Wednesday, April 9, 2008

Relevance Theory

This semester I’m attending a course on Dan Sperber and Deirdre Wilson's (1987) Relevance Theory, a major theory of how communicative understanding arises in the pragmatic satiation of face to face discourse. I think I’ll write a bit about it time and again, and especially look at how their proposals relate to the stuff I usually write about on this blog.

In the first Chapter, on Communication, Sperber and Wilson pose the basic question:

“How do human beings communicate with one another?” (1)

But to answer this question, we first need to know what exactly is meant by the fuzzy term ‘communication’. Sperber and Wilson define it as a process that involves two “information-processing devices” (i.e. people, their mind/brains or what have you) in which one of the devices ‘modifies’ the others physical environment in some way or other. Taking the example of speech, If I talk to you, I modify your acoustic environment by sending acoustic speech signals out into your surroundings.

If you read this blog (Hi dad!), I modify your visual environment by presenting colored marks against a white background which you have to decipher. What happens as a result is that you (given that your are the second, receiving “information-processing device”) construct a (mental) representation which bears resemblance to the representation that is stored in my head (or in the “first device”).

It is difficult to say how similar our mental representation of what I’ve said or written are exactly, but we’ll come to that again, and presumably, there is to be some overlapping of representations, given that we both speak English, live in the same physical world, have basically the same general neural architecture, a similar genetic makeup, and similar basic cognitive skills (that is, we at least both know how to sit down, how to switch on a computer, how to type and how to read, for otherwise how would you’ve gotten her in the first place? (Yes I know, you wouldn’t necessarily need to be able to type if you bookmarked my blog or read it via a newsfeed (I know that there are thirty of you somewhere out there…) but that would be nitpicking)

Coming back to the theory, the basic questions now are:

  1. What exactly it is that is communicated? Sperber and Wilson postpone an elaborate answer to a later chapter, and suffice it to say that what is transmitted are thoughts, i.e. “conceptual representations", and assumptions, i.e. claims about a factual state in the real world, or information, which isn’t really defined at that point.
  1. How is communication, i.e. the establishment of functionally overlapping mental representations, achieved exactly?

To answer this question, Sperber and Wilson first draw our attention to the paradigm of semantics they are opposed to, namely the “dualistic code model.” This model proposes that

communication is achieved by encoding and decoding messages.” (2)

Sperber and Wilson champion another approach, developed by philosophers such as Paul Grice and David Lewis, called the inferential model, according to which,

“communication is achieved by producing and interpreting evidence.” (2)

Both models aren’t totally opposed to each other, but lay different emphasis on which parts of understanding are really important. What Sperber and Wilson argue is that these two ways of understanding are independent of each other.

But first, let’s look a bit closer at the code model. Sperber and Wilsin define the three key elements of the model as such:

Code: “a system which pairs message with signals, enabling two information-processing devices (organisms or machines) to communicate.” (3f.)

Message: “a representation internal to the communicating devices.” (4)

Signal: “a modification of the external environment which can be produced by one device and recognised by the other.“ (4)

A classical example of such a way of information transmission would be the morse code. Sperber and Wilson also give the example of honeybees, who have been shown to be able to communicate the location of nectar they’ve found by ‘dancing’. To be able to do so, a honeybee, seen as an information-processing device, has to possess the following properties: a) some kind of memory, from which the relevant information about where the nectar is and how to get there can be taken or put b) some kind of “encoder-decoder device“ which can pair the message of the flight plan with the signal of ‘dancing’. (5)

The ‘semiotic’ code model goes way back to the ancient greeks, and is also endorsed by Ferdinand de Saussure, one of the founders of modern linguistics:

“Language is a system of signs that express ideas” (Saussure 1974: 16)

But Sperber and Wilson claim that there is the model is deficient, especially when it comes to complex devices (i.e. us) processing and transmitting information. If we interpret a linguistic signal, we do indeed encode a message, but the information that we get as a whole, our representation of it, doesn’t only come from the encoding of the message alone. In some situations there needn’t be a ‘coded message’ at all. As the authors argue

“What a better understanding of myth, literature, ritual, etc., has shown is that these cultural phenomena do not, in general, serve to convey precise and predictable messages.”. (8)

But how then can we explain this apparent

“gap between the semantic representations of sentences and the thoughts actually communicated by utterances” (9) ?

First, Sperber and Wilson differentiate between ‘sentence’ and ‘utterance’. An utterance is the actual communicative act in a given situation with all its linguistic and non-linguistic (e.g. situational) properties, whereas a sentence is meant to describe the more abstract general semantic of a content.

Consider the following examples:

(1) I write a lot about Zombies

(2) George likes brains

(3) Michael is glad that the members of the Max Planck Institute for Evolutionary Anthropology haven’t been eaten by zombies

if we ask what the content of these ‘sentences’ is, i.e. what their general semantic representation is we only get that (1) An agent writes a lot about Zombies. The same holds for (2). In (3) without any situational, pragmatic content, we just don’t know who this Michael guy is. Therefore, for the interpretations of these ‘utterances’ to be successful, they need to

“involve an interaction between linguistic structure and non-linguistic information.” (10)

Moreover, "utterances are used not only to convey thoughts but to reveal the speaker's attitude to, or relation to, the thought expressed” (10f.)

In fact in a lot of utterances there is a lot of ambiguity judging from the linguistic material alone. Consider for example, irony (which of course can be partly interpreted due to prosodic features of an utterance such as tone of voice, stress, etc.), or Chomsky’s famous example that

“flying planes can be dangerous.”
In general, it can be said that there is an amount of semantic flexibility in the signal for which the hearer must compensate by non-linguistic considerations such as situational/pragmatic cues. Often, a linguistic signals' meaning isn’t exhausted by its semantic content, but has to be combined with a lot of contextual information they carry implicitly, such as “Do you know what time it is?”, by which we of course mean, ‘Please tell me what time it is if you know”.

If we want to delve even deeper, we can now ask what exactly the mechanisms are that allow us to successfully draw inferences as to the full representational content of an utterance. However, most rules aren’t as simple as

(4) “Substitute for 'I' a reference to the speaker.”

(5) “Substitute for 'tomorrow' a reference to the day after the utterance.” (12)

This of course would work for sentences such as 'I write a lot about Zombies', so that we would get that it is me, Michael Pleyer, who writes a lot about Zombies (well, in fact I think, my amount of Zombie references is moderate).

But according to Sperber and Wilson, it is hard to think of an exhaustive list of such ‘decoding principles’, and then it still seems dubious that such an endless process of feature checking and decoding would be psychologically plausible.

As most pragmatists, Sperber and Wilson advertise the view that comprehension is in fact an inferential process.

To get a clear view for the differences between the inferential model and the code model, they give the following definitions:

“An inferential process starts from a set of premises and results in a set of conclusions which follow logically from, or are at least warranted by, the premises” (12f.)[1]

A decoding process starts from a signal and results in the recovery of a message which is associated to the signal by an underlying code” (13) [2]

To achieve successful communication(i.e. partly overlapping mental representations), then, two people need to have the same basic premises and use the same inferential mechanism. How exactly this comes about is addressed in later sections, but to hint at the basic principle, Sperber and Wilson think that one of the most basic premises of interactions is that of relevance, i.e. that we only try to draw attention to propositions, thoughts, assumption, information. etc. that wee deem relevant, i.e. of interest to you and me.

I’ll post a bit more on the book some time in the future.


Saussure, Ferdinand de (1974): Course in general linguistics. Translated from the French (1916) by Wade Baskin. London: Peter Owen.

Sperber, Dan and Deirdre, Wilson (1995): Relevance: Communication and Cognition. Second Edition. Malden et al.: Blackwell.

[1] The most classic inference process is the following:

All men are mortal

Socrates is a man

Therefore Socrates is mortal.

[2] Sperber and Wilson’s discussion of the relation between inferential models and the code model is actually quite complicated, because in their view, inferential models can be used to decode messages, but not vice versa, as you will see in this example

No comments: