
next two sections I want to draw attention to three practical ways of testing
hypotheses about the properties of connectives.
3.3.
Testing
hypotheses about connectives
3.3.3
Testing hypotheses in texts
We have
just looked at one way of testing one's hypothesis about the meaning of a
logical connective: that was to apply the hypothesis to some given text or
text-portion, and see if it allows for an interpretation that is arguably
consistent with the principle of relevance, making use only of the notions
supplied by the theoretical framework. If no such interpretation can be
proposed, then the assumed meaning of the connective may be wrong.
Suppose,
for example, that we found occurrences of -m
where no interpretation consistent with the principle of relevance could be
construed on the assumption that the meaning of -m corresponded to that of logical 'and'. This would naturally
suggest that our hypothesis is inaccurate, and we would have to look for a
better one instead.
3.3.3
Testing the context-accessing
function
However,
relevance theory allows us to test our hypotheses in another way, too, and this
is by exploiting the fact that a connective may have a context-accessing function.
Thus according to our analysis in its focal use -m should suggest a certain kind of context to the audience.
Here is
a little experiment that I did that seems to corroborate the hypothesis that -m is focus-associated. In this
experiment I gave the following sentences one by one to three native Silt'i
speakers:
(13)
a.
wut'at ayaam laam liyookb ulbaarag
heeda.
Monday
day cow he-to-buy Ulbarag he-went
'On
Monday he went to Ulbarag to buy a cow.'
b.
wut'atim ayaam laam liyookb ulbaarag
heeda.
Monday-m day
cow he-to-buy Ulbarag he-went
'Also
on Monday he went to U. to buy a cow.'
c. wut'at ayaam laamim
liyookb ulbaarag heeda.
Monday
day cow-m he-to-buy Ulbarag he-went.
'On
Monday he went to U. to buy also a cow.'
d.
wut'at ayaam laam liyookbim ulbaarag
heeda.
Monday
day cow he-to-buy-m Ulbarag
he-went
'On
Monday he went to U. also to buy a cow.'
e.
wut'at ayaam laam liyookb ulbaaragim
heeda.
Monday day cow he-to-buy Ulbarag-m
he-went
'On
Monday he went also to U. to buy a cow.'
I did
not provide them with any contextual information, but rather as we considered
each sentence, I asked the test person whether it communicated to him anything
beyond what was actually said, and if so, what. Their answers, given
independently of one another, are summarized in the following display:
(14)
a'.
(No interpretation given beyond the information expressed in (13a).)
b'. He seems to have gone previously.
c'. He wanted to buy something else
besides a cow or he had gone previously and now went again to also buy a
cow.
d'.
He has also other business. (The K'ibbat dialect speaker inadvertently used (c)
when referring to (d), and one other speaker said that he preferred (c) instead
of (d).)
e'. The K'ibbat dialect speaker
rejected (e). The other informants interpreted it as suggesting that he had
gone elsewhere before.
Now it
is not hard to see that the answers differ according to the position of the
suffix -m, and as I try to show in my
paper (Gutt 1988), the differences correlate well with the predictions of
relevance theory with regard to the focus-background distinction. Thus, in (b) -m marked the time phrase wut'at ayaam 'on Monday' and in (b') we
find this replaced by another time phrase: 'previously'. In (c) -m marked the direct object laam 'cow', and in (c') we find the
substitution 'something else besides a cow', and so forth.
To
briefly examine just one of these examples, (b) could have the following
underlying structure:
[wut'at-m ayaam] laam liyookb ulbaarag heeda.
Here the
focused constituent would be the time phrase wut'at ayaam 'on Monday', with the -m suffixed to its leftmost subconstituent wut'at 'Monday'. Variable substitution at the focus yields the
background entailment:
[ X ]
laam liyookb ulbaarag heeda.
The X is
meant to indicate an appropriate semantic variable, ranging in this case over
time expressions. By using the focused sentence (b), the speaker induces the
hearer to look for a contextual proposition corresponding closely to the
background entailment, but with some value of X substituted at the focus. This
seems to be borne out by the fact that in our example the informants suggested
that the person had gone 'previously'.
This
brief sketch may suffice here to give an idea of how one can test whether or
not a connective constrains the use of contextual assumptions in a particular
way. For further discussion of the data itself the reader is referred to Gutt
(1988).
3.3.3
Testing by predicted failure of
communication
Relevance
theory accounts not only for 'regular', 'normal' communication, but it explains
communication breakdowns as well. In accordance with this, I tried to think of an
utterance where the presence of -m
would be predicted to lead to a breakdown of communication.
What
sort of case could that be?
According
to relevance theory, the focused constituent must be the relevance-establishing
part of the utterance. But suppose one construed an example where the focused
information is already highly accessible in the context. In that case the focal
use of -m should be felt to be
inappropriate. If this prediction turned out to be true, it could be taken as
supporting evidence that the analysis is likely to be true.
I used
the following utterances to test the hypothesis.
(15)
(a) safiiya 'saalo araashin; irasoot
ishlaan' baat.
Safiiya Salo
farmer-he-is to-plough he-can
she-said
'Safiya said, "Salo is a
farmer; he can plough."
(b) safiiya 'saalo araashin; irasootam ishlaan' baat.
Safiiya Salo
farmer-he-is to-plough-m
he-can she-said
As can
be seen, the two utterances differ only in the occurrence of -m in the (b) utterance. Yet the three
informants to whom the data were presented independently of each other rejected
the (b) utterance while accepting the (a) utterance. This seems to confirm our
prediction, and can be accounted for along the following lines.
Interpreting
-m as focally-used, we obtain the underlying
representation and the background:
underlying: [irasoota-m] ishlaan.
background: [ X ]
ishlaan.
According
to focus theory and by consistency with the principle of relevance, the
increment of information that one needs to add to the background in order to
obtain the underlying representation must be the main point, i.e. the context
modifying element, of the utterance. However, as the example has been
construed, the underlying representation is immediately preceded by:
saalo araashin.
Salo he-is-farmer
In
Silt'i culture this would normally entail very obviously that Salo is able to
plough - this is a stereotyped assumption stored in the encyclopaedic entry of araashi 'farmer'. Furthermore, according
to relevance theory, the content of the immediately preceding utterance
constitutes a highly accessible context for the utterance under consideration.
Taken together this means that there is a highly accessible context for the
underlying representation that contains what is supposed to be the main point.
In other words, the focused part, which should establish its relevance, is, in
fact, already a highly accessible contextual assumption. Under these
conditions, the audience would find it difficult to arrive at an interpretation
consistent with the principle of relevance, and it is this difficulty which can
be taken as the underlying cause of the informants' rejection of this example.[18]
4. Conclusion
In
summary, we see that relevance theory offers us a cognition-based framework
that can help us to arrive at a relevance-based reconstruction of the
speaker-intended meaning. Any such reconstruction is open to falsification by
the criterion of consistency with the principle of relevance. These
reconstructions of the speaker-intended meaning allow us to test hypotheses
about the semantic (and other) linguistic properties of connectives, both in
running text and by experimentation. There is no need to look for or develop a
set of interpropositional relationships in order to spell out how the various
utterances of a text relate to each other.
References
Ballard,
D. Lee, Robert J. Conrad, and R. E. Longacre 1971 'The deep and surface grammar
of interclausal relations' Foundations of
Language, 7, 70-118
Beekman,
John and John Callow 1974 Translating the
word of God, Zondervan, Grand Rapids
Blakemore,
Diane 1987 Semantic constraints on
relevance Blackwell, Oxford
Blass,
Regina forthcoming Relevance relations in
discourse: A study with special reference to Sissala, Cambridge University
Press
Callow,
John 1982 A semantic structure analysis
of Second Thessalonians, Summer Institute of Linguistics, Dallas
Callow,
Kathleen forthcoming Man and message
Grice,
Paul 1975 'Logic and conversation' in Peter Cole and Jerry L. Morgan (eds) Syntax and Semantics, v. 3, Academic
Press, New York, pp. 41-58; quoted from reprint in Aloysius P. Martinich (ed.) The Philosophy of language, Oxford
University Press, Oxford 1985, pp. 159-170
Gutt,
Ernst-August 1988 'Toward an analysis of pragmatic connectives in Silt'i' in Proceedings of the Eighth International
Conference of Ethiopian Studies, Addis Ababa 26-30 November 1984, ELM
Publications, Huntingdon vol. 1, pp. 665-678
Larson,
Mildred L. 1984 Meaning-based
translation: A guide to cross-language equivalence, University Press of
America, Lanham
Longacre,
Robert A. 1983 The grammar of discourse,
Plenum Press, New York
Mann,
William C. and Sandra A. Thompson 1986 Rhetorical
structure theory: Description and construction of text structures,
Information Sciences Institute Reprint Series ISI/RS-86-174, University of
Southern California, Marina del Rey
Mann,
William C. and Sandra A. Thompson 1987 Rhetorical
structure theory: A theory of text organization Information Sciences
Institute Reprint Series ISI/RS-87-190, University of Southern California,
Marina del Rey
Pike,
Kenneth L. and Evelyn G. Pike 1977 Grammatical
Analysis, Summer Institute of Linguistics, Dallas, and The University of
Texas at Arlington, Arlington
Sperber,
Dan and Deirdre Wilson 1986 Relevance:
Communication and Cognition, Blackwell, Oxford
(E.-A.
Gutt, October 18, 1989)
[1] Excerpt from Eugene Loos' letter of
invitation dated June 15, 1987.
[2] For a very insightful treatment and
much broader treatment of discourse connectives in general from a
relevance-theoretic point of view see Blakemore (1987) with data from English,
and Blass (forthcoming) with data from a Ghanaian and other languages.
[3] As you will notice quite soon,
relevance theory approaches communication from a somewhat different perspective
from what many of us are used to. It may seem tempting to "translate"
the concepts of relevance theory into your own framework. However, as I know
from my own experience, this is not a straightforward matter, and you may be
better off by first of all concentrating on understanding relevance theory on
its own terms, and then try to relate it to other models later on.
[4] If the utterances mentioned are
intended to convey the message that it is time to go, then they would
constitute what has come to be known "indirect speech acts". However,
this does not mean that inference plays a role only in indirect speech acts.
Rather, as Sperber and Wilson (1986) have shown, inference is basic to all
verbal and non-verbal instances of human communication.
[5] In her presentation of "Man and
Message", Kathleen Callow referred to this metaphorically by saying that
the new information needs 'to plug into an existing frame'. Relevance theory
spells out that this 'plugging in' is done inferentially.
[6] As Sperber and Wilson (1986) show,
this single principle takes care of all that the four maxims suggested by Grice
(1975) were meant to do.
[7] In the literature it is sometimes
said that communicators do not express information already known to the
audience because this is unnecessary. Relevance theory shows it is not only
unnecessary but detrimental to the success of the communication act: it makes
the audience spend processing effort without increasing the contextual effects,
and hence reduces the overall relevance of the utterance. This explains our
feeling of irrelevance when someone starts cluing us in about a person who is
actually well-known to us. (Often we tend to cut such introductions short by
saying, "Oh, I know Phil.") On the other hand, expressing known
information is not always inappropriate, reminders being an obvious example.
Relevance theory can account for these cases as well: it can be consistent with
the principle of relevance to express known information if, for example, it
would cost the audience more effort to find that information in memory than to
derive it from a verbal reminder. For
further discussion see Sperber and Wilson (1986), pp. 149ff.
[8] I use the term 'logical' here with
some reservations since it can easily be misunderstood as referring to matters
of standard logics only. It would perhaps be better to use the wider term
'inferential' here, but since the workshop referred to them as 'logical'
connectives, it seems best to follow that terminology here. Cf. also note 17
below.
[9] Note that the term 'constraint on
relevance' does not refer to a lessening of the degree of relevance; on the
contrary, constraints on relevance can be seen as increasing the relevance of
an utterance in that they aid the audience in finding the intended
interpretation. For an extensive treatment of this notion see Blakemore (1987).
[10] This is not to suggest that
connectives always must have one sense nor that there cannot be homophones.
However, as will become clearer below, there is reason to suspect that
relationship-based accounts involve unnecessary multiplicity of meaning.
[11] As Blakemore (1987) and Blass
(forthcoming) have shown, connectives and particles tend to work by imposing
such constraints on the relevance, hence on the interpretation of utterances.
[12] I am not entirely satisfied with
the formulation of these syntactic conditions yet. Thus it may be possible to
eliminate the disjunction here if one assumes that the "focus" can
extend to the sentence as a whole. Furthermore, there is some evidence that -m is not attached to the leftmost constituent
when the focus is on a whole subordinate clause. However, since these syntactic
details do not affect the thrust of the general argument, I will not pursue
them now.
[13] This follows from the assumption
that the logical form of an utterance is incomplete in a number of ways,
including the exact time reference; thus a sentence in the simple past tense
can refer to virtually any point of time in the past. The time appropriate to
that particular utterance is worked out pragmatically by consistency with the
principle of relevance. Since the background is an entailment of logical form,
it also has no fixed time reference yet.
[14] Relevance theory assumes that the
strength of the conclusion of an argument depends on the strength of its
weakest premise.
[15] It should be noted that in our
discussion we have interpreted the hyena's utterance from the point of view of
his audience, i.e. of the donkey. If we continued to interpret it from the
perspective of the storyteller's audience we would find further contextual effects;
thus the naiveté of the donkey in assuming that the hyena would let her go
would almost certainly have the effect of strengthening and solidifying the
cultural assumption that donkeys are stupid, and conversely that hyenas are
clever.
[16] The qualification "minimally
needed" means that by this one assumption the argument could be completed.
However, as I point out below, this assumption itself is more likely the
conclusion to another sequence of inferences.
[17] It may be objected that my own
analysis of -m relies on an
interpropositional relationship - that is, the logical relationship established
by 'and'. Hence are we not just reducing the number of interpropositional
relationships to a smaller, more basic set, perhaps equivalent to the basic
operations of standard logics? Firstly, Sperber and Wilson do not assume that
there is a "radical distinction between concepts such as and, if
... then, and or, which are
regarded as proper logical concepts, and concepts such as when, know, run, bachelor, which are considered non-logical.
Following another tradition, we regard these other concepts as also determining
logical implications. Which concepts do or do not have logical entries, which
rules these contain, ... are all matters for empirical investigation." (1986,
p. 87) Thus relevance theory does not assume there to be a small set of logical
relationships in the first place.
Secondly,
some argumentative connectives do not encode a logical or inferential
relationship at all. For example, Blakemore shows that anyway can indicate "that the proposition it introduces is
relevant in a context that does not include the immediately preceding
remark" (Blakemore 1987, p. 141). This 'meaning' is not reducible to any
more basic, logical relationship, and there is no reason why it should be. As
pointed out already, the meaning of such connectives is largely a matter of
language-specific semantics.
[18] The test results need not be as
straightforward as this; in fact, in a second test the informant reactions
seemed to disagree. However, this disagreement, too, provided valuable clues
about the interpretation process and could be accounted for in terms of
relevance theory. For details see Gutt (1988).