03 - Relevance Theory

Pros and Cons of Grice

Grice's theory improves on Speech Act theory as he explains how implicatures are triggered (violation of maxims combined with the cooperative principle). However, he does not describe the inferential process in detail.

In addition, interaction is not always cooperative, e.g.

1. A: Where have you been?
   B: Why are you asking?

B's response triggers an implicature without presupposing that they are cooperative. Also, the maxim of quality (truthfulness) does not work in a binary fashion, e.g.

1. He has a square face
2. Holland is a flat country

(NB his face is not literally square and Holland is not literally falt, like a billiards table).

Enter Relevance Theory!

Proposed by Dan Sperber, and Deirdre Wilson, with further work from Robyn Carston. It is a general theory of pragmatic communication which explains a wide range of phenomena. Unlike Grice, it details how inferences are made. It is also more specific about the issue of cognition versus culture. Whereas Grice did not account for his maxims / the cooperative principle, e.g. do they arise from cultural / cognitive process, RT firmly argues that inference is driven by cognitive processes.

How / why we make inferences

Positive contextual effects

The inferences we make lead to "positive contextual effects", i.e. an increase in our knowledge about the world. These arise from three processes:

a. Strengthening / confirming an assumption b. Contradicting / eliminating an assumption c. Combining with an existing assumption to yield a new assumption.

1. A: Would you like a coffee?
   B: Coffee would keep me awake.

This may

a. Confirm/strengthen A's assumption that B might want a coffee b. Confirm/strengthen A's assumption that the exam is tomorrow. c. Contradict/elimiate A's assumption that B might want a coffee

Implicated premises and conclusions

Inferences are often made using a 2-stage process: (a) Activating an existing assumption, called the implicated premise, and creating a new asusmption called the implicated conclusion, e.g.

1. A: Is he a good cook?
   B: He's Italian.

IMPLICATED PREMISE: Italian's are good at cooking. IMPLICATED CONCLUSION: He's good at cooking.

1. A: Was Fred invited to Ruby's party?
   B: Fred isn't a friend of Ruby.

IMPLICATED PREMISE: Only friends are invited to parties. IMPLICATED CONCLUSION: Fred wasn't invited to the party.

1. A: Did Tanya pass the exam?
   B: Well she was wandering around with a great big smile on her face.

IMPLICATED PREMISE 1: Passing exams makes you happy IMPLICATED PREMISE 2: People who are happy walk around smiling. IMPLICATED CONCLUSION: Tanya passed the exam.

Knowing when to stop

There is nothing to stop us making very complex premises and arriving at very complex conclusions, e.g.

1. A: Is he a good cook?
   B: He's Italian.

IMPLICATED PREMISE: Italian's are good at cooking pizza and pasta but aren't very good at making curries. IMPLICATED CONCLUSION: He's good at cooking pizza and pasta, but not very good at making curries.

We avoid this via an effort principle. We do just enough inferencing to genrate a contextual effect, and then stop.

The principle of relevance

Sperber and Wilson encapsulate their account via a single principle, the principle of relevance. An utterance is optimally relevant if (a) generates maximum contextual effects, or (b) minimal effort.

The cognitive principle of relevance

This principle of relevance depends on a broader cognitive principle: the cognitive principle of relevance. According to this, all inferences, whether linguistic or non-linguistic, are driven by a quest for relevance. For example, if you hear the doorbell you could make the following inferences:

a. That could be Jenny (strengthening your assumption that Jenny is planning to call) b. That could be the The TV licence person (strengthening your assumption that a TV licence person is likely to call) c. The doorbell works (contradicting your assumption that the doorbell is broken) d. John must have fixed it (combining with an existing assumption that John is planning to fix the doorbell, to yield a new assumption, that John has finally got around to fixing it).

Notice that you could also make an overly complex inference, e.g. the doorbell works on a Tuesday, but this is ruled out by the least effort principle.

The communicative principle of relevance

This can be seen as an analogues of the cooperative principle, i.e. an overaching principle. Each utterance creates an expectation of relevance. Every time someone speaks they guarantee that their utterance is relevant. This entails that it is worthwhile for the hearer to expend mental energy in order to make inferences and arrive at the implicature. Because RT does not insist that spakers are cooperative it can explain non-cooperative communication, e.g.

1. A: Where have you been?
   B: Why are you asking?

Because it has no maximum of quality it can explain utterances which are partially truthful, e.g.

1. He has a square face
2. Holland is a flat country

Why bother with implicit communication?

This is a key question in pragmatics. Because pragmatic inferences are defeasible, there is a risk that we will make the wrong inference. Why depend on a system where such errors are possible?

With pragmatic inference we can

a. Convey far more than we could convey explicitly b. Do this in such a way that we can deny any inferences.

So the following example

A: Would you like a coffee?
B: Coffee would keep me awake.

may generate a number of important inferences, e.g.

a. No. b. I have an exam tomorrow (and therefore wish to go to bed early). c. Please don't play your music loud.

According to RT, each utterance carries a guarantee of optimal relevance, which makes it worth the listener's while to engage in substantial inferential process. Up to a certain point, the more effort we make to recover inferred meaning, the greater oure reward will be. This helps use to convey substantial inferred content.

Another example is as follows

1. A: How is your new tennis partner?
   B: He's a bit of a John MacEnroe.

We can generate a wide variety of inferences from this, e.g. he has a temper, he shouts a lot, he argues every point, he is willing to resort to subterfuge to win a point, he is mercurial etc. This is far more information than can be conveyed if we communicate explicitly, e.g. B: "He's badly behaved".

Informative versus communicative intentions

For the sentence

1. That's my ice cream

The informative intention is "I'm informing you that the ice-cream belongs to me". The communicative intention is "I wish to communicate to you that I'm informing you that the ice-cream belongs to me." The "communicative" intention is the deeper level of intention that describes the main speech act. Notice that if we were to merely interpret the informative intention, we might respond "Thank you very much for that information", which is clearly not what the speaker intended. A speaker who is responding to the communicative intention might say "Oh, I'm sorry, I didn't realise it was yours", which is a more appropriate response.

Explicit versus implicit communication

RT is very good at describing phenomena that lie part way between implicit and explicit communication. It specifies a level called the explicature, which is a development of the form of the utterance, but is also subject to inference. There are 3 key inferential processes which take place at the level of the explicature:

a. Reference assignment: He went to the bank (He refers to Peter) b. Disambiguation: He went to the bank (Bank refers to a financial institutions, not a bottle bank, or a river margin) c. Enrichment: Resolving semantic incompleteness / making lexical meanings more specific

Here are some examples of enrichment:

1. A: Will you post these letters?
   B: I will [post them]
2. Jack and Jill are married [to each other]
3. Bill insulted his bos and [as a result] got fired
4. You'll get promoted if [and only if] you work hard
5. Ralph is ready [to go to work]
6. Nina has had enough [pasta to eat]

Here are some examples where enrichment is used to make lexical meanings more specific:

1. He ate a large meal (compare with Felix is a large cat, or Alpha Centauri is a large star)

And less specific:

1. John has a square face (not literally square)
2. Holland is a flat country (not literally flat)

Putting it all together

Pragmatic communication can break down at a number of different levels: the implicated premise, the implicated conclusion, or the explicature, e.g.

A: Is he a good cook?
B: He's Italian.

Explicature: He = new flatmate, Italian = speaker needs to work out what is intended by the word Italian in this context, e.g. of Italian nationality / from a family of Italian nationality / of Italian heritage (but not necessarily of Italian nationality)

Implications and criticisms

It makes more specific predications regarding the source of pragmatic constraints. RT is based on cognitive processes. Grice by contrast was non-committal regarding the source of his maxims. Some RT researchers have also suggested that inferential processes may be relatively modular and therefore can be impacted in individuals who have pragmatic difficulties. It is also very comprehensive, covering many aspects of pragmatic theory.

One potential Achilles heel is the least effort principle. How do we know that we have done enough processing to yield contextual effects?