All ideas expressed in Thor's Articles and The Passionate Skeptic are entirely those of the author, who has no aim to influence, proselytize or persuade others to a point of view. He is pleased if his writing generates reflection in readers, either for or against the sentiment of the argument.

University of Melbourne Postgraduate Linguistics Conference, May 1992



Thor May

University of Melbourne, 1992


Abstract: This paper questions the sources of linguistic creativity by considering the corpus of an idiolect (that is, one individual's grammar). The objective analysis of this corpus led the researcher to speculate that the use of mental constructs, specifically language, in real time had a kind of immediate "resonance" in the brain which increased the likelihood of their repetition, either exactly or with simple grammatical modifications. The phenomenon is defined in this paper as "mnemonic resonance". If this resonance patterning were general then it would have profound consequences for listener decoding strategies which depend heavily upon collocational predictability. At a theoretical level, mnemonic resonance would also have consequences for many existing linguistic models. 


Table of Contents: Introduction / 1. Mnemonic Resonance / 2. A cognitive processing model / 3. The domain of mnemonic resonance / 4. Textual analysis / 5. Make / 6. The tripartite relationship of Agent-Verbal_Idea-Theme / 7. "Idea particles" and parallel processing / 8. Have / 9. Rhetorical parallelism / 10. Dream logic to social logic / 11. The mechanics of a recollection and utterance / 12. Know / 13. Not knowing / 14. [know] & mnemonic resonance / 15. Branching collocations / 16. The non-viability of static collocation constraints / 17. Mnemonic resonance & fluctuating collocation weightings / 18. [Everything].. : The role of VEX forms in resonance and recall / 19. Summary / 20. Conclusion  / Appendix 1: Extract from CH Corpus  / Appendix II: Item line references for CH Extract / References / Notes





This paper is speculative. Reflecting that, I have often chosen to insert a first person pronoun to remind the reader that there is no pretended scientific certainty about what is proposed. Some of the suggestions might be naive. The argument relies directly on no model known to the author, and makes rather few scholarly references. However the observations are drawn from studying an extract in a quite large corpus of 133,000 words from one individual. The main purpose of a larger research agenda**, of which the ideas here are a small part, is to determine just how linguistically creative one speaker is in generating her language, and in particular the extent to which she uses pre-constructed formulas[ii], as opposed to generating unique strings of language every time she speaks. In the course of trying to understand how this speaker — let's call her CH[iii] — generates her language, I have been driven back to theorize about the fundamental cognitive processes which may be involved. The suggestion here is that mnemonic resonance may be one of those processes.


     1. Mnemonic Resonance


What is mnemonic resonance? Mnemonic resonance is the description adopted in this paper for the persistence in memory of an idea after use — since we are talking linguistics, let's say of a linguistic element after use — and the heightened availability of this element to a progressing discourse for recall or for evoking a field of linguistic associations.

To imagine this concept of resonance more clearly, think of a collection of orchestral instruments. When a note is struck on one instrument, it does not die immediately, but persists with a diminishing resonance for several seconds. Ultimately the note may fade, it may be renewed by the same instrument, or amplified by others, or it may set the harmonic pattern for a developing theme.

In an unfolding discourse it is often noticeable that words, phrases, syntactic patterns, semantic fields, speech rhythms, and indeed other discrete elements of the language, can be seen to recur in some form over a stretch of utterance. The recurrence is often not exact (though it may be), yet can be recognized as a variation on the theme.

The argument developed here is that recurrent patterns in discourse are partly a functional consequence of what I have called mnemonic resonance. There are also instances when the resonance may be dysfunctional, causing generative or communicative breakdown. The underlying assumption is that this resonance is a direct property of electrical or chemical activity in the neural centers where language is processed. While the neural hypothesis itself must remain an untested proposal in this linguistic analysis, it still constitutes a useful base for constructing a model a) to partly explain the patterning of discourse, as well as (paradoxically) pattern distortion; and b) to suggest how speakers are able to generate a linear stream of linguistic symbols at the amazing rate of six thousand or so words an hour.



     2. A Cognitive Processing Model


Before we get on to discussing particular language examples, it may be useful to put this idea of mnemonic resonance into the wider context of cognitive process. For argument's sake, let us suppose that cognitive processes can roughly be broken into the following, which are summarized in Table 1.


input processes — that is, those processes involved in acquiring stimuli to be manipulated;

management processes — that is, those devices built into the design of the machine which help it to manage the stimuli productively;

transformation processes — that is, some of the patterns of changes which typically occur to stimuli in the process of being utilized and manipulated;

output processes — which may generate behaviour observable to those outside of the individual, or may simply feed back into the cognitive cycle.


Table 1  Notional Diagram of Cognitive Processes



The input processes would be mostly concerned with perceptual mechanisms, together with the inherent and acquired pathways for distributing stimuli passing through the perceptual mechanisms. We won't be discussing that level of process any further.

With management processes we will be fairly centrally concerned, because these impact upon everything which follows. Two elements of the management processes seem especially important. The first is mnemonic resonance, which is the real subject of this paper. I will come back to that presently. The second would be cognitive organizing principles, which apply not only to language, but to all cognitive processes. Of these organizing principles I think two kinds of patterns are very pertinent to language. One is the principle of syntagmatic association, which in language refers to the collocation of word classes to construct a string, and the other organizing principle is paradigmatic association, an instance of which in language would be the ability to substitute alternative lexical items into a phrase-structure slot.

Syntagmatic association appears to rely partly upon acquired or inherited rule ordering. Paradigmatic association draws strongly on what linguists call semantic fields, and these apply quite differently to different kinds of word classes.

It is assumed for convenience here that the transformational patterning of cognitive process takes place in what we could call the holding bay of memory, which I have suggested is partly sustained by mnemonic resonance. "The holding bay of memory" can be conveniently divided into facsimile memory, substituting memory and generalizing memory.

Facsimile memory is, in a sense, the absence of transformation — yes, I know I am being naive here but brevity is useful. An extreme example of "pure" facsimile memory would be the gift that some individuals seem to have of so-called photographic recall. A more pertinent example might be entirely fixed and memorized content, such as the words of a song.

The second category of transformation patterning in memory would be substitution, where drawing on the organizing principle of paradigmatic association, a lexical item (for example) is substituted by another lexical item or phrase. Note that the process is not necessarily linguistic; (e.g. a word can substitute for a smile, and vice versa).

The third category transformation patterning I've called generalizing, which draws on a combination of paradigmatic and syntagmatic association.

The output phase of cognitive process has its analogues in transformational patterning. For example, at one linguistic level, facsimile semantic memory would probably be represented by simple denotation (although we know of course that the listener will always attribute connotation as well as denotation to any message). One example of substitution at output would be metaphor, and an example of generalizing output would be some kind of classification.




3. The Domain Of Mnemonic Resonance


For the purposes of linguistic analysis, mnemonic resonance may be inferred from dynamic relationships amongst the elements of a text. Let us consider the salient elements of such a text as expression fields (exfd). (The reader is begged here to tolerate a series of abbreviations which will make it easier to organize proposals later). Expression fields may be words or phrases or any manifest linguistic element, and may embed other expression fields. Crudely, each expression field is initiated by a non-linguistic mental event (mtev) to which is normally applied a selection from the word/phrase store, ie. the lexicon (lex), and some kind of syntax (stx). Where a shared context (shcx) is  presupposed, or where there is phrase duplication (phdu), the expression field may be wholly or partly ellipted (ellp). Where the shared context is very familiar(fam), it may be expressed as a linguistic formula (forml). Thus






mental event ->

mtev -> [lex x stx] + ...


expression field

.. +{(shcx)-> exfd


ellipted exp-field

.. +{(fam)(phdu)-> ellp exfd


formula exp-field

.. +{(fam)(shcx)-> forml exfd

Table 2:  Derivation of Expression Fields from Mental Events

I propose that the process does not end there. The mnemonic resonance (mnr) from each expression field may be transformed by a whole range of other mental events, which will lead to a variety of linguistic consequences.  

There may be a decision to sharpen (shp) the mnemonic resonance from the expression field (exfd), maybe because of positive listener response (poslr), or maybe for an internal reason. This is often achieved by repetition (rep), and/or an elaboration of detail. On the other hand, the focus of the mnemonic resonance may be blurred (blur), often by a trailing vague expression (vex) at the end of a list. This broader but weaker resonance can occasionally excite more peripheral associations, if they happen to be prominent in recent memory, but more often signals a shift in concentration to a new topic (newtop).  

The mnemonic resonance may give rise to a semantic echo (seco), generating another expression field, usually expressed as some kind of lexical substitution (sub): a pronominal, a synonym etc. The mnemonic resonance may also give rise to a phonological echo (peco) which is semantically unrelated to the first expression field, and may not even be part of the same topic set. The second expression field generated in this instance will be a homophone (homp) of the first expression field. An attempt to extend the expression field into a full proposition (prop) may result in organizational failure (ogfa) and thus be truncated as an anacoluthon (anco).

The expression field itself may distract the speaker from her discourse goal (disgl), with a couple of possible consequences. One outcome could be a digression (digr). Another could be temporary memory loss (meml), covered with a vague expression (vex). Where the speaker perceives negative listener response (neglr) there may be a reformulation which is ambiguous (ambig) or even contradictory (contra) to the original expression field. Finally, when the expression field has exhausted a topic (endtop), it may be accompanied by a terminal marker (term) such as falling intonation, or spontaneously generate a new topic (newtop). This catalogue certainly does not exhaust the possibilities, but our progress can be conveniently summarized in the following table:







shp,mnr,(poslr) -> rep(exfd1)


vague expression

blur,mnr -> vex



exfd1,mnr -> seco -> sub



exfd1,mnr -> peco -> homp



exfd -> ~prop -> anco



exfd -> ~disgl -> digr


vague expression

exfd -> meml -> ~disgl -> vex



exfd1 -> neglr -> ambig



exfd1 -> neglr -> contra


terminal marker

exfd,endtop -> term


new topic

exfd,endtop -> newtop

Table 3: Modification of Expression Fields  [~=not;failure]


4. Textual Analysis


The clearest way to present a model of mnemonic resonance is to  examine an actual text. Please refer to Appendix 1 (text extract) and Appendix 2 (line references). Certain lexical items will be selected from the extract to act as fulcrums for the discussion. I will start with a couple of verbs, because verbs have important functional properties in controlling the internal patterns of language strings, as well as invoking sometimes colourful reference to non-linguistic entities. As each issue emerges a digression will be made to examine general principles before returning to the empirical analysis.



5. Make


The verb [make] and its past form [made] occurs four times in the extract:


up to make a switchboard 21

an old oil can he'd make into 15

have a board made 41

nb: see 11 (fitted up etc)


they made a jammer 114

Table 4: Contexts for [make, made]
In Appendix I Text; (numbers in the table refer to line numbers).

The frequency of [make, made] in this extract is 0.74% compared to 0.14%  ([make]x100 + [made]x81) in the CH corpus overall. In other words, it is about 500% more common in this extract than elsewhere. that is a very high ratio, but it is perhaps the sort of clustering about a specific mental event that you would expect. If you are discussing a particular thing, then you are going to use a particular word a great deal. However it is the idea of making, rather than any particular event, which is recurrent in the above examples. This becomes even more striking if we extend [make] as a "semantic resonance"  to the extra five references in item 11 [fitted up; wired up/into; pulled apart].

This general verbal idea of making or unmaking becomes a thread to stitch together the Agent (Item 1 [my son, he, him] or later, Item 30 [boys, they]) with the broad theme of electronic objects, which take a whole variety of forms (items 5, 6, 8, 9, 18, 28, 44, 66).




6. The tripartite relationship of Agent-Verbal_Idea-Theme


The tripartite relationship of Agent-Verbal_Idea-Theme gives the CH extract whatever formal coherence it has, but the linguistic details often mesh awkwardly, as if each strand were evoking associations independently. (I distinguish "linguistic details" here from the register and general cultural posture, which is very consistent). It is also noticeable that the focus of thematic relations becomes progressively dispersed from Agent (two items: [my son, boys]), to Verbal_Idea (four, or possibly seven items of a class: [make, fit up, wire up,pull apart] etc.) to the Theme (eight items — eight electronic objects — in a general class).

The dispersal of thematic relations across the discourse seems to call into question the psychological reality of any model which proposes a tight generative dependence between the semantic and syntactic elements of linguistic strings, especially above clause level. What seems more likely is that every activated expression field ( mostly lexical items) can be thought of as a briefly "resonating idea particle" which evokes a sympathetic resonance in other dormant elements.

In the case of grammatical function words that sympathetic resonance will be dominated by syntagmatic collocations which are narrowly linguistic: suitable members of other word classes that are sufficiently close to recall for stimulation. In the case of open-set lexical items like nouns, the sympathetic resonance will relate more to a general idea that can be conceived of independently of its linguistic form. It will range more freely through paradigmatic options: repetition of the same item, homonyms, hyponyms, homophones, synonyms, antonyms, general thematic associations, and so on. Verbs seem to resonate strongly for both syntagmatic and paradigmatic associations, which accounts perhaps for their central role in generating and binding discourse.



7. "Idea particles" and parallel processing


It might seem that a cacophony of resonating "idea particles" could never give rise to the orchestrated syntax which marks natural language and makes communication possible. However where different classes of expression field acquire characteristic tendencies of collocation, of the kind suggested in the previous section (though obviously more complex) then we could predict that any synchronic population of resonating particles would rapidly pattern on a best-fit basis. Indeed another way of thinking about the whole paradigm is in terms of node weighting of the kind found in parallel processing models[iv]. Even the simple existing versions of such models have been able to generate, complex and ordered outcomes. Chaos Theory[v], which has recently been applied to an astounding range of ordered outcomes from complex natural events, may also have some bearing on what is going on, but is beyond our brief in this paper.

In one sense then, linguistic generation might be described as a mathematical process. If this is the case, then mnemonic resonance, the dispersal of thematic focus across a discourse (noted above), and the progressive loosening of syntactic arrangement from the level of morphology to the level of text may all conceivably be properties of the same mathematical design. Of course, that process can be constrained by other variables, as in planned discourse. For example, repetition stemming (perhaps) from mnemonic resonance in a novelist's mind, could be stylistically undesirable. However, repetition stemming from a technical requirement, as in a legal or engineering document, might also be stylistically required. Indeed, an essential difference between language generation and other natural processes is a creative, but controlled, tension in language generation between automatic patterning and the conscious interference of a persona with communicative goals. In the CH text we are dealing with language which seems to show little conscious social manipulation, but is subject of course to all the usual sociolinguistic pressures of its domain.


8. Have


Along with [make], the verb [have] is one of the major organizing ideas in this extract. In fact it occurs seventeen times in various incarnations. I have had to omit contracted auxiliary forms of [have]. They were just too difficult to extract from the overall corpus for comparison. This is not a minor omission. But even with the remainder of full lexical expressions there are at least five different [have] forms. The overall frequency of [have] in the extract is 3.3% (70.6%, 12/17 occurrences, are the PAST form [had]) compared to an overall frequency in the corpus of 1.2%. (In the corpus 58%, 911/1563 occurrences, are [had]. In total for the corpus 0.7% [had] + 0.5% [have]= 1.2% lexical overall frequency).


I wouldn't have let him 110

I had one time the toaster talk back to me  55

they had to do  88

Table 5a: Contexts for [have=Aux], [have=Experience] &

                                     [have to] in Appendix I text


he'd have it pulled apart  14

he'd have a board made  15

he had it all fitted up  30

he'd have this wired into that  45

he's had wires hooked through  49

he'd have intercommunications  50

Table 5b: Contexts for [have=Modal+NP+V-Participle] in Appendix I text


my son had one  4

if he had one  6

he had this board  19

I had a real old one  95

it had the maggie* and all 97  [*maggie = magneto]

it had a few valves  101

asked me could he have it  106

they had a different.. 115


 Table 5c: Contexts for [have=Possessive] in Appendix 1 text


[have to]

5.9% (1/17)




5.9% (1/17)



5.9% (1/17)



  + NP + Participle]

35% (6/17)




47% (8/17)



Table 5d: Relative distribution of [have] homonyms in Appendix 1 Text

In the sample there is a heavy preponderance of [have=Modal + NP + Participle] and [have=Possessive], evenly divided, and both controlling a class of object NPs with the general semantic property of "electronic device". As we noted, the frequency of [have] here exceeds that in the main corpus by a factor of three, which is probably consistent with its function as the second principal verbal idea ([make] is the other one) stitching together discourse in the extract.



     9. Rhetorical Parallelism


Note that occurrences of [have=MODAL + NP + PARTICIPLE] all appear between lines 14 and 50 where they frame a litany of generalizations about what CH's son would [have-verbed to] electronic items. What is happening here? It is a shadow, almost, of the rhetorical cycle in oral epic poetry. The Bible is not my usual reading, but I picked it off a bookshelf and opened it at random. This is what I saw: .. And he will appoint him captains../ And he will take your daughters../ And he will take your fields.../ And he will take the tenth of your seed... [Samuel 8:12-15]). 

 From CH we have:


And two weeks later he'd have it pulled apart

And he'd have a board made

And he'd run around all his mates

And all the broken down trannies

And god knows what

And he had this board

        with um all the little trannies

        up to make a switchboard

And god knows what


Perhaps CH is not generating great literature, but it seems to me that the same inherent process is going on somehow in that extract from the King James Bible, and the extract from CH. Is she being a clever raconteur? Is this clever talk? No, I don't think so. She is trying to "paint me a picture" of her son's typical behaviour, but stylistic control seems no part of her conscious repertoire. I'm pretty sure of that because I sat down and listened to the lady for a week. However, CH's mind at this point is being dominated by the idea of "he had X". The patterning is something which forms from the layers of recollection as they emerge, and for this analysis, that is the point.

The discourse structure, and even the phrase structure seems to be determined in large part by systematic laws which govern recall, and those in turn may be largely governed by systematic laws which govern the ways in which we store information in memory ("information" here being a non-linguistic concept). At each layer of translation something will be lost, and the micro-generative rules of that layer will reshape the message somewhat, but the final code passed by voice into the airwaves may owe as much to the original patterning of general memory and recall as to the patterning of conventional linguistic syntax and phonology. 


     10. Dream logic to social logic


Anyone who has ever set out to write down a dream knows that the uncensored files of private memory have their own strange logic of association, but there is a logic. It seems to flag events for personal significance rather than sequence or "real" cause & effect, "real" size, and so on. Language, by and large, is a waking phenomenon, but there is no reason to suppose that the ideas which it makes manifest differ in their genesis from those of dreams. Arguably, language mediates between the private logic we sometimes access through dreams, and the culturally defined logic of our external world where significance must have a social dimension. Since this mediation is lifelong rather than a temporary compromise, it seems difficult to deny that it must shape not only each momentary message, but the medium itself. If these propositions are valid, then linguistic study must make reference both to social outcomes (sociolinguistics) and to general cognitive patterns (cognitive science).




     11. The mechanics of a recollection and utterance


Imagine then, CH locked onto a composite memory of her son. This memory will be defined by a myriad of old events, each made prominent or diminished by their private significance to CH. Sticking out of the heap is a bag labeled "electronic objects". She has given this a tug and from it tumbles a string of image fragments: "trannies", "boards", "record players" and so on. They are tagged with the idea of  "pulled apart/wired up".

She has already formulated the syntactic string "if he had one of them" , which began her discourse (see Appendix 1). That was an anaphoric link and acknowledgement to the preceding speaker, so the lexical unit [have] exerts a current resonance.  "He'd have it pulled apart" thus represents the most available lexical collection to express her recollection.

The recalled image fragments are essentially similar, and in enumerating them through language CH takes advantage of this in two ways. Firstly, the syntactic pattern of [have=MODAL + NP + PARTICIPLE] is repeated and reinforced.

Secondly she generalizes them as a class through loosely applied labels — what I call vague (VEX) expressions. A "vague expression" is defined as a word or phrase whose exact reference class remains unspecified and extensible. To this end CH uses indeterminate quantifiers ([one], [all]), unreferenced collective nouns ([things]) and trailing list expressions ([and god knows what]). VEX forms are patching devices with a multitude of uses. In the context being discussed, VEX labeling supports an arbitrary class of items that can be described with a single resonating verbal idea.

Now let's consider a different kind of verb.



     12. Know


Excluding [you know] (which has its own properties), [know] occurs seven times in the extract:


and god knows what  18, 22

I don't know how it was/whether it was..70

but if I'd knew what for I wouldn't.. 109

you never knew what they were... 126

I didn't know whether I was Arthur or Martha or whether it was gunna start raining boxes 59

Table 6: Contexts for [know; know + whether] in Appendix I text

The seven occurrences of [know/knew] have a frequency of 1.3% in the extract compared to a frequency of 0.36% in the total corpus (53 [knew] + 432 [know]. The corpus frequency was obtained by subtracting 975 occurrences of [you know]).

There are a few striking properties of these collocations containing [know]. One is that they tend to have a formulaic feel about them. The pre-eminent example of this is the phatic phrase [you know], dealt with separately. The idioms on lines 18, 22 and 59 are frozen forms. The [Arthur or Martha] expression doesn't occur anywhere else (useful proof that CH has a store of formulas which are not necessarily deployed often), while [and god knows what] is a relatively popular vague expression with 38 further occurrences in the corpus. There are interesting observations to be made about all of these expressions, but here I will merely note in passing a rhetorical parallelism in the [Arthur or Martha] expression which was already observed with HAVE. The parallelism seems to be an elemental consequence of amplified mnemonic resonance.



     13. Not knowing


Every use of [know] in the extract is about not knowing. Only lines 59 and 70 mark this with the standard NEG particle, but the semantic sense of the other phrases is unmistakable. Even the high frequency [you know] may express a disclaimer by the speaker about her own unique state of knowing. The negative associations of [know] could be governed perhaps more by the expectations of CH's sub-culture than any objective state of knowledge. It is socially distasteful in traditional Anglo-Australian culture, and especially to a woman of CH's generation — she was in her late 50's at the time of recording —  to be seen as a "know-all". Although English has nothing approaching the politeness-markers of Japanese or Korean to constrain linguistic outcomes, there are probably collocation sub-sets like the [know-NEG] association with a similar origin, and they may well be marked for gender like their Japanese analogues. The formulaic dominance of examples in these KNOW-extracts, noted above, reinforces this conclusion.

The social disclaimer role of [know-NEG] is supported by its general distribution in the corpus. There are 103 occurrences of [I don't know] but only 4 uses of [I didn't know]. Since the CH tapes are essentially personal recollections, most relating to quite old events, the predominance of the present tense [know] is difficult to explain on contextual grounds. Note that psych-verbs like [know] do not even participate in discourse styles such as the "historic present". The subordinating conjunction [whether] also tends to occur in negative contexts, which is consistent with the [know] collocations. The phrase [I don't know whether] occurs 38 times in the corpus (or 39 times with the rephrasing of line 70), whereas [know whether] alone without the NEG occurs only twice. 


     14. [know] & mnemonic resonance


How then does the behaviour of [know] relate to our general theory of mnemonic resonance? I believe that it is another instance of a resonating idea inserting itself into the discourse, more or less independently of topic. It is also independent of the event-syntax (which the thematic verb structure represents with PAST tense morphology). Thus the use of [know] here is morphologically 1st-PERSON-PRESENT, it is NEGATIVE, again independently of the event-syntax, and it is concentrated in this extract at 400% of the corpus level. [NEG+know] has powerful salience in CH's culture, and once evoked seems to have considerable mnemonic persistence.



     15. Branching Collocations


The non-frozen phrases from Item 32 display a property which needs further investigation. I will call them branching collocations.


[I don't know]  x 103 times

[I don't know whether] x 38

[I don't know whether I] x 1  



[I don't know whether it] x 7


[I don't know how] x 11

[I don't know how I] x 1



[I don't know how it was]x 1

you never  x 15

you never knew  x 5

you never knew what x1

Table 7: Examples of branching collocations in the CH corpus 

The pattern is scarcely surprising. Even in a totally generative system the probability of say, six elements combining more than once is exponentially smaller than the probability of two elements combining more than once. No natural language system is purely generative however. The most enthusiastic generative grammarian would concede that individual lexical items in word classes are weighted for probable combination with individual lexical items in other word classes. To the extent that these weightings for collocation are increased, the number of different strings likely to be generated in a given time frame is constrained. We have already referred to this in the speculation about parallel processing.




16. The non-viability of static collocation constraints


CH uses some 6,600 unique word forms (ignoring homophones) in a corpus of 133,000 words, a simple ratio of about 1/20, but obviously with a pyramid of lexical items from high frequency function words to one-off nominal references. When syntactic requirements are also considered (obligatory determiners etc.), the occurrence of a 5-word collocation (counting NEG) like [I don't know whether] 38 times suggests a system that is much more severely constrained by collocation weighting than existing models assume. The branching collocation pattern, pruned by collocation weighting for all lexical items in the corpus, could rapidly reach a stage of gridlock. There is a real sense in which we must forget the collocations of old strings in order to generate new sentences.



     17. Mnemonic resonance & fluctuating collocation weightings


However a possible circuit-breaker is available to the collocation weighting paradox. If the weighting for each collocation remained a generative constant, or a match that could increase in probability for the idiolect over time (more likely), then linguistic creativity would rapidly diminish with age.

If on the other hand collocation weighting were a fluctuating value, partially governed by the recollection of historic collocations, but largely determined by immediate recall via something like mnemonic resonance, then the scope for linguistic creativity, within certain bounds, would be much more viably established. The clustering of word and phrase frequencies in the extract being analyzed, relative to their much thinner distribution in the corpus, suggests that the model of fluctuating collocation weightings is indeed closer to psychological reality than the assumptions of most existing models of grammar or discourse structure.



     18. [Everything].. : The role of VEX forms in Resonance and Recall


Item 16# [every; everything]; Item 49@# [he was onto everything]

every room in my house has got..    47

everything would be wired up..   37

everything 'd be wired..  42

in every room sort of thing..   51

but everything you picked up.. it'd talk...   52

he was onto everything..  81

Table 8: Contexts for [every; everything] in Appendix I text

The two occurrences of [every] have a frequency of 0.37% in the extract compared to a frequency of 0.1% in the total corpus (133 occurrences), and the four occurrences of [everything] have a frequency of 0.74% compared to a frequency of 0.17% in the total corpus (227 occurrences).

As with other quantifiers like Item 2 [one] and Item 4 [all], the significance of a high ratio of occurrence for Item 16 [every; everything] may turn on how unique its reference is to particular mental events. That is, if a particular event is being discussed, it is to be expected that lexical items salient to that event will have a high frequency relative to their distribution in parts of the text dealing with other topics. On the other hand, if the quantifier is a generalized vague expression, with text-internal functions (unlike [you know]) but habitually used to "blur" resonance, then a much more even distribution in the corpus could be predicted. In this case a concentration in the extract would require an independent explanation.

I placed [every] and [everything] under the same item number because it struck me that with the exception of line 81 (a likely formulaic expression), the occurrences for both seemed to cluster in one part of the extract and possibly reflect dispersion from a common resonance. In the case of [every], line 51 is a repetition of line 47. On the face of it they appear to represent unique reference to a particular mental event: the rooms in CH's house are obviously known and accountable to her. Yet the repetition itself undermines this assumption with the trailing vague expression, [sort of thing]. We are left with the impression that CH is referring to a general class of places where her son exercised his mischief. If this is so, it closely matches [everything] , which refers to a general class of objects.

We may dispute the specificity of [every], but there is no question that [everything] is a canonical vague expression. Although a whole class of VEX-forms are distributed throughout CH's language, and do much to define her style, [everything] itself is concentrated in the extract to a ratio of 7/1 over the corpus. To understand this concentration, it will be useful to reflect further on the actual function of such expressions in the psychology of language generation.

The organization of information in memory seems to be governed by two competing requirements. Firstly there is the need to structure and preserve a facsimile of sensory information which consistently represents the mind's belief of "how the world out there" works. That is, a firm view of "external reality" is necessary in order for the human organism to recall useful features of its environment in the daily struggle to survive. Anyone who has had their reality-view severely shaken by a physical or emotional crisis knows that even routine activity becomes precarious and exhausting when the ground under one's feet cannot be taken for granted. A certain kind of memory has a central role in such reality construction. It reaches its acme of development in the apparent "photographic memories" of certain individuals, and is enthusiastically rewarded by most formal education systems. In discourse structure, the facsimile role of memory is expressed by repetition, and those other discourse functions which sharpen mnemonic resonance.

The second (and competing) requirement for mnemonic organization is a capacity for extracting general patterns and principles which will allow isolated instances of experience to become a basis for predicting future events. It is this generalizing capacity which, above all, has given homo sapiens a competitive advantage over other species. Generalizing sensory experience (whether linguistic or otherwise) is a complex process that involves a stripping away of superficial characteristics to reveal underlying patterns, and best-fit matching of incomplete patterns with an expanded field of other, roughly similar templates. In other words, the process involves both destruction and synthesis.

The destruction of our facsimile reality carries a price in loss of focus. At the level of discourse function this loss of focus emerges through what I have called "blurring" above. Recall that blurring has the effect on mnemonic resonance of rendering it at once shallower and  broader: a wider range of associations may be evoked occasionally, but the low amplitude of the resonance makes it more likely to dissipate without further impact. Many of the devices of formal argumentation are props to capture and preserve weak resonance long enough to make general associations with similar classes of event. This paper itself is a perfect example of such a procedure. In casual utterance like the CH corpus, there is little desire or opportunity to develop rigorous generalizations from carefully constructed arguments. Nevertheless, the machinery of generalization proceeds automatically, with or without the discipline of formal training.

VEX forms in the CH corpus are clear manifestations of the generalizing trend in mnemonic organization. At the level of personal psychology they have several applications. They allow CH to preserve self-respect where facsimile recall has failed, and indeed mean that near-facsimile storage is only necessary long enough to allow events to be pared down to a skeleton of familiar type-forms. At the level of social interaction, VEX forms permit CH to package experience which is not merely private but, it is asserted, has a general pattern relevant enough to be recognized and valued by other people. The notion that a generalizing VEX term like [thing] can give experience more social currency may be flawed for some listeners, but it is strongly encouraged in CH's sub-culture. Excessive fluency, meaning precise, logically organized, unhesitating expression without errors, may be viewed with distrust in working-class Australian society. It seems too much like the fake knowledgability of a politician reading from an autocue.

Returning to [everything] in the context of a generalizing VEX form, we can begin to understand its concentration in the CH extract. For CH, [everything] is comparable to my coining a neologism such as "vague expression (= VEX)" to collect a common class of phenomena under one label. The differences are a) that CH's collective label can be recycled for other generalized sets, and b) that for her the process is unconscious. Nevertheless, the temporary assignment of [everything] to one class of events in a sub-narrative of the corpus will give it a high concentration within that sub-narrative. The problem for the listener attempting to decode CH's monologue is that her assignment of [everything] as a restricted class constant is implicit and transient. CH offers encouragement such as [you know], but will the listener be motivated enough to guess at her intended class of events, or merely bored ?

If the preceding argument for VEX development is valid, it suggests the mnemonic resonance itself is a process motivated by the general (non-linguistic) processes of cognitive organization. By concentrating clusters of repeating ideas almost mechanically through mnemonic resonance, the mind sets up classes to which can be applied generalizing labels. A generalized label is much more economical to store than a myriad of individual events. However, the recall and linguistic presentation of such general labels depends on the decoder having made similar categorizations. The exceptionally non-specific VEX labels minimize the chances of a category matching conflict between interlocutors, and hence are likely to be socially valuable. Further, they facilitate the rapid generation of linguistic strings since careful, detailed recollection is not required, and automatic adjustment to the listener's nuances of understanding is maximized.



     19. Summary


Mnemonic resonance was defined at the beginning of this paper as "the persistence in memory of an idea after use — in linguistics, of a linguistic element after use — and the heightened availability of this element to a progressing discourse for recall or for evoking a field of linguistic associations". It was suggested that mnemonic resonance, or something like it, was a general property of cognitive processing with actual electrical or chemical expression, usefully captured by the metaphor of a "resonating idea particle" which could evoke sympathetic resonance in related "particles". This form of language had the value of permitting the analysis to transcend normal assumptions about language generation as an autonomous cognitive process.

An extract from the corpus of an idiolect was used as a vehicle for exploring speculations about mnemonic resonance. In particular, the verbs [have], [make] and [know], and the VEX forms [every, everything] were shown to occur in patterns of use that at least arguably supported the mnemonic resonance hypothesis.

Thematic relations, specifically Agent-Verbal_Idea-Theme, depended for their coherence and cohesion in the text upon loose recollections of general classes of items rather than closely defined referents. This observation supported the idea of mnemonic resonance at the level of paradigmatic association. The text also yielded examples of unconscious rhetorical parallelism which seemed best explained by some kind of resonance model.

Elements in the text like [know], which seemed to be relatively independent of the topic/event syntax, invited explanation where their frequency clustered in the extract at levels greatly exceeding the general corpus. It was suggested that the cultural potency of [know-NEG], for example, gave it a high resonance potential (hence persistence) in the text, once invoked. [Know] also offered an excellent example of "branching collocations". It was argued that such collocation sets could rapidly lead to generative gridlock if they were static or cumulative over time. At some point, old collocations had to be forgotten before new sentences could be created. A workable compromise could be fluctuating collocation weightings, supported by mnemonic resonance in recent recall.

The terms [every, everything] were explored as vague expressions which "blurred" mnemonic resonance, but also broadened it, with the possibility of evoking more distant associations. VEX forms were shown to be integral not only to a speaker's cognitive processing, but to pattern matching between the world views of interlocutors. One conclusion from the evidence adduced in the analysis seemed to be that the persistence of a linguistic element in memory or use is sometimes separable from the persistence of what it precisely signifies. VEX forms are indispensable to this kind of abstracted experience. It is a special case of their generalizing role. Unfocused mnemonic resonance  with its power to evoke "sympathetic resonance" in loosely related "idea particles" appeared to make generalization in language, and especially the VEX form, inevitable.



20. Conclusion


It is exceptionally difficult to evaluate the evidence for or against a proposal such as the mnemonic resonance hypothesis. There are so many intervening variables between memory access and speech production. The evidence is further confused by the associationist nature on mnemonic resonance: it may be made manifest not only by precise lexical repetition, but also the occurrence of lexical substitutes, metaphorical forms, and so on. Note that we discuss language, but the genesis of mnemonic resonance is essentially non-linguistic. Given severe handicaps on evidence it is normal (and usually sound) scientific procedure to abandon an untestable hypothesis. My defence for proposing the tantalizing but elusive mnemonic resonance hypothesis is that it leads a linguist down paths that remain otherwise unexplored, and may give a glimpse of vistas that can later be approached on a less reckless tangent.


Thor May
June 1992



Appendix 1: Extract from CH Corpus







Yeah well




1b play 21a round with it 37a 4a all day 38a


1c3b,33b, . 2b


35a When 1dhe was little 39a


I used to 40a


1e he told


"I want a trannie 5a mum for Christmas"


or 1fhis birthday..


I'd buy it


and two weeks later 41a


1g33c it 11x


and 1h33d 6a7a


and 1i run 21b round 42a 4b all 1jhis mates


and 4c43a5b




and 1khe had 33e this 6bboard


with um 4d] 5c




8b .


1l12a fiddling.. with things 44a


I refused to go to 1mhis room in the end 45a


It would electrify y'


ha ha


it was like the one 3c I was gunna take






1nhe had 33f it 4eall 11m


and I was sitting there one 3d night


and 1ohe 10a :


"want a cup of coffee mum"


and the 9b 15a


'n ..,






1p He went into 13a telecommunications*..




but ob*,






An old oil can 1r make 7c into 13b a(?)


you know 36a




and it


16c in my house


has 33h got a little hole bored through 23a the skirting board 6d




1u33j intercommunications


in 16d sort of thing 46a


but 16eeverything you picked up



ha ha.


I had 33k one 3e time




one 3f morning


ha ha ha


17a32a 34a Arthur or Martha


or 17b,34b gunna start raining boxes 47a




Yes + ...yes...yes




Peculiar behaviour of CB radios




They reckon one 3g of these 20a








17c32b how


17d,34c 19b


­or whatever


and someone 48a would drive past you with a 22a


you know 36b


22bwith a CB 18b going


and it'd 10c15x










1v he was onto 16f everything 49a


I mean um..




CH son's pranks in Telecom school




They 30c went to Sydney


silly one 3i (?)..


They 30d had to 33p do


for twelve months schooling






in the PMG.


They 30f blacked out the whole of Sydney 25b


ha ha


I had 33l a 26a


one 3j of those um.. mantle radios 28a


27a the maggie and 4fall in it


It was






27b,33m a a few valves


but the 26b


you know 36c




And 31a30g,24b,25c


1w He asked me could 1xhe have 33n it


and so I said


" Oh I suppose so" 50a


but if I'd knew 32x what for


I wouldn't have 33q let him 51a


Thirty three boys 30a about sixteen to seven...teen years old : can you imagine 52a them 30h?


Anyhow, um,




and they made 7d a jammer.


They 30j had 33o a different 19c


4g all the boys 30b from the Sydney 25d branch right through 23c


posted because




you know 36d


"he's heading this way*"


so righto


they 30k.. finally got about a mile off


you know 36e


and they 30l dismantled it


ha ha..


You never knew 32y what they 30m were going to get up to 53a








Appendix II: Item line references for CH Extract



CORPUS STATISTICS  29/11/91: pages = 323; total word count = 133283; unique words = 6619;characters = 772,000


EXTRACT STATISTICS: total word count = 544


Key: @ = likely formulaic expressions; # = vague form



Line No.


my son; he; him

4,5,6,7,9,11,14,15,16,19,23, 24, 30,32,38, 41,43, 45,49,50, 81,106,110






if [my son] had one of them

4, 6











make, made (up) [ref. No.11]

21,43 | 15,114

8 #

and god knows what



record player



come(ing) out; [10x] first started

32,67,76,118 | 67,91


fitted up; wired up/into; [11x] pulled apart; [Ref.No.7]

30 | 37,42,45 | 14

12 #

he was always



[x]  into [y]



a) he'd have this wired into that; b) where he's had the wires hooked through



talk back; talking to you; [15x] come out [of z] / [ref.10]


16 #

every; everything

47,51 | 37,42,52,81


I didn't know whether [x] was   [ref.34]





19 #

different [Xs]


20 #

these [Xs]



[x] round



with a; with

73, 75 | 5,20,23





down .... Sydney






real old



it had



wireless; radio

104 | 64,96


when ....... out    [ref.35]



boys; they

111,116 | 86,88,91, 93, 105,113,114,115,124, 126, (66,122)


they took [x] down



knew; know ;  [Except "you know"; ref. 36]

(18,22), 59,70 | 109,126


have; had  [=possess]:a,b,e,h,l,m,n,o

4,6,19,95,97,101,106, 115


have; had  [=do]: c,d,f,g,i,j,



have[=experience]: k



have/had to: p



have [=AUX]: q



whether  [ref. 17]



when  [ref.29]


36 @

you know   [ref.32]


37 @

play round with (it)


38 @#

all day


39 @

when (he) was little


40 @

I used to


41 @

(two) weeks later


42 @

run round (x)


43 @

broken down


44 @#

(always) fiddling with things


45 @

in the end


46 @#

sort of thing


47 @

I didn't know whether I was Arthur for Martha, or whether it was gunna start raining boxes


48 #



49 @#

he was onto everything


50 @

oh, I suppose so


51 @

but if I knew what for I wouldn't have let him


52 @

can you imagine


53 @

you never knew what they were going to get up to







Aitchison J  (1987)  Words in the mind : an introduction to the mental lexicon. Oxford: Basil Blackwell


Aitchison N & Chiat S (1981)  "Natural phonology or natural memory? The interaction between phonological processes and recall mechanisms", Language & Speech 24: 311-26


Becker, J (1975 )  "The Phrasal Lexicon", in Schank R & Nash-Webber B (eds) Theoretical Issues in Natural Language Processing, Cambridge, Mass: Bolt Beranek & Newman; 1975:60-63


Chafe W (1980) (ed) The Pear Stories: Cognitive, Cultural & Linguistic Aspects of Narrative Production. NJ:Ablex


Elman J (1988) "Finding structure in time", Centre for Research in Language Technical Report 8801. University of California, San Diego


Gleick, James (1987) Chaos; London: Cardinal/Sphere Books.


Karen L (1990) "Identification of topical entities in discourse: a connectionist approach to attentional mechanisms in language", Connectionist Science, Vol.2, Nos. 1 & 2


May, Thor (1994, 2004) "Generative Oscillation: A Cognitive Model for the Emergence of Language" [unpublished MS]


May, Thor (1994) "Postsupposition and Pastiche Talk: Mediating Order and Chaos in Language". University of Melbourne Working Papers in Linguistics 14 (1994): 31-46


Pawley, Andrew (1985) "On speech formulas and linguistic competence", Lenguas Modernas 12:84-104; Universidad de Chile


Pawley A & Frances Syder (1983) "Two puzzles for linguistic theory : nativelike fluency and nativelike selection", in Language & Communication, (eds) J.Richards & R Schmidt, UK: Longman, pp 191-228


Pawley A (1985) "Lexicalization", in Tannen D & J Alatis (eds) Languages & linguistics : the interdependence of theory, data and application, Georgetown U. Round Table on Languages & Linguistics 1985: 98-120


Wong-Fillmore L (1976) The Second Time Around. Stanford University PhD dissertation.







[i]         A shorter version of this paper was first delivered to the University of Melbourne Linguistics Postgraduate Conference in May 1992. It emerged as part of research for a PhD that was later discontinued (much later the writer completed a PhD in a quite different area). Other material which centered around the same research and has since been put online includes:

          May, Thor (1994, 2004) "Generative Oscillation: A Cognitive Model for the Emergence of Language" [unpublished MS]

          May, Thor (1994) "Postsupposition and Pastiche Talk: Mediating Order and Chaos in Language". University of Melbourne Working Papers in Linguistics 14 (1994): 31-46

[ii]         Idioms have been studied for many years by linguists. The idea that a large percentage of all language may be prefabricated and stored as collocation sets has been harder to substantiate. See Andrew Pawley (1985, 1987) and Joseph Becker (1975) for arguments along these lines. A strand of language acquisition research has also grown from Lilly Wong-Fillmore's (1978) proposal of acquisition "chunks".

[iii]        CH, an Australian woman in her late 50's, talked to me for a week in 1979 about anything and everything in fairly relaxed surroundings. Her speech was picked up clearly by a radio microphone and recorded. The microphone did not pick up other interlocutors clearly or reliably. For the purposes of this analysis that is not important. CH utterly dominated these encounters and remained more or less undistracted by whoever happened to be the audience. She quickly came to ignore the miniature radio microphone. The recorded corpus came to sixteen C90 cassette tapes. Linguistics apart, CH emerged as a woman of great character and compassion. She would be frankly amused by this kind of nit-picking academic paper, though she was aware of the researcher's purpose and gave written permission for use of the corpus. I beg her indulgence.

[iv]        So-called connectionist models in linguistics and cognitive science studies  are broadly sympathetic to ideas of parallel processing. See for example, Elman J (1988) and Karen L (1990).

[v]         Probably the best popular introduction to Chaos Theory is  James Gleick's (1987) book, Chaos.



Department of Literature & Language, University of the South Pacific P.O. Box 1168, Suva, FIJI


The Passionate Skeptic 
[and what this website stands for ..]

Doubt well, do what you can, then let it be. Presidents, priests, wage slaves, hustlers, men and women, kids, we all live by the grace of those we love to despise...


Direct Link to Thor's Aphorisms


©2012 Thor May