top of page

From Post-Behaviourism to Language

 

The term 'Behaviourism' is as well known within the Psychological Sciences as it is notorious. The term is most closely associated with the Psychologist B. F. Skinner. Although Pavlov and Watson started the movement, Skinner became its high priest, lauding upon his subjects from his Harvard throne for almost 60 years. The David to Skinner's Goliath was a young MIT linguist called Noam Chomsky, who in 1959 correctly pointed out the major flaw in Skinner's 'Verbal Behaviour' theory- its inability to incorporate the kinds of hierarchical representations required by language codes. Skinners reputation never fully recovered from this event.

 

However, the bare bones of the story as presented above bely an inescapable underlying truth- that Skinner's theory (a movable feast) provided the dominant scientific narrative for over half a century. This was because it was correct enough to be very useful. Skinner had a habit of called his latest theoretical update by the same name- 'Neobehaviourism'.

 

At the heart of Skinner's view is an idea which is still a sound one- that  language is just another type of behaviour, albeit one whose 'effectors' (the mouth, tongue, teeth and vocal tract/the hands*) and 'sensors' (the ears/the eyes*) interoperate via the invisible medium of acoustic vibrations in the air.

In the diagram, the perceptual features recognised by the subject are divided into parallel (concurrently processed) sensori-motor behaviour loops (from both self and others/world) and serial synto-semantic language loops. Of course, when several speech streams are being processed at once (as in the cocktail party effect), language processing becomes parallel, although not symmetrical. The point being made is important- speech (and, by implication, writing) has been deliberately constructed to be made of serial symbol streams that mimic the predominantly serial cause-effect chains observed in the world at large. That is, language has 'deliberately**' evolved in order that humans are equipped with a generic cause-effect modelling mechanism with the ability to infinitely nest (ie 'quote') expressions in a mathematical sense.

 

Following Skinner(1959), and deferring to Chomsky(1960), an updated, modern form of Skinner's Neo-Behaviourism is presented which we have called Post-Behaviourism(PB) as much to differentiate it from Skinner's Neo-Behaviourist paradigm as any other reason. In PB, branched S->R and R->S forms are introduced as well as their more familiar chained forms, so that PB is able to incorporate hierarchical representation forms. The transfer of theoretical benefit is not only from linguistics to Behaviourism. PB has given to linguistics the new concept of class-exemplar equivalence principle (CEEP).

 

The implications of this new principle could not be more profound and wide-reaching for the moribund institution of contemporary linguistics. Until the discovery of this principle by the author, no one had been able to resolve the contradictions between the semantics of 'surface' and 'deep' grammatical structures, ie between Constituency and Dependency analytical viewpoints. For the first time, CEEP offers a pathway out of this dilemma. The application of the CEEP principle toward the solution of this problem is rather subtle, and the serious student will find that several rereadings will be necessary to understand it fully- after all, if it has eluded a mind as great as Chomsky's, it must therefore present significant latent barriers to comprehension.

 

Imagine that, consistent with the Dependency view, an infant learns where to use the term 'breakfast' by means of the physical situation of being fed in the morning, and having the mother saying, "Baby want some breakfast". The infant learns that 'breakfast' is the morning meal.

 

As it builds up its experience of other eating situations, the infant learns to differentiate between various types of mealtimes. From the Post-Behaviourist (PB) viewpoint, the Dependency association of the word 'breakfast' in the infant's mind is constructed via classical conditioning (PB-observation). The infant's mother (or father, less often) acts like Pavlov to his eponymous hound, by supplying the anticipatory acoustic cue to the arrival of food.

 

Consider the word 'table'. Different flat surfaces of this type require small or large adjustments in motor programming (ie so-called procedural knowledge), depending on such factors as evenness of the surface, friction coefficients etc. For the PB-experimentation to produce consistently unique solutions, some variables such as added Constituent word order must be fixed. In English, we place adjectives like 'orange' in front of nouns like 'table', as in 'orange table', whereas in some Romance languages like French, the adjective appears afterward, as in 'table orange'. The infant's mind somehow learns to ignore secondary terms appearing consistently before (or after) the primary Dependent semantic association. CEEP is the mechanism that resolves the puzzle of this 'somehow'.

 

In CEEP, the term 'table' is not just a terminal semantic value, when used alone, but a class descriptor also, when preceded by English adjectives. The same mechanism applies to verbs, and adverbs too. To the infant's mind, 'table' and 'orange table' are actually treated like 'branches' of a descending data 'tree' or hierarchy, as in '/table' and '/table/orange', to use the nomenclature of Operating System Hierarchical File Systems (OS-HFS). It is the ubiquitous OS-HFS that serves as the prime computational analogy.

 

Some time ago, I had the sudden realisation that the semantics of SVO sentence forms can be modelled by command line script formats in computer operating systems like Unix, PC-DOS and Windows. This is my 'Einstein moment', equivalent to the moment when he realised that accelerating (non-inertial) frames and gravity were entirely equivalent (I hope that doesn't sound too pompous). Suddenly, via this insight, the way that language fits into the Pierceian triad (see figure below) became clear. The mind is equivalent to a computer file system at the knowledge-base level. This analogy explains the way that words can be used in almost any semantic role, after all, you can make a new directory (using mkdir or md) with any name you choose.

 

In other words, another way of coming to the required insights is to note the essential similarity between valid linguistic sentences and directory::file pathnames in the scripts used to customise digital computer operating systems for sub-domains and individual users. When file locations are invoked as arguments in top-level OS scripts (eg Windows .bat and .cmd files and Unix shell scripts), the lines in the script exhibit key morphological resemblances to sentences. The quickest way to explain this is to use a simple SVO (Subject-Verb-Object) example- "John has a blue shirt".

 

We can compare this to a scripted line by recognising one of the syntactic conventions of batch files, that of latent path contexts, and rewriting the sentence as-

C:\Users(speakers)\John C:\Programs(verbs)\To Be\Present Tense\has a C:\Users\John\possessions\clothing\blue shirt.

 

Far from being a contrived example, the above comparison highlights an essential communicative property of language which is inadequately treated in the literature, that of situational set sub-selection*** (S4). Language is a flexible communicative behaviour controlled by cognitive representations. These representations are not only used to represent physical situations, but to also virtual (ie rerepresented) ones. In structural terms, we have the communicative equivalent of mirrors facing each other, and reflecting within each others frames a potentially infinite regress of reflected (ie re-represented) images.

This phrase indicates another fault with the conventional approach to AI- the blind assumption that any Boolean variable must refer to a proposition (eg true or false). In the example above, 'John', 'blue' and 'shirt' are all perceptual features (see Treisman). Their presence or absence in a situation is a Boolean variable. Whether John has or doesn't have a blue shirt is a proposition, but the component parts are not. It is a fallacy to allocate 'true' or 'false' values to non-propositional forms, akin to the Liar's Paradox, 'This statement is false'. Propositions must be of the form x=(y=z).

*referring to speech/writing as the modalities, and phonemes/lexemes as the perceptual features that act as building blocks.

 

**we can safely use the word 'deliberately' to indicate a simpler process than full-featured human thought, by understanding it to mean a teleological (tropic, connative) or goal-driven computation. Evolution is goal-driven, but the goal is not a conventional, statically-defined one, it is a dynamic, recursive target, that of reproductive fitness. since the phrase 'survival of the fittest' is a logical tautology, unless a recursive approach is taken.

 

***its poor treatment is reflected by the fact that there is no widely agreed name for the topic, hence the rather clumsy one chosen here. Maybe 'directed ambiguity resolution ' could also be used. The underlying idea is that ambiguity is deliberately used by language as its MAIN SEMANTIC FUNCTION

bottom of page