CHAPTER ONE:
METATHEORETICAL PREREQUISITES FOR COGNITIVE
LINGUISTICS
1.Introduction
In this chapter I proceed from the assumption that scholars (in whatever field) cannot afford to ignore the question of the validity of the fundamental concepts that they in fact use in their day-to-day professional activities. This is a particularly serious problem in Linguistics, as some Linguists appear to offer strong resistance to accepting the methodological consequences of lucid conceptual analysis.
I demonstrate that Autonomist Linguistics (or those branches of it that make their underlying assumptions explicit to any degree) is essentially metaphysical in nature. This is shown via an analysis of the confusion surrounding the terms GRAMMATICALITY, ACCEPTABILITY, INTUITIONS, and COMPETENCE-PERFORMANCE DISTINCTION. Concrete and positive proposals are made to overcome the problems that are discussed.
The relationship (or otherwise) of grammar to psychology comes up more than once, but principally in the context of my discussion of the Competence-Performance Distinction (henceforth CPD). The CPD is analysed into sixteen or seventeen versions present in the Generative literature. These are shown to be logically distinct from each other (with one possible exception).
The main aim is to make it clear that linguists are free to accept or reject any given CPD, without this entailing that they need necessarily accept or reject any of the others (again, with one possible exception). I see this as an essential first step, before there can be informed debate as to the validity of the various idealisations involved. Idealisations are of course essential in linguistics as in other fields -- but so is intellectual clarity as to the exact nature of the idealisations that are being made. As a secondary and distinct facet of this discussion, I make concise suggestions as to which particular subsets of the CPDs I would myself accept and (respectively) reject. These suggestions are put forward tentatively.
The notion of descriptive grammaticality is replaced by comprehensibility in the case of language-processing, and by a combination of presumed comprehensibility and normative grammaticality in the case of language-production. Intuitions as to descriptive grammaticality are replaced by intuitions as to comprehensibility. The crucial difference is that nativespeakers routinely make judgements about comprehensibility as part of their normal language-processing, whereas intuitions about descriptive grammaticality are artificial and incoherent in theory and unreliable in practice.
2.Grammaticality and intuitions
2.1 Background remarks
I will discuss the notions grammaticality and intuitions to some extent separately, and to some extent conjointly -- as seems appropriate. The former notion is, of course, methodologically dependent on the latter in Generative Grammar: native-speaker intuitions are the means by which grammaticality or its absence is ascertained -- according to most Generative sources, at any rate.
Snow and Meijer (1977)
argue that ... syntactic intuitions are secondary to other kinds of language behaviour and even to other kinds of linguistic intuitions in three senses: developmentally, pragmatically and methodologically (ibid,163).
They claim that children learn to use language before they have any intuitions about it -- and, even then, they have semantic intuitions long before they develop syntactic intuitions. Pragmatically, the main use of language,they say, is communicative, and trying to decide whether a sentence is syntactically deviant is secondary to the communicative purpose or role. They also claim that intuitions are methodologically secondary vis a vis corpuses/texts.
As far as formal approaches to linguistic semantics are concerned, it is noteworthy that Spencer-Smith(1987), which I take to be representative (in this respect) of Discourse Representation Theory in general, nowhere mentions grammaticality. Understanding seems to be his central concern: what it is that is necessary for a hearer/reader to understand a certain segment of natural language. Since the central innovation of Discourse Representation Theory consists of denying the semantic relevance of the sentence boundary, it is hard to imagine that Spencer-Smith would want to retain the term grammaticality for intra-sentential phenomena, while restricting understanding for the inter-sentential/discourse/textual realm.
He gives no metatheoretical inkling of why he prefers to talk in terms of understanding, rather than grammaticality, or even acceptability. But I share his preference to some extent -- as I will explain.
Spencer-Smith's preference for understanding over grammaticality is consistent with my interpretation of Discourse Representation Structures (DRS's) as thinly-disguised psychological models of some aspects of the human mind -- peculiar only through the fact that they have been developed by philosophers, rather than psychologists. Spencer-Smith's preference is consistent with a psychological model for the reason that no one would claim that the goal of the languageunderstanding process is to isolate just those utterances or sentences that are grammatical, ambiguous, or pairwise synonymous, etc..
Historically speaking, the concentration on grammaticality may have been (and may still be) appropriate in terms of the objectives of prescriptive or language-instruction textbooks, where such terms as correct, etc., obviously imply some normative notion of grammaticality. However, if we aim to explain directly the language-user's actual behaviour, then grammaticality must be said to be irrelevant to the hearer/reader in the first instance, and it can serve only as a crude set of guiding parameters to the speaker -- Sociolinguistic considerations aside, of course.
In Autonomous Linguistics, of course, the language-user's behaviour ("performance") is described only indirectly -- in a model incorporating various idealisations (see below), where grammaticality is a descriptive notion, rather than a normative one.
It has long been (even popularly) fashionable to differentiate normative from descriptive statements about the grammaticality or otherwise of sentences. In effect, this distinction has served to cut the ground from under the feet of lay or journalistic normativists, as they could now be seen as expressing mere personal preferences, as opposed to authoritative pronouncements that issued somehow from the very nature of the language itself. As was mentioned above, however, there is a distinct tendency detectable in many non-Linguists (e.g. philosophers and psychologists) studying language: these scholars tend to concentrate on what is understood or comprehensible -- as far as processing (not production) is concerned.
From a Cognitive perspective it appears to me that, of these three possible objects of study (norms, inherent grammaticality, and comprehensibility), the middle one [the concept that there is some kind of inherent grammaticality, such that it is either present or absent (or even present or absent to some degree) in a given sample of language] is not essential to linguistic theory.
Consider an extremely unusual sentence such as:
(1) Up ran stream dog the the.
Such a sentence would presumably be considered to be objectively ungrammatical as well as unacceptable in Autonomist Linguistics.
In Cognitive Linguistics, as I see it, this would be considered to be normatively ungrammatical, though perhaps marginally comprehensible. There is a parallel with other normgoverned behaviour, such as table manners. Table manners are a set of norms that govern eating, drinking, and associated behaviour at table. One aim is a high degree of efficiency in getting the food and drink into the alimentary canal -- but norms in this area do not favour any and all efficient (or even the most efficient) methods of achieving this aim. Likewise, the fact that (1) is more or less comprehensible is not enough to give it a pass-mark as far as the norms of grammaticality are concerned. Langacker (1987:36-37,60) makes some remarks which could be considered compatible with the above point of view. Consider also:
(2) Dog ran up stream.
This is much more comprehensible than (1) -- yet it comes not much closer to meeting grammatical norms,. Similarly, the following normatively grammatical sentence receives a pass-mark, despite not being much more comprehensible than (2):
(3) The dog ran up the stream.
I discuss the Descriptive Grammaticality-Intuitions (DGI's) used in Generative Grammar below. As far as Cognitive Linguistics is concerned, however, I would like to suggest that intuitions are valid if and only if they are intuitions of comprehensibility or of normative grammaticality. This is because both occur as part of normal language behaviour. If we find a given sentence incomprehensible in some normal linguistic situation, then we are most certainly aware of it. We are also latently aware of the converse: the default case where we find a sentence comprehensible. Likewise for degrees of comprehensibility. And the only type of grammaticality that naive informants can reliably report on is precisely the normative (prescriptive) kind.
One place where normative grammaticality plays an important role is in language-teaching. Teachers of languages necessarily set norms, otherwise they would have no means of knowing what rules to teach their students, or what to mark right or wrong in their work. Of course, they assume that their norms reflect in some way the objective reality of rules of the language involved. But in practice they seldom, if ever, read objective linguistic studies of that language -- relying on textbooks which have a tacit normative undercurrent which relates to the fact that the variety chosen for teaching is the standard (this word itself is equivalent to norm) dialect or variety.
The teacher is not actually teaching the rules that describe the linguistic behaviour of any particular group of speakers -- he is in fact actively setting standards for his students -- the same standards which native speakers of the language measure themselves against if they aspire to speak the standard dialect/variety of their language. The tradition of studying and teaching grammar separately from the roles of speaker, writer, listener, and reader seems to have been carried over relatively uncritically from earlier applied linguistics into modern theoretical linguistics.
2.2 Intuitionising
Let us now look at the theory and practice of intuitionising with respect to descriptive grammaticality in Generative Grammar. Although the literature as a whole gives the impression that, in Generative Grammar, native speakers have intuitions about (amongst other things) the grammaticality of sentences, and that these intuitive judgements comprise the data against which proposed grammars are matched for adequacy, Newmeyer (1983) disagrees. Though he does not explain it this way, it is clear that his stance is motivated essentially by the problem that native speakers cannot be relied upon to give grammaticality-judgements that can be represented as being a direct reflection of their Linguistic Competence. As he says (p.51):
... speakers are ... prone to give ... judgements that depart both from their linguistic competence and from their everyday use of language. Thus people will find sentences unacceptable if they contain taboo words or nonstandard forms, and so on even though they commonly use such sentences. Conversely, speakers have been known to give favourable judgements to sentences that are demonstrably not part of their (nonstandard) dialects or that belong to stylistic registers they do not control.
Given that a linguistically naive native speaker cannot be expected to distinguish between grammaticality and acceptability (however these may be theoretically defined) in his judgements, a Generativist has to choose whether to consider his judgements to be judgements of grammaticality or of acceptability. For the reasons given in the quotation above, Newmeyer has chosen to regard them as judgements of acceptability.
The problem now arises of what actual data is available to the Generativist as a check on the non-imaginary status of all the theoretical apparatus that makes up his elaborate grammars and theories. Since Grammaticality, according to Newmeyer, is now 'not directly accessible to the intuitions of the speaker of the language', and since Generative Grammars are set up to generate all and only the grammatical sentences of a given language, we are in a good position to make the formal and explicit claim that Generative Grammars are purely metaphysical in nature, as they are not, in principle or in practice, matched against instrumental or sensory data in the usual "scientific" way. Points very similar to this have been made, on other grounds, by other authors. Now that Cognitive Linguistics exists as a plausible alternative to Autonomist Linguistics, it is timely and appropriate that such arguments be taken seriously, and that linguistic theory and practice be made consistent with the conclusions that one must draw from them.
Whitaker (1974), for example, refers to Fodor and Garrett (1966), which he criticises for equivocation on the central issue of the "psychological reality" of grammars. He also cites the following passage:
It should be emphasized that, in showing that a predicted complexity order fails to obtain, one has not shown that the grammar is disconfirmed. A grammar is simply an axiomatic representation of an infinite set of structural descriptions, and the internal evidence in favor of the structural descriptions modern grammars generate is so strong that it is difficult to imagine their succumbing to any purely experimental disconfirmation. Rather, one would best interpret negative data as showing that an acceptable theory of the relation between competence and performance models will have to represent that relation as abstract, the degree of abstractness being proportional to the failure of formal features of derivation to correspond to performance variables.
It is easy to satirise this passage, from a Philosophy of Science point of view, but Whitaker (op.cit.) limits himself to saying:" On the face of it, Fodor and Garrett's stance is untenable, if it is still agreed that linguistic theory has empirical content." (ibid, 79). In addition, we can now, with hindsight, totally reject the only argument for Autonomist Linguistics that Fodor and Garrett put forward: namely, that the internal evidence for "modern" structural descriptions is strong. In fact, the history of Generative Grammar has been marked with schisms and internal disagreements on every type of issue from the most abstract and fundamental to the most detailed aspect of structural descriptions. This is proof that the internal evidence is far from sufficient to achieve consensus -- just as in other metaphysical systems, such as religions.
Newmeyer, like some other Generativists, takes the position that the grammar of a given language will be developed by Linguists on the basis of those data which are noncontroversial, and that this grammar will then itself decide which of the controversial examples are grammatical, and which are ungrammatical.
This position seems completely incoherent: If the native-speaker can make judgements only as to acceptability, rather than grammaticality, then there is no way -- even in principle -- that a core of noncontroversially grammatical sentences can be built up from scratch. All that can be agreed upon, at best, is a set of noncontroversially acceptable sentences. But, as we have seen, the grammar is meant to account for all and only the grammatical, not the acceptable sentences. So this core set of data is totally irrelevant to the task of writing a Generative Grammar.
Other authors, such as Bever (1974a), have come close to pointing out this fundamental and debilitating flaw -- but only close. He talks of a "basic set of pretheoretically determined cases" which may be of heuristic value in the infancy of any given science. But the question here is: "cases of what ?" As we saw above, investigating sets of acceptable/unacceptable sentences is of no help if what we are supposed to be doing is investigating cases of grammatical/ungrammatical sentences.
As far as the use of native-speaker intuitions per se is concerned, it is noticeable how frequently this use is attacked, and how rarely generativists attempt to defend it. In fact, apart from Newmeyer (op.cit.), the only such attempt that I am aware of is Valian (1982). Her basic approach is to try to equate intuitionising with experimentation as much as possible. She says that intuitive judgements 'may be thought of as informal experiments', but she does not explain at all why intuitionising is to be considered more similar to experiments (however 'informal') than to any other natural phenomenon such as (to mention some possibilities at random) dreams, hallucinations, or psychedelic highs . The term "experiment" is, of course, sociologically prestigious, and Valian gives the sceptical reader no reason not to suspect that that is the sole reason she has for drawing the comparison between the two terms.
Valian also claims that "formal psycholinguistic experiments (including observational studies) test performance models directly, competence models only very indirectly"(ibid, 179), whereas intuitionising, however fallible, tests competence models more directly. However, she provides absolutely no evidence for this claim, and we have already seen that Newmeyer (for one) considers that intuitions reflects acceptability (i.e.performance), rather than grammaticality (competence).
Valian reviews some of the criticisms that have been made of intuitionising, and then writes:
All experimental procedures are fallible. They all yield data which are subject to more than one interpretation. There is always some datum in conflict with other data. The same procedure may yield one result in the hands of a scientist with one theoretical orientation and another in the hands of a scientist of a different orientation. Some experimenters get results when others do not.(ibid,184)
Valian thus sees intuitionising and experimentation as being equally liable to error. However, there are two main differences which she does not mention: first, such problems are much more serious and more frequent with intuitions than they are with experiments. Secondly, experimental procedures can be externalised (written up) so that intersubjective agreement can often be achieved by having others repeat the experiment as often as seems necessary until a convincing picture emerges, by varying the experimental method,by carrying out additional experiments specifically to test rival interpretations of the first experiment, and so on, whereas none of these options is available to intuitionisers, as intuitionising is a private, subjective matter.
Another claim that Valian makes is that people vary in their skill at intuitionising, and that training can increase this skill, so that professional linguists (or perhaps only Generativists who agree with Valian on a case-by-case basis ?) are usually better intuitionisers than lay people. She does not propose any criterion by which one could distinguish 'good' from 'bad' intuitionisers.
The following passage includes a proposal similar to that made by Valian:
...problems with native speaker intuitions can be countered by saying that there are privileged native speakers, viz. the linguists: linguists are then considered better judges of grammaticality than laymen, precisely because linguists are better aware of the theoretical issues and distinctions involved in assessing grammaticality (Dirk Geeraerts, personal communication).
This argument is untenable on methodological grounds. The native speaker intuitions are the principal data of Autonomist Linguistics. If the researchers are the ultimate authority on both the data and the theories that are constructed on the basis of the data, then Linguistics denies itself the possibility of intersubjective agreement or "objective" statement. If two Linguists differ as to the grammaticality or acceptability of a given sentence or utterance, there is no principled way of deciding who is right. There is no sense of the word "scientific" broad enough to include a discipline with this eccentric attitude to data.
The following passage proposes (facetiously) a link to punning:
The alleged ability of speakers of a language to distinguish between 'grammatical' and 'ungrammatical' strings of words is about as rare and perverse as the ability to construct puns, an ability to which I believe it is closely related (McCawley 1982,78)
There is also another problem with intuitionising (see Chapter Six) which I call the Real Estate Institute approach to examples. This problem affects intuitions as to comprehensibility just as much as grammaticality intuitions. The problem is, quite simply, that a given example will tend to appear more grammatical/comprehensible in the context of a clearly ungrammatical/incomprehensible example that it is paired with. Conversely, if it is paired with a clearly grammatical/comprehensible example, then it will tend to appear less Comprehensible/Grammatical than in the former context.
This is only one of a number of methodological problems which have been somewhat neglected in the Autonomist literature. Nevertheless, a science is ultimately only as sound as its methodology. By that criterion, Autonomist linguistics is very unsound indeed.
2.3 Proposals
Having proposed the replacement of Descriptive-Grammaticality-Intuitions (DGI's) with Normative-Grammaticality-Intuitions (NGI's) and Comprehensibility-Intuitions (CI's), it is incumbent on me to explain my view of the logical, evidential, and scientific status of NGI's and CI's, so that the reader may judge whether they avoid most of the above-mentioned drawbacks of DGI's.
Ringen (1977) is a praiseworthy study of the scientific status of DGI's. He concludes, in short, that there is a true parallel between native-speaker reports of DGI's and first-person reports of sensations in Psychophysics. In relation to the latter, he claims that:
A person's reports of (for example) his own pains can be shown by others to be false or mistaken. The speaker can be shown to be lying, to have misspoken, or to be incompetent in using parts of the language (or, perhaps, even to be deceiving himself). Since there are objective ways of determining these things, speakers do not have "privileged access" to their mental states, and first-person reports about which a speaker has final epistemic authority are not subject to the charge that their veracity is in principle uncheckable.
Where I differ from Ringen (op.cit.) is in the applicability of the objective-verifiability characteristic, which he mentions above, to DGI's, NGI's and CI's. Even in the case of reports of pains, I would myself assume that the most the experimenter could be reasonably sure about was that in some cases a report was unreliable for some reason (lying, etc.).
The converse, however, is not the case: he could never be quite sure that the speaker was NOT lying, or otherwise in error. If there were, in fact, adequate objective tests for such things, then it seems to me that the use of Psychophysical data would long ago have become redundant. Whether or not I am correct in this point, I feel it is totally implausible to argue that there are objective tests for the reliability of a particular instance of a Native Speaker's NGI, DGI, or CI. I certainly have never heard of any, and can imagine none. I leave it to the reader to either agree or come up with a refutation.
This problem, however, is not so serious for my approach as it is for Ringen, who writes:
If linguistic intuitions are treated as logically private entities, assessment of the reliability of intuitive data is, in principle, impossible. If there is to be any hope of developing a principled methodology for assessing data concerning linguistic intuition, linguists must reject any view according to which linguistic intuitions are mental entities to which each speaker/informant has privileged access.
I see no need to "develop a principled methodology for assessing data concerning linguistic intuition". When a physical scientist looks at an instrument reading, what principled methodology exists to check if his eyes are deceiving him ? What principled methodology exists to check if the instrument is malfunctioning on that one particular occasion ? When Sociological surveys are carried out, what principled methodology exists to check if the responses are accurate or sincere ? And even if such methodologies exist, how often are they actually implemented ? The most that happens is that some (seemingly arbitrary) statistical allowance for error is built into the published results. It is obvious that some things have to be taken on trust in science, on the assumption that repeating experiments with different experimenters, subjects and equipment will tend to iron out freak results in the long run.
I do therefore consider that informants have essentially privileged access to their DGI's, NGI's or CI's. In the case of CI's, comprehensibility is not absolute, but relative, for any given informant; and ultimately comprehensibility is to be understood as a statistical measure across linguistic populations.
Thus I suggest that (relative) comprehensibility be intuited by informants as a means of providing data in the first instance. In case of dispute about data, however, the final authority would be a statistical survey of the relevant linguistic population, as in some forms of Sociolinguistics. The same applies to NGI's (mutandis mutatis), although I see NGI's as peripheral to Cognitive Linguistics per se.
DGI's I see as devoid of theoretical interest, though people may of course feel that they have them. It should be noted that most of the disagreements I expressed with Valian (op.cit.) (above) apply only to DGI's, rather than to CI's and NGI's as I have described them in the context of Cognitive Linguistics.
Carroll, Bever, and Pollack (1981) reports experiments which show that intuitionising is an actual Psychological process ("performance") upon which nonlinguistic factors can exert influence in systematic ways. Thus it is illegitimate to claim, as Valian (op.cit.) does, that DGI's mirror Competence directly. Carroll et al.(ibid) were studying intuitions about sentence relatedness, rather than intuitions about Descriptive Grammaticality, Normative Grammaticality, or Comprehensibility, but I think it is likely that similar factors affect these other types of intuitions, and that they should be thoroughly investigated, as they have to do with the reliability of data which are fundamental to both Autonomist and Cognitive Linguistics.
3.Competences and Performances
The central role that the Competence-Performance Distinction (CPD) plays in Autonomist Linguistics can be seen from the statement that:
a grammar of a language purports to be a description of the ideal speaker-hearer's intrinsic competence (Chomsky 1965,4).
And "competence" is defined in terms of its opposition to "performance".
3.1 Factoring out
For a statement of the CPD, we will refer to what is perhaps the Locus Classicus:
Linguistic theory is concerned primarily with an ideal speaker-listener, in a completely homogeneous speechcommunity, who knows its language perfectly and is unaffected by such grammatically irrelevant conditions as memory limitations, distractions, shifts of attention and interest, and errors (random or characteristic) in applying his knowledge of the language in actual performance.... We thus make a fundamental distinction between Competence (the speaker-hearer's knowledge of his language) and Performance (the actual use of language in concrete situations). (Chomsky 1965,3-4)
It seems to me that Chomsky here is making eight separate distinctions, all of which must be termed, it seems, Competence-Performance Distinctions. Yet his summary at the end of the quoted passage includes only three of them. Here are the eight:
C1/P1: abstracting away from the speaker-role
C2/P2: abstracting away from the listener-role
C3/P3: abstracting away from the writer-role (by implication,
though not actually stated)
C4/P4: abstracting away from the reader-role (by implication,
likewise)
C5/P5: abstracting away from any deficiencies in a person's
knowledge of his language
C6/P6: abstracting away from the characteristics of memory
C7/P7: abstracting away from the effects of variable
attention and concentration
C8/P8: abstracting away from all sources of error (it is not
clear how many exist apart from those covered by other
C/P distinctions)
Only the last three are actually related to the contrast that Chomsky draws at the end of the passage between knowledge of language in the abstract and the use of language in concrete situations.
There obviously are abstract and constant factors involved in the separate speaker/listener/reader/writer roles, quite apart from factors which may intervene (and perhaps cause errors) in concrete situations. For example, there are systematic differences between the speaker-role and the listenerrole, for example. But, within the speaker-role itself, a given situation may be more or less stressful, resulting in a given amount of "performance error", such as stuttering.
Likewise, a person's knowledge of a language (whether deficient or not, and however deficient it may be) is totally independent of how this knowledge may be applied in concrete situations (presence or absence of stuttering, reduction of abstract vocabulary when speaking to less educated people, and so on ). Such confusions have long undermined Generative thinking.
The fact that the CPD has been used ambiguously, as an umbrella-term, is something that has been pointed out by various writers (principally George Lakoff -- see below). However, I feel I am able to make two contributions to this discussion.
One is to argue that the full extent and internal structure of this umbrella have not been hitherto revealed. My second contribution is to point out that the term Competence has been implicitly held constant -- even when it has been realised that there is more than one type of 'Performance' which it contrasts with. In other words, it seems that such Linguists (e.g. Lakoff, see below) as have subdivided the CPD into various subdistinctions on the basis of the various uses of the term Performance have not taken the further step of questioning whether the Competence which contrasts with a different Performance in each case remains the same notion of Competence throughout. Obviously, it cannot be the same notion, as the two terms are defined in opposition to each other. Specifically, we have already seen that the Performances that emerge on the basis of the first five distinctions must (at the very least) be once more subdivided into Competence and Performance on the basis of the next three distinctions, which relate to concrete situations. It is obviously impossible to argue that two uses of Competence are identical when one is included in Performance, as defined by the other use. Likewise, it is impossible to argue that two uses of Performance are identical, when one is a subset of the other, and the residue is non-null. This fact becomes more and more obvious, the more distinct CPD's one culls from the literature. Here are some more:
C9/P9: abstracting away from hesitation phenomena
C10/P10: abstracting away from the linguistic
context/cotext/discourse of the sentence
C11/P11: abstracting away from the nonlinguistic context of
the sentence
C12/P12: abstracting away from the reasons why people
say/write what they do
C13/P13: abstracting away from Sociolinguistic factors
C14/P14: abstracting away from nonlinguistic mental faculties
(assuming such boundaries can be drawn) in general
C15/P15: knowledge of grammar, as opposed to the use of some
parts of that knowledge on specific occasions (this
may be identical to C12/P12)
In addition, Lakoff (1973) isolates the following distinction:
C16/P16: abstracting away from stylistic reordering rules
This distinction comes from Chomsky (1965,126-7), where he says:
... it should be emphasized that grammatical transformations do not seem to be an appropriate device for expressing the full range of possibilities for stylistic inversion. It seems, rather, that there are several underlying generalizations that determine when such reordering is permissible, and what its semantic functions are.
Chomsky does not spell out exactly why he thinks transformations are not appropriate here, but it can be assumed that he means that such a treatment would not capture the appropriate generalisations in an optimally simple manner. He suggests, in fact, that the rules involved are probably Performance rules. Lakoff points out that there is a borderline problem here, as well, and that Chomsky has no principled way of distinguishing C16 from P16 -- particularly as these stylistic inversion rules make reference to such categories as lexical item, noun, clause, adverb, etc., just as transformations do. It is important to note that Chomsky's problems here are caused by the fact that he insists on a CPD so that his theories can be autonomous of the findings of Cognitive Psychologists. Since he excuses himself from the goal of empirical adequacy in this sense, all he is left with is the goal of observational adequacy (which is not sufficiently challenging or difficult to achieve),on the one hand, and the goal of maximising simplicity, on the other. Simplicity has therefore assumed a disproportionate importance in his writings -- at least rhetorically.
It is this essentially unavoidable preoccupation with simplicity that disinclined Chomsky -- one must assume -- to want to introduce into the grammar a special rule-type just to cope with stylistic reordering, so he shoved it aside into the longfamous Pragmatic Waste-Basket (Bar-Hillel (1971)) , where, quite possibly, it might create complications commensurate with the simplification of the grammar that Chomsky achieved by excluding it from there.
Lakoff (1982) analyses the CPD lucidly without referring to it by name. He contrasts what the calls the autonomous (principally = Generative) with the experiential view of 'how a theory of language ought to fit into a general theory of cognition, human development, and social interaction' (page 145). He then summarises seven sets of research results which refute various C/P distinctions, including the last one I will list:
C17/P17: abstracting away from non-literal meanings.
Some of the above distinctions are relevant only to the writer/speaker. The following is a tentative graphic portrayal of how these various CPD's are interrelated. This is not claimed to be the only possible, or even necessarily the best way of arranging the CPD's into a hierarchy. It should be noted that many of the distinctions are only vaguely described in the literature, and that some distinctions could be subsumed under others. The diagram should be read as a hierarchy, where the Performance part of each non-terminal CPD is subdivided again, at the next level down, into Competence and Performance, as per the relevant distinctions listed above.
COMPETENCE-PERFORMANCE HIERARCHY
C/P 11,12 and 13
C/P 10, 16, and 17
C/P 8
C/P 5
C/P 15
C/P 1-4
C/P 6, 7, 8 (again) 9 and 14
This can be read as meaning that the speaker/writer has reasons for saying/writing what he does, and that these (C/P12) count as Performance vis-a-vis what he might have said (Competence). At more or less this level, too, the nonlinguistic context (C/P11) and Sociolinguistic factors (C/P13) also influence what the speaker/writer produces or the reader/listener understands (Performance).
At the second level of the hierachy, we have the more arguably "linguistic" of the CPD's -- C/P10 (the sentence vs its linguistic context), C/P16 (stylistic reordering), and C/P17 (literal vs non-literal meanings). This is the stage at which literal meanings and sentence grammar (Competence) are affected by these supposedly "Performance" factors.
At this point I have inserted C/P 8 (sources of error), which is by far the vaguest of the CPD's. It also occurs lower down in the hierarchy, and, in fact, there may also be other positions in which it could justifiably be inserted.
Once any errors have been committed, the Performance part of C/P8 becomes subject to another source of error: C/P5 (deficiencies of knowledge of the language). Obviously, C/P8 could be taken as including C/P5, which need not really be listed separately.
The Performance part of C/P5 (i.e. the person's actual extent of knowledge of his language) is then subject to C/P15 (the contrast between his knowledge and how he actually applies it in concrete situations).
This application (Performance) is then subdivided into C/P 1,2,3, or 4, according to what role the person in question is playing in the linguistic situation.
At this point (and perhaps elsewhere as well), nonlinguistic aspects of Mind play their part (C/P14). Related, possibly subsumed factors include memory (C/P6), attention and concentration (C/P7), and hesitation (C/P9). Hesitation may have many causes, and thus could appear in more than one place in this hierarchy. C/P8 reappears here, as all these factors are sources of error.
3.2 Some Generative proposals
Smith and Wilson (1979) provide an interesting discussion of the CPD from a Generativist perspective. One point that they make is that:
...along with the distinction between competence and performance, grammaticality and acceptability, comes a parallel distinction between sentence and utterance. (p.45)
This three-way terminological parallelism is woth noting, as it seems to me that, when we investigate individual C/P pairs, we are also implicitly investigating a separate grammaticality/acceptability distinction, and a separate sentence/utterance distinction as well -- in many cases, at least.
Smith and Wilson (ibid) also construct a straw horse, if that is a fair characterisation of it, from which I feel the need to distance myself:
...it might be possible to argue that there is not a distinction between competence models and performance models, but merely between more and less abstract rules of performance, each of which has its part to play in the full production and understanding of utterances... we have never seen a fully coherent outline of a theory based on a single notion of performance, which could account in an adequate way for the facts which can be accounted for in terms of the competence-performance distinction. (p.48)
I certainly do not wish to argue that there is no distinction between Competence and Performance -- my point is that there are several distinctions being labelled by those names, and that they probably differ in the extent to which a linguist might want to employ them as tools in his work. The closest I come to this (somewhat vaguely described) straw horse is to say that the first four C/P pairs should be dispensed with, so that there is no Abstract Grammar which is separate from listening, speaking, reading and writing strategies. This is another aspect of my remarks (above) about replacing Descriptive Grammaticality with Comprehensibility and Normative Grammaticality. However, it would be logically possible to agree with me as far as the first four C/P pairs are concerned, while differing from me in whatever view I take of the validity of other C/P pairs. I sympathise with the above authors' complaint that they had hitherto never seen a coherent Cognitive Linguisics alternative to the CPD, but I hope that since that time the issues have been gradually sharpening and the alternatives becoming more coherent.
Smith (1979) also contains an interesting and relatively detailed discussion of the CPD, particularly in relation to Lakoff and Thompson (1975), which includes the example:
(4) The aardvark was given a bagel by Irving,
and also the statement,
Grammars are just collections of strategies for understanding and producing sentences.
Smith comments as follows:
On hearing "the aardvark" the listener will probably form the "current semantic hypothesis" that it is the Subject of the sentence being uttered: after hearing "was given" this hypothesis will be modified to make "the aardvark" into the Object of the sentence, and after meeting another NP, "the bagel", this hypothesis will in turn be modified so that "the aardvark" is indirect object, and "the bagel" is Direct Object of the sentence.... The strategies will (ultimately) be more sophisticated so that the different hypotheses will be engendered by NP's containing different semantic features or having different pragmatic associations, so that "the aardvark was given..." and "the biscuit was given...." will not be treated identically. It is worth noting indeed that these semantic and pragmatic hypotheses will need to be extremely, perhaps impossibly, powerful if the hearer is going to distinguish satisfactorily among:
(5) The cognitive grammarians were given most of their ideas by their students
(6) The cognitive grammarians were given to the enemy for intellectual dissection
(7) The cognitive grammarians were given to extravagant claims
where the leftmost, identical NP is respectively indirect object, direct object, and subject. (ibid, 52-53)
Smith's argument is based on a misapprehension: The semantic and pragmatic hypotheses referred to are not intended to replace, but only to supplement the strategies. Thus the strategies will apply in the normal way (successfully) to (5-7) (above), but, obviously, no semantic or pragmatic hypotheses will be any use in speeding the process up.
Smith does make what seems to me to be a valid point, however, when he criticises Lakoff and Thompson for aiming at a set of rules which are neutral as between speaker and hearer, and yet producing rules that are one-sidedly hearer-strategies. Another valid point he makes is when he says that Lakoff and Thompson's distinction between "grammar" and "production mechanisms" is just (one component of) the CPD in disguise. It will, I hope, be by now clear to the reader that I would not see Cognitive Linguistics as being forced to adopt the two positions that Smith rightly criticises: I myself see no need for rules which are neutral as between speaker and hearer (except in normative grammars), and I would also not subscribe to any distinction between "grammar" and "production mechanisms".
Another of Smith's misunderstandings is worth pointing out:
... in any account of perceptual strategies some provision for abstract linguistic rules will always be necessary to account for the decoding we are able to carry out on those occasions when our first approximation -- e.g. that NP-VP-NP corresponds to agent-verb-object -- fails. Only if such rules are then claimed themselves to be or to correspond to perceptual strategies can we cover the whole of speech perception by performance strategies, and without independent justification such a claim trivialises the content of all performance strategies. (ibid, 54)
Here Smith has turned the logic of the argument upsidedown: Autonomist abstract linguistic rules exist in parallel to perceptual strategies which would have to be described (ultimately), once Competence-Models are embedded into Performance-Models. What independent justification is needed for is this Autonomist parallelism and redundancy. It is a trivial, rather than a substantive matter whether a given linguistic generalisation is worded as an abstract linguistic rule or as a perceptual strategy. I would see Cognitive Linguistics as having the advantage that it would need only one of these two wordings (i.e. the latter). This would need no independent justification, although the Autonomist Linguistics approach to this issue in fact does require such justification -- as explained above.
3.3. Discussion
I would like to outline my views as to which CPD's are useful and justified (if any), and (more generally) what conclusions might be drawn from the above discussion with respect to the goals and assumptions of a truly Cognitive linguistics. These remarks are additional to my main thesis, which is that the CPD is not unitary, but hybrid, and should be examined as such.
I repeat the complete list of CPD's here for easy reference:
C1/P1: abstracting away from the speaker-role
C2/P2: abstracting away from the listener-role
C3/P3: abstracting away from the writer-role
C4/P4: abstracting away from the reader-role
C5/P5: abstracting away from any deficiencies in a person's knowledge of his language
C6/P6: abstracting away from the characteristics of memory
C7/P7: abstracting away from the effects of variable
attention and concentration
C8/P8: abstracting away from all sources of error (it is not clear how many exist apart from those covered by other C/P distinctions)
C9/P9: abstracting away from hesitation phenomena
C10/P10: abstracting away from the linguistic context/cotext/discourse of the sentence
C11/P11: abstracting away from the nonlinguistic context of the sentence
C12/P12: abstracting away from the reasons why people say/write what they do
C13/P13: abstracting away from Sociolinguistic factors
C14/P14: abstracting away from nonlinguistic mental faculties (assuming such boundaries can be drawn) in general
C15/P15: knowledge of grammar, as opposed to the use of some parts of that knowledge on specific occasions (this may be identical to C12/P12)
C16/P16: abstracting away from stylistic reordering rules
C17/P17: abstracting away from non-literal meanings.
I deny the need for C/P 1-4 as far as linguistic description is concerned. "Abstract Grammar", as distinct from models of speakers, readers, writers, and hearers/listeners, is a theoretical necessity only for application such as translation and language-teaching. Duplication can be avoided, and more insightful conclusions drawn if empirical linguists study these roles separately in the first instance. However, I can imagine that a point might be reached where aspects of these four separate lines of research might well converge. This convergence would only be partial, in my view.
I also deny the need for distinctions C/P6, C/P10, C/P14, and C/P16. That is, I feel that the mind is not so "modular" that language can be usefully separated from the functioning of other parts of the mind, such as the various forms of memory. Too many generalisations apply both within and between sentences for the distinction between a sentence and its linguistic context to have much theoretical value. And we have already discussed the nature of stylistic reordering (above).
The remainder of the present book contains many specific examples of the crucial role that such distinctions play in Autonomist linguistic descriptions. Many examples are also provided of alternative descriptions that do not involve such distinctions.
As to the remaining nine CPD's, I have no strong feeling that they need (or, for that matter, need not) be abolished.
IT IS IMPORTANT TO KEEP TWO ISSUES SEPARATE AND DISTINCT:
1. THE NEED TO APPRECIATE THE HYBRID NATURE OF THE CPD.
2. THE QUESTION OF WHICH CPD'S ARE ACTUALLY DEFENSIBLE.
4. Conclusion
Dretske (1974) outlines lucidly the Mentalist (including Cognitivist) approach to the study of language, but rejects it on the grounds that it is not a true characterisation of what (most) Descriptive Linguists are actually doing in practice.
From a Cognitivist perspective, I would prefer to conclude that, if theory and practice differ, then so much the worse for the practice. In other words, if the type of research that Descriptive Linguists (in the main) are doing does not mesh well with the notion that Linguistics is a branch of Cognitive Psychology, then my solution would be to propose changes to the methodology of the working Descriptive Linguist so that it does so mesh.
One point that Dretske (op.cit.) makes is that Linguists write about (behavioural) rules, rather than (scientific) laws.
Grammars ... are systems of rules, not laws. But rules are the sort of thing we expect to hear from moralists, referees, and etiquette columnists, not scientists. Rules do not purport to tell us what will happen all the time, not even most of the time - much less what must happen. One can break the rules; one can't violate the laws. And this, according to some, is why laws give us explanations in a way that rules never do. (Dretske, ibid, 26).
Dretske seems to confuse two distinct kinds of rules:
1. prescriptive/normative rules (e.g. etiquette);
2. descriptive/generative rules.
The point is that generative linguistic rules are supposed to form a Competence Model which will be incorporated into a Performance Model, and the Performance Model will be written in the form of scientific laws. The fact that there is as much chance of this actually happening as of Socialism in Eastern Europe developing into full-blown utopian Communism is a separate issue.
Dretske goes on to explain how systems of rules (parking rules, the rules of chess, etc.) can be used to explain someone's behaviour, provided that one assumes that he is aware of such rules and wishes to obey them. Such explanations are in terms of scientific laws, which may be of statistical nature. He says that, in Mentalist Linguistics, Descriptive Linguists study the rules (= Grammar) of Language. Psychologists then can use the assumption that speakers are aware of these rules and want to obey them in order to formulate laws of human Speech (Dretske makes a Language/Speech distinction which is equivalent to the C1-4/P1-4 pairs).
This proposed division of labour between Psychologists and Linguists is characteristic of Autonomist Mentalism (but not Cognitivist Mentalism), on the one hand, and also of what one might call Autonomist Dualism, on the other. The latter is the position that Dretske himself goes on to argue for.
He distinguishes between brute acts and institutional acts. The latter are distinguished by having some conventional meaning. He regards Language as being a system of institutional acts, and the role of the Descriptive Linguist as being to relate brute acts (e.g. making sounds) to institutional acts (meaning something by making sounds).
Despite Dretske's apparent confusion as to the meaning of "rules" in Generative Grammar, I think there is a lot to commend his view of language and linguistics. In Cognitive Linguistics, however, I do not think that such sharp boundaries need to be drawn between the activities of psychologists and those of linguists -- or between "language" and "speech". Speakers and writers are guided by prescriptive rules, and linguists/psychologists need to take account of these in their speaker models and writer models. Listener models and reader models, on the other hand, do not need to incorporate these prescriptive rules.