CHAPTER 17

The Meaning of Meaning

17.1 Internal Representation and Predictions

When a person hears a statement, he or she translates it into an internal representation of some kind. At its simplest level, this representation is descriptive of some aspect of the world about us. If the statement is 'your book is lying on the table in the room next door' the internal structure which is formed would contain the representation of an entity {book}, of an entity {table} and of two entities {room-I} and {room-2}. The speaker and the hearer will also be represented as two entities {speaker} and {hearer}. Both will be represented as being {inside room-I} and the table and book will be represented as being {inside room-2}.

The two rooms will be represented as being {adjacent} and the book will be represented as being {on the table}.

We have placed angle brackets round the representations of entities to indicate that we do not want to get bogged down at present in the detail of exactly how these things are represented.

[NOTE (added in 2004.) Angle brackets have been replaced by curly brackets to avoid a clash with the notation of HTML]

The hearer will then be able to process this composite representation and make predictions about hypothetical situations. For example, if the person concerned goes into the room next door, the book will brought into view and can thus be obtained. The speaker will have a similar representation and this representation, or model, will include a representation of the hearer's represen­tation. It will also include a representation of the hearer's motivation, that is, his wanting to obtain the book. Based on this guess about what is going on in the hearer's mind, the speaker will predict that the hearer will take action to fmd the book or at least express gratitude for the information which makes this possible. If such action is not observed, the speaker may conclude that the hearer has either not heard or not understood the statement.

This somewhat complicated description summarises the nature of a conversa­tion. Two people engaged in a conversation are constantly constructing representations of the meanings of the statements made by each other, and they also form representations of each other's representations. The internal condition of another person's mind is for all of us an important piece of information. In an environment which is constantly being changed by the actions of other people, we would be in dire straits if we had no idea what to expect of other people's behaviour.

Understanding then, requires two activities: the construction of an internal representation, and the processing of the representation to make appropriate predictions.



17.2 Context and Implications

The phrase 'appropriate predictions' is used advisedly. There are many predictions which could be made from any such representation, most of which would be irrelevant. For example, if the room next door was to be filled with water, then the book would get wet. If the book is very heavy then the table may be damaged, and so on. These predictions might become appropriate however, if we have additional information which provides a context for the statement. For example, we may know that there is a burst water pipe in the room next door, or we may know that the table in the room next door is at present being glued together and is not capable of bearing heavy weights. The internal representation constructed by the person who 'understands' the statement will include a representation of the context. The context will determine which of the possible predictions are relevant. The inclusion of information about the context, far from being an unnecessary complication, is the only hope we have of avoiding the embarrassment of dealing with an infmity of possible predictions. We shall call these relevant predictions the 'implications of the statement'. It is our contention that people do not understand a statement (as we do) unless they are aware of the implications (of which we are aware).

17.3 Multiple Meanings and Partial Meaning

Different persons will have different experience and, in consequence, different contextual information available to them. They will be aware, therefore, of different implications for any given statement. It may be supposed that the person who utters a statement, or who writes it for others to read, is the true arbiter of its meaning, but there are circumstances where this is not so. It is easy to envisage a scenario where a speaker addressing a meeting might make a statement which, to his surprise, causes instant panic or hilarity in his audience because it has some implication of which he was unaware. In such a case he might be described as not (fully) understanding what he has just said. The parenthetic 'fully' is important, and suggests the notion of partial understanding and therefore of partial meaning.

Since we have no reason to suppose that anyone person is in a privileged position in this respect, it follows that any given statement may have several different and equally valid 'meanings'. At the same time, although each person may be aware of a unique set of implications in the context of their unique experience, each will be aware to some extent of how much of that experience is common to many people. They will therefore be aware of how much of their interpretation is shared. This judgement about what is shared is what we regard as the meaning of a statement. We do not expect others to understand our unique and innermost thoughts, but we do expect a degree of common understanding, and it is this that serves as the 'meaning' used in communication. It is easy to see how misunderstandings can arise.

17.4 Uniformity of Representational Form

Another thing emerges from this example. The contextual information which affects the internal representation of a statement may not have been gained by hearing other statements. It may have been the result of direct observation. It has been found that people are sometimes unable to distinguish between the representation they construct as the result of visual observation and that generated by processing linguistic utterances. In laboratory experiments, witnesses to a motor accident (on fIlm) were asked the questions:

(I) 'Did you see broken glass on the road after the two cars SMASHED INTO one another', and

(2) 'Did you see broken glass on the road after the two cars HIT each other'.

They were more inclined to say 'yes' to question (I), although in the fIlm there had been no broken glass at all. The mental representation of a 'smash' is more likely to generate an inference of broken glass than a mere 'hit', and this inference influenced the answers. The witnesses seemed unaware that they had allowed an inference based on linguistic clues to modify what they thought they had actually seen.

If the two types of information can be combined, then they must be constructed from the same kind of constructional units. We deduce that the internal representation formed when we 'understand' a linguistic statement is made of the same kind of stuff as our other mental constructs, and that the information which is derived from several different sources can be integrated into a single form. In other words, we do not have a special kind of semantic representation which is used only in the processing of language. Language processing makes use of the mental processes and representations which are associated with intelligent behaviour, and which were developed long before the ability to use language.

17.5 Perception

We perceive the world by means of our senses, and our view of the world is governed by the physical characteristics of those senses and by the processing which we carry out upon the data input. Sensory stimulations generated by sight, hearing, smell, taste and touch are recorded and stored as exact images. They are called eidetic images. Under normal circumstances, however, the information is only accessible to us in that form for a very short length of time. During that time it is processed.

'Perceptions' are the results obtained by processing sensory data. Much of this processing is more or less automatic, being to a large extent 'hard-wired' or 'firm­wired' into our sensory systems.

An important aspect of this processing is the identification of 'features'. In vision, the features associated with shape consist of lines, edges, comers, etc. Processing is fast and unconscious, so that we are not aware that what we see is actually a collection of these features and a mosaic of colours. What we perceive are entities such as people and other objects. We also appear to have standard templates for the purposes of recognising and distinguishing specific and important patterns, such as that associated with the human face.

Our perceptual apparatus is extremely poor at determining absolute values, but is extremely efficient at detecting subtle differences. Thus most of us can tell that two musical notes are not identical with amazing discrimination, but are poor at identifying the absolute pitch of one note. We are able to detect the motion of objects which we would overlook completely if they remained still. Encoding is in the form of change and difference. We remember faces in terms of the way they differ from a standard template pattern.

In general we classify entities by the properties they share and we distinguish individual members of a class by the differences between them and the standard for their class.

17.6 Memory

There are at least three types of memory store involved in human cognition.

Naturalistic memory traces. There is evidence which suggests that eidetic memory traces persist in the brain indefmitely for all persons, but although they might be physically present, they become 'forgotten' after a very short time and can no longer be retrieved (normally). People with so-called 'photographic memories' appear able to access precise details of past experiences, so that at a later time they can 'read' from their memory trace information of which they were not consciously aware at the time of the actual experience. For most practical purposes, however, we can ignore the existence of eidetic or naturalistic memory.

Episodic memory. Our memory of past events, or 'episodes', is normally recalled in a degraded form with details deleted and, as the example of the traffic accident witnesses shows, details concocted from our knowledge of what 'ought' to have happened. When asked to recall sentences read to them a short time before, people are often unable to recall the exact wording, but are usually able to paraphrase the sentences. What gets remembered is the meaning of the statement. Meaningless data is much more difficult to commit to memory than meaningful data. Episodic memory, therefore, is a trace of past experience, f1ltered and interpreted in terms of its significance.

Semantic memory. Semantic memory is a store of 'concepts' in which the memory traces of events and entities have been analysed and classified to become 'general knowledge'. In semantic memory even more detail has been lost, and what is retained is a compendium of several experiences and observations which have been classified and associated to form concepts. A concept represents a class of perceptions. Concepts are associated with other concepts for ease of retrieval.

General knowledge. Two of the most important aspects of general knowledge are:

(a) the ability to recognise the classification of a specific entity, which may never have been observed before, by virtue of the properties it shares with other, known entities. This allows us to predict some of its less obvious properties (e.g. whether it is likely to be dangerous).

(b) the ability to predict the occurrence of events on the basis of events currently or recently observed. This is done by abstracting the patterns of 'causation' from the trace of past events. 'Causation' is the way in which events of one type follow upon events of another. Current events can then be matched with these past experiences and classified so that the likely outcome of events can be predicted.

Concept organisation. To build an effective semantic memory it is necessary to process the episodic memory and to organise and classify what is found there. It is likely that such classification is multi-dimensional, so that a single experience or episode will be classified along with similar experiences according to different criteria. For example, episodic concepts could be classified on the basis of similarity of outcome, similarity of cause, similarity of entities involved and, perhaps most important of all, similarity of emotional effect on ourselves.

Other concepts are associated with distinct objects which could be classified by their shared attributes, such as having the property 'edible'. Evidence for this is provided by patients who have suffered brain damage (e.g. as the result of a stroke) and who are sometimes found to have lost, selectively, the ability to use a particular class of words, such as the class of words associated with the 'edible' property, or with verbs of action (i.e. not including the verb 'to be'). Since this loss is associated with damage to a particular location in the brain, the obvious conclusion is that for a given patient that location had been part of the 'index­table' or 'classific tion' structure. By analogy, if we delete a node in a tree structure we lose all of its subordinate nodes. Such an organisation of memory would be helpful for a speaker in finding appropriate words to match his or her meaning.

The everyday experiences of a typical adult are not usually greatly different from other past experiences. The memory trace of these experiences is therefore quickly analysed and subsumed within existing concepts in semantic memory.

However, completely fresh types of experience, particularly those associated with charged emotions, do not fit easily with past experiences and may trigger the formation of a new concept. In such a case, the 'general' case is based on only one example, and therefore retains much of the detail of that example. Slowly, as other examples of that kind are added, the particular details of the first experience (that is, those which are peculiar to that experience), are deleted, while the aspects of the experience which it has in common with others (those which characterise the concept) are reinforced. The concept gradually becomes more general and loses specific detail. Unique experiences remain etched in some detail in our memories, since no other experience of that kind is available to generalise it.

17.7 Labelling Concepts and Learning new Concepts

Labelling a concept. It is possible to label concepts symbolically. The symbol used could be a visual pattern, a gesture, a sound or a word. A label is used like an index in a filing system to access the concept. A language is a system of symbols together with rules governing their use.

Any concept can be labelled linguistically, and many can be described by patterns of lingistic symbols. In this way we can transfer concepts to each other verbally without the need for direct experience. Not all concepts can be described adequately by language, however. Motor actions, such as the actions necessary to achieve a parallel turn in skiing, or a good golf swing, can often defeat efforts to analyse them linguistically. The only effective way to have a mental concept corresponding to these things is to have them encoded within the mind in terms of the muscular patterns necessary to achieve them.

Learning a new concept. Learning the meaning of a new word may mean identifying a concept which we already have and simply indexing it by that word. For example, most people will have seen and noticed the vertical and horizontal strips of wood in a window which separate and hold the panes of glass. These strips will be part of the concept 'window' , and they will also constitute a concept in their own right in the minds of most people. Perhaps not so many people have a word-label for this concept. When told that the correct term is 'astragal' they are able to append it to the existing concept.

On other occasions the learning of a new word-label may mean constructing a new concept from an assortment of existing structures in semantic memory. As an example of this, let us invent a new word and a concept to go with it.

'Glunging'. An example of 'glunging' takes place when the driver of a car decides to overtake the car in front, pulls out, and is almost irrevocably committed to the manoeuvre when another oncoming car appears round the bend ahead. Our glunging driver breaks dramatically and aborts the overtake procedure, suffering as he does so a mixture of fear and embarrassment. Glunging also takes place when a man decides to proposition a woman at a party, and after launching himself on his 'lead in' patter suddenly realises that she is his boss's wife, and has to extract himself from the situation in some confusion.

The reader should now be in a position to think of other situations not involving motor car driving or parties to which the term 'glunging' might be applied.

We now share an understanding as to what 'glunging' means. Our ideas may not coincide exactly, but the degree of common knowledge will suffice to allow communication. To an extent this account will have modified the reader's semantic memory store, so that he or she now has a concept which was not there before. Previously the constituent parts of the 'glunging' concept were dispersed through several other concepts, and were obscured by detail. It is unlikely that it was recognised as a concept in its own right. But now it has been identified and labelled. Language not only labels concepts but modifies our view of the world. The concept 'to glunge' has a structure roughly as follows:

There exists an entity PERSON. PERSON is in a state of indecision. PERSON wants to take some ACTION but is inhibited by fear of some ADVERSE CONSEQUENCE and by the disapproval of others which might accrue if the venture fails and therefore appears stupid At length the PERSQN decides, impetuously, to take the ACTION. At a point when there is almost no chance of taking an alternative course of action, ADVERSE CONSEQUENCE suddenly appears certain. The PERSON aborts the ACTION in circumstances where it is fairly obvious to others that PERSON has blundered The PERSON suffers embarrassment in consequence.

The words in upper case represent parameters which can be replaced by specific identifiers or descriptions to suit the particular circumstances. By replacing the parameters with specific values the concept above could generate all possible examples of glunging. Hence, it is a 'generic' structure. This is strictly analogous with our concept 'triangle', used in the micro-graphics example, which could generate all possible examples of 'triangle' by giving the parameters (vertex coordinates) the full range of numeric values. In the notation of predicate calculus we would have:

glunge( {person}, {action}, {adverse consequence}

To use the verb 'to glunge' in natural English, we need to invent a grammatical structure for it so that the various parametric values can be attached conveniently. We might describe our first example of glunging above with these words:

'He glunged the overtaking of the car away from a head on collision'. In the second case we would say:

'He glunged a pass at the woman away from an entanglement with his boss's wife.'

The structure is {person} glunge {action} away from {adverse consequence}.

'To glunge' has some similarities with the verb 'to abort', but has overtones of embarrassment and absurdity.

Note the use of the words 'away from' in this structure. They play an important but subordinate role. They help to identify the role of the entity corresponding to the formal parameter which follows them. They trigger an action, which is to evaluate or process the description of the entity which follows, and then to fit that description into the role-slot provided by the verb 'to glunge'. In a similar way the word 'the' in the phrase 'the man' triggers a search for some previous mention of an entity 'man', so that the new description of that entity can be linked to the past description. These considerations suggest that certain words are labels not for passively descriptive concepts but for active processes or 'functions'.

17.8 Types of Concept

We note that concepts exhibit the characteristics of:

(a)  entities
(b)  episodes
(c)  functions



A given concept may incorporate more than one of these characteristics, being partly descriptive of some entity, partly episodic and partly functional. The concept 'tennis ball' for example, involves a description of a physical object but also a description of a typical tennis match (episodic). The concept which is labelled by the word 'and' describes a single (conjoined) entity and also triggers a search for two entities which are to be conjoined.

Roughly speaking, 'entity' concepts are associated with nouns, episodic concepts are associated with verbs, and functional concepts are associated with prepositions and conjunctions. We shall argue later, however, that this association with nouns, verbs and prepositions is imprecise, and that, for example, some nouns are as 'episodic' as many verbs.