An Expedient Mind: The development of LANGUAGE

AN EXPEDIENT MIND

Tartan Hen Publications : Home | more books | Contact : feedback@tartanhen.co.uk

Layer 5: Language

Word-Concept Associations
The first step in the acquisition of language, is to establish an association between the sound of a word, and a concept structure which already exists in the robot's concept-store. No new mechanism is required. The sound of a word is a sensory experience like any other. All that is required, in this first phase, is that the word-sound-experience should occur repeatedly at the same time as the concept is recognised. So the word "FIDO", for example, is heard as the object FIDO is seen. Learning a set of associations like that is easiest for physical objects. The next in line will be actions like "RUN", and "SING". More word-concept associations, involving more complex concepts, will be learned later.

Focus of Attention
Note, however, that the possession of even one single word-concept association, can be advantageous. One person can draw the attention of another, to some aspect of the environment, by just pointing. Saying a single word, like "Mummy", or "lookout!", can be helpful in exactly the same way. These, single-word communications are effective because they affect the focus of attention. Later, as more sophisticated forms are developed, that effect (i.e. the direct effect on the focus of attention), will continue to play an active role in linguistic communication. Indeed, it is not possible to understand most verbal statements, without being able to identify the current (and constantly changing) focus to which attention is being directed. As I remarked earlier, evolution operates upon what already exists, and what previously existed, invariably remains and is incorporated into each new form.

The Construction Kit Theory of Language
We saw earlier that the interpretation process operates on concept structures and uses them as the components of a more complicated and dynamically changing #interpretation structure which represents the circumstances currently being experienced. It is a simple extention of that arrangement, for the component concepts to be delivered to the interpretation process, not by the mechanism of direct sensory experience, but by hearing a string of words. Each word is associated with a concept, so the interpretation process can do its work on these concepts and construct an #interpretation of the spoken utterance.
This is what I call "The construction kit theory of language". Language, in this view, is a mechanism for transferring an #interpretation structure from one person to another. The speaker puts the interpretation process into reverse. He disassembles the #interpretation structure into its component concepts. He utters the word-labels of each concept. The listener hears the words, accesses the associated concepts, froom her own concept-store, makes a copy-version of each, and builds those copy-versions (with suitable modifications) into an #interpretation structure which will be similar, but not necessarily identical, to the #interpretation structure in the speaker's head.

Objections
This idea, or some version of it, has often been suggested. It has equally often been dismissed as infeasible. Here is Jean Fodor, an acknowledged expert in the field of semantics, first explaining the construction kit theory, and then rejecting it ...

‘... this theory is something like the following. I have a thought or idea. I formulate a sentence, I utter it to you, and when you hear it you come to have the same thought as me. This may seem a crude picture of the way language is used but it is not obviously a false one. As a theory of meaning it says that an expression has a meaning if and only if it is associated (universally? standardly?) with some idea, and that two expressions have the same meaning if and only if they are associated with the same idea. ...... but ...... mental imagery is shifting and arbitrary, and differs both in extent and kind from person to person and from occasion to occasion. On one day the word ‘tablecloth’ may conjure up in me the depressing thought of laundry that needs to be done, on another it may make me think of an elderly aunt who sends the same gift every Christmas, for someone else ‘tablecloth’ may be associated with a party game, and yet another person may have no image associated with it at all. There is simply no stable correlation between imagery and the meanings of expressions heard or uttered. And there are probably no images at all which would serve to explicate the meanings of words like “how”, ”despite”, ”to”.’ (Fodor 1982).

Here is another objection, this time by Kirkham. Kirkham raises the issue of word ambiguity. His example concerns the word ‘bank’ (financial institution or the margins of a river).

‘Another example is “John’s bank is eroding”. Here the meanings of ‘bank’ and “eroding” are mutually dependent. ... [they] ... can be assigned meanings only as a married couple. These examples show how it is not possible to assign meanings to individual words one by one and build up the meanings of sentences therefrom, and the capacity of a theory to do just that was supposed to keep it to a finite size.’ (Kirkham 1995, p243).

Counter-arguments
My response to each of those objects are as follows -

(1) There is no standard version of any concept. That is true. Each individual person (or robot) will have a unique structure for each concept. I do not think, however, that that in any way invalidates the construction kit theory. Consider an analogous situation. Two DIY enthusiasts, Tom and Fred, are discussing the construction of a rabbit hutch by telephone or email. Each has a workshop and each has a junk-box full of bits and pieces discarded from previous DIY projects.

Tom: "It is easy if you start with a packing case."
Fred: "What size?"
Tom: "Big enough to accommodate a rabbit in comfort."
Fred: I don't have one big enough."
Tom: "Then you will need five sheets of plywood. Four for the walls, and one for a sloping roof."
Fred: "Don't you need a floor?"
Tom: "Depends. If you give it a floor you will need to lift it off the ground to stop it rotting."

.... and so on ....

No one would suggest that a conversation of that kind would be impossible unless Tom and Fred had junk-boxes containing identical bits and pieces. If each builds a rabbit hutch based on this discussion, it is very unlikely that the two rabbit hutches will look the same. But they will perform the same function. They will both keep accommodate a rabbit in relative comfort. And that is what is important. If you and I have a conversation about "a house", it does not matter, if you are thinking about one made of wood while I have in mind a brick structure. If it was important, we could add a few more words. The phrase "a wooden house", would bring our mental models a little closer. And if they need to be brought closer still, we can add still more words. So I reject the first of these objections.

(2) There is no obvious concept with which words like "how" and "despite" can be associated. That is true only if we discount the possibility of a concept referring to the contents of someone's MIND. The #interpretation structure I envisage, however, specifically includes the contents of the MIND. The concept associated with the word "how", contains an implicit reference to a PERSON, and to that person's MIND. Within that MIND structure is the representation of some circumstance. That representation is deficient however. It lacks any description of the causal factor which created that circumstance. It is also shown within this structure that the fact that that causal link is missing, causes a NASTY condition in the person's MIND. There is, therefore, an implication (or prediction) that if the missing information is supplied by another person, the first person would be pleased (cause a NICE condition).
Some may find that explanation a bit convoluted, but I cannot see any reason why it is invalid. My claim is that my system would work. I did not claim that it would be aesthetically pleasing to purists who attach great significance to mathematical elegance.
A similar explanation for the meanign of the word "despite" can be constructed. It does not involve elements which are missing from someone's mental representations. It does involve the expectations which a person might reasonably have and a circumstance which contradicts those expectations.

(3) Words are often ambiguous. That is true. According to Kirkham, that means that in some circumstances, meanings cannot be assigned to individual words, but to word pairs (and, presumably, also to larger groups of words). That assertion is based on the assumption that meanings (or concept structures) are fixed and cannot be changed.
That , however, is not true of the system which I envisage. The copy-version of each concept can be changed as circumstances require. If each concept is like the piece of a jig-saw puzzle, and the #interpretation structure is like the finished jig-saw picture, then each concept (or jig-saw piece) is, in effect, made out of rubber. It can be bent and twisted, and bits of it deleted, to accommodate the requirements of the other pieces.
Furthermore, if the word "bank" is associated (ambiguously) with two distinct concept structures, the interpretation process can select the one which best fits the requirements of the concept associated with "eroding". The concept labelled by "eroding" will refer to some physical object which comes into contact with another moving physical object, and is, as a result, gradually diminished. When "bank" and "eroding" are placed in the same sentence, the physical object which is being eroded will be linked to the margin of a river, and the moving object which does the eroding action will be linked to the water in the river. It might, in an other context be possible to interpret the sentence as the gradual undermining of the financial stability of a commercial institution. The wider context will determine which is most appropriate. My system can accommodate both interpretations, and, as the context continues to expand, will always be open to a re-assessment of what is the correct interpretation,

The Communication of Emotions
In the previous chapter (on the Theory of Mind) I argued that an understanding of the behaviour of other people, is one most important (if not THE most important) aspects of our mental abilities. The reading of body language and of facial expressions are a vital part of that understanding. The same argument applies to linguistic communication. Language is, par excellance, a means to communicate feelings. It come as no surprise, therefore, as is made clear in later sections on representation, that the meanings of the majority of words, carry within those meaning structures, reference to that inner world of the MIND. The meanning of even a simple word like "yes", is almost entirely devoted to the mental condition of the speaker. I do not understand how language processing would be possible, without that mental dimension.

Tartan Hen Publications : Home | more books | Contact : feedback@tartanhen.co.uk