Consistency and Coherence
In the last chapter, I started my critical examination o f P-BSA theory with the assumption that it was the way to proceed with a design model derived from an analytic linguistic base but ‘stiffened’ by the introduction of planning for the practical purposes o f design. I attempted to show that in spite o f what I claimed were its promising features it fell short o f fiilfilling its promise. I want now to step back and consider how such a theory as Cohen & Perrault’s should be viewed and if it is not adequate, what kind o f theory or model would be; or in what way it could be
supplemented. I take account o f what kind o f model Cohen & Perrault think that it is, as well as arguing that it can be classified with respect to some general and well used categories. Further, that if it is inadequate, is there a theory or model otherwise available which could compensate for this inadequacy? And if there is a theory or model which falls into a complementary category, can some more general conclusions be drawn about knowledge which will allow us to assess NLD’s conformity with an engineering view of cognitive matters?
A further less specific reason for this need to gain some perspective is that much of the argument deployed so far has a bearing on linguistic theory but I have not addressed, except obliquely, some o f these points o f difference with the linguistic
theories which abound. Different linguistic theories may well bear on the issues I raise at many points but I cannot deal with them individually. I propose, therefore, by concentrating on these more general properties and categories o f linguistic theories, referred to above, firstly, to individuate my viewpoint with respect to others; and, secondly, at the same time, to focus on just those features which I need to exploit for the more general NLD fi-amework and its relationship with Human-Computer
Interaction as an engineering discipline (HCIe (Dowell & Long, 1989)), as well as for the foundational fi*amework which I intend to develop, on the basis o f the terms of the NLD design firamework, in the second half o f the thesis.
It is clear, then, that the presuppositions and assumptions o f P-BSA theory, after the analysis o f the last two chapters, are not going to instantiate the framework, but only a stage on the way to that framework. Cohen & Perrault themselves acknowledge that not only is their work incomplete but that it is by its very nature partial with respect to dialogue: that their theory is only a ‘competence’ theory (that they cannot account for ‘performance’ or ‘process’), and, in addition to this partiality, they concede half the dialogue ground to the work o f Allen and Perrault (see Allen (1979), and AUen & Perrault (1980)), whose work concentrates on the other half o f the dialogue - plan- based recognition o f the speech act. Even taken together, their work will not account for ‘coherence’. One way of describing this partial treatment by Cohen & Perrault and Alien & Perrault might be to say that they deal only with that component o f the dialogue model which allows us to judge for consistency - the necessary conditions of the dialogue. The missing component is the one which would allow us to assess the sufficiency conditions o f the dialogue: a step closer to an explanation o f what
‘coherence’ might be. I want first to examine what sort of theory the P-BSA account is claimed to be by positioning it with respect to some others, and ask how its status or type might reflect on what is needed to complement it, for my purposes. I would like, then, to see how it might fit other more general categories to suit my more general aims.
Competence and Performance
What do Cohen & Perrault mean by saying that they are setting out to construct a P- BSA theory which fulfils the requirements o f a ‘competence’ theory? They write that
it is “analogous to competence theories o f grammar” and refer to Chomsky (1965) as their source. It appears to be analogous in the sense that their theory “describes the set o f possible plans underlying the use o f particular kinds o f speech act, and thus
states the conditions under which speech acts o f those types are appropriate”. At the end o f the paper they describe it as follows:
“a plan-based competence theory, given configurations o f beliefs and goals, speech act operators, and plan construction inferences should generate plans for all and only those speech acts that are appropriate in those configurations” (my italics). Chomsky (1965) writes that he is concerned “with the syntactic component o f a generative grammar, that is, with the rules that specify the well-formed strings of minimally syntactically functioning units”. In the context o f discussing the changes Chomsky made to refine the relationship o f syntax with the sounds and meanings o f language, Gamham (1985) writes, “Chomsky intends to relate sounds and meanings by generating in the syntax all and only the sentences o f the language, and then associating with each sentence a sound and a meaning” (my italics). One difference between Cohen & Perrault and Chomsky is the relative completeness o f the scope o f application o f their respective theories. Cohen & Perrault are only concerned with dialogue; Chomsky addresses competence for language as a whole. A second important difference is what kind o f status is being attributed to this ‘competence’, since Chomsky believes that his grammar springs fi*om an innate and mentalistic source. This difference has implications for what goes beyond the minimal claims of such as the P-BSA theory, but the common ground between Cohen & Perrault and Chomsky is that, with respect to the realm o f the application o f the theories or models in question, the two theories aim to account for all possible expressions without paying attention to the causes or rationalisations for the actual (or performed) dialogue or general linguistic expression. Both Cohen & Perrault and Chomsky contrast their approach with a fuller empirical one: in Chomsky’s case, that wider scope is covered by ‘performance’ theory; in Cohen & Perrault’s, the reference is to ‘process’ theory. In both cases, the narrowing o f the scope o f their attention is determined by the purposes o f their projects.
Cohen & Perrault’s contrast with ‘competence’ is ‘process’, which presumably lies somewhere between ‘competence’ and ‘performance’, since it is defined as follows,
“aprocess theory...concerns how an ideal speaker/hearer chooses one (or perhaps
more than one) plan out o f the set o f possible plans. Such a theory would
characterize how a speaker decides what speech act to perform and how a hearer identifies what speech act was performed by recognising the plan(s) in which that utterance was to play a part.” (my bold)
The mention o f the ‘ideal speaker/hearer’ appears to preclude it fi*om being a true ‘performance’ theory^^, whüe a competence theory goes no further than presenting (but not selecting fi*om) a set o f possible plans. At any rate, they seem to see this ‘process’ type o f theory as going some substantial way towards a ‘performance’ theory. Further, they identify it with the work on plan-based speech act recognition by Allen & Perrault, who deal in addition with the phenomenon o f the indirect speech act (ISA). A distinction which Cohen & Perrault do make clear is that whatever they, or Allen & Perrault, are doing it is not contributing to ‘cognitive process’ theory, which “claims would require empirical validation”, and Cohen & Perrault comment, “it is unclear whether there could be just one process theory o f intentional behaviour since each individual might use a different method” (my italics). In sum, this view is close to what Gamham (1985) calls
“an alternative formulation o f the competence/performance distinction, originally proposed by Stanley Peters (see Johnson-Laird, 1983, p i 67).... In technical terms, the linguistic theory specifies the function, in the mathematical sense, to be
computed. Psycholinguistics has to determine which o f the indefinitely many possible procedures the human language understanding system uses to compute the function”.
It is difficult to square this formulation exactly with what Chomsky characterises as performance theory - a linguistic theory which deals with the “false starts, deviations from rules, changes o f plan in mid-course, and so on” (Chomsky, 1965). He says that performance would be “a direct reflection o f competence” if it were
^'^"Psycholinguistics can thus explain deviations from the ideal in what Chomsky calls linguistic performance (‘the actual use of language in concrete situations’ (Chomsky, ‘65))" (my italics) (Gamham, 85).
“concerned primarily with an ideal speaker-listener, in a completely homogeneous speech-community, who knows its language perfectly and is unaffected by such grammatically irrelevant conditions as memory limitations, distractions, shifts o f attention and interest, and errors (random or characteristic) in applying his knowledge o f the language in actual performance”,
and then he adds that this is, o f course, impossible. The implication is that the serious study o f linguistics must concern itself with linguistic competence, which, for
Chomsky, is absolute and knowable; that any other knowledge o f language is relatively ad hoc, and deals with the chaotic and unpredictable results o f error and imperfect capacity. There is no indication that Cohen & Perrault, by contrast, are holding to the view that their model is a representation o f an innate mentalistic ability, nor that nothing serious or rigorous can be done in the wider field of, what they call, process or cognitive process theory. What, for them, makes the study rigorous is the assumption o f ideality, and one can distinguish between the ideality o f a conpetence theory and that o f a performance theory (which they carefully distinguish as a ‘process theory’ insofar as its application is not studied empirically)'^^. (It is the more normally acceptable view that all theory is essentially ideal: it postulates components and regularities, tentatively, and in its applications to real-world situations operates on the qualifying principle o f ‘ceteris paribus’.) Perhaps, however, the categories
‘competence’, ‘process’ and ‘performance’ are employed too differently in different contexts to have much general import. Can one, therefore, employ more generally used categories to bring these theories, or kinds o f knowledge, into line with the wider world of linguistics and design?
Semantic/Pragmatic
Origins and Rationale
In 1938, Charles Morris introduced the trichotomy - semantics, syntactics and pragmatics - in order to refer exhaustively to the three parts o f the theory o f signs, semiotics (Gamham, 1985). Semantics was concerned with the relations o f signs to what they signified; syntactics, with their relations with each other; and pragmatics
'^^However, in "Rules and Representations" (BBS ‘target article’, v3, 1980), Chomsky writes "that it is possible in principle for a person to have full grammatical competence and no pragmatic competence"(my italics). Given his technical definition o f ‘competence’ this quote implies that a performance theory could be idealised away fi*om actual performance, and looks closer than his original notion (see Chomsky (1965) quote in the text above) to Cohen & Perrault’s ‘process’ theory.
with the relationship o f signs to the user (deictic terms such as ‘I’, ‘here’ and ‘now’). The salient contrasts which arose were semantic/syntactic and semantic/pragmatic. ‘Semantic’, therefore, developed a dual aspect. Unfortunately, the two dichotomies were both used in the field o f linguistics, though the latter - semantic/pragmatic - is applied more widely (see below). As a consequence, some confusion can arise. A theory, for example, which, unlike Chomsky’s might not emphasise the primacy o f syntactic rules, could be assumed to be a semantic theory, i.e., one which emphasises the primacy o f meaning. However, though concerned with meaning, it might be so but in a pragmatic manner. Further, how this semantic/pragmatic distinction is made varies fi"om writer to writer. For example, Gamham (1985) points out that, according to Montague Grammar, both formal semantics and formal pragmatics deals with truth conditions. But if Gazdar’s suggestion that any theory of language which deals in truth conditions is a semantic theory (Gazdar, 1979) then Montague’s pragmatics would, contrary to his own convictions, fall within the purview o f semantics.
At the same time, syntax (the form of language) and semantics (the content of language) are inextricably linked; and the particular position which Chomsky adopts rests, I think, on a particular assumption: that to understand language in a scientific way one must restrict the scope o f the theory to what ‘competence’ covers, and only the form or syntax can be expressed explicitly. However, since what one intuitively accepts as a sentence (with a given syntax) must be an expression which can be applied meaningfully in some circumstances or other, then, in those circumstances it could be meaningfully true or false, and, therefore, according to Gazdar - a semantic theory. Chomsky’s concentration on syntax is partly methodological and
epistemological in motive, but the overall goal is the understanding o f the meaningful and demonstrably valid use o f language. Perhaps we should accept, as Ramsay (1990) does, that syntax and semantics are inseparable.
In the beginning, then, Morris’ categories drew attention to aspects o f linguistic study, on more or less equal terms with each other. As the discipline developed, the context or world played a critical role in the validation o f linguistic knowledge. The boundary between the semantic and the pragmatic became epistemologically problematic. This dichotomy took on a more than merely categorial significance, and this more general
distinction, I believe, is that which exists between theories which are, in some sense, circumscribed and complete and those which are essentially open-ended. Although the dividing line between them is disputed and/or fuzzy, most people agree that these categories are nevertheless distinguishable; and have a crucial relevance to the holder’s world-view.
Implications fo r Competence/Performance Dichotomy
Therefore, if performance is understood as adequately empirical, the distinction between competence and performance should perhaps best be seen as defining the boundary between the autonomous core of linguistic knowledge (or dispositions), on the one hand, and the relatively ad hoc understanding o f the manifold diversity of actual linguistic behaviour, on the other; leaving aside the validation o f any ontology. The general aim o f the P-BSA theory is to concentrate on what can be defined
rigorously, accepting that there are no direct empirical cognitive implications. And although Cohen & Perrault’s undertaking does not appear to carry with it the kind of ontological presuppositions which Chomsky’s does, they seem to agree that they are dealing with linguistic/conceptual issues unsullied by foreign factors such as memory limitation, or other more general cognitive influences.
Putting the details o f the process/performance distinction on one side (and also sidelining ontological issues), I want to suggest that the semantic/pragmatic dichotomy used with respect to the division within linguistic theory is a better
categorisation of what are referred to, respectively, as ‘competence’ theory on the one hand, and ‘performance’ theory on the other; but it comes without the implication that one part is superior to the other. Rather, as I have said, it indicates different realms of application or function. What is characteristic o f Cohen & Perrault’s or Chomsky’s theories is that they tend to the more or less formal. Chomsky has separated his theorising fi-om any empirical consequences. For Chomsky, as Gamham (1985) puts it.
“if psycholinguistic data did not confirm the predictions o f a performance theory, it was not the competence theory that was wrong, but the account o f how linguistic knowledge is used”.
This is so because Chomsky has arrived at his theory as a rationalist. Although Cohen & Perrault have adopted an analytical approach they have not argued that it is
apodictic in the sense which is claimed for Chomsky’s. Their theory, therefore, arguably falls within the category o f semantic theories in that it attempts to impose rigour, and adopt a version o f the principle o f compositionality, as I have already noted: features which, if achievable, should guarantee consistency. Cohen & Perrault do not, as I said above, and unlike Chomsky, make any cognitive reality the object o f their theory.
It is characteristic o f pragmatics, however, in contrast with semantics, that the rules for those linguistic phenomena with which it is concerned are defeasible, i.e., they may always be overridden by higher principles (or the application o f the principle at one remove). Consider the maxims o f Grice, a writer thought o f as dealing with
pragmatics: they are guides which are supposed to indicate the conditions for certain utterances, but the conditions under which they themselves operate cannot be
specified unless by reference to the principle o f cooperation, and then only given the circumstances. But when Cohen & Perrault specify the conditions, they do so with the intention of making them precisely implementable.
They are, indeed, embodying the conditions in a formal setting, using the possible world’s model, and are attempting to assess how to make the above specifications in a consistent fashion; they are ignoring those factors which determine how a topic might arise in a dialogue - either in terms o f its generation or its comprehension. When they allude to the work (by Perrault & Allen, 1980) which might be done on indirect speech acts by employing similar techniques to the ones they have developed, they do not claim that it is anything other than a competence theory like theirs (though Perrault & Allen do factor in heuristics to support the search through the problem space - a pragmatic device). In the terms in which I am operating, I would suggest that theirs too is an endeavour in semantics, since, as Cohen & Perrault’s also does, it adheres largely to the standard principle o f compositionality. Pragmatics is, therefore, what is left over in accordance with the Levinson’s (1983) catchphrase ‘pragmatics is meaning minus semantics’ when semantics is concentrated on truth-conditionally determined meaning, conforming fairly closely to Gazdar’s view (Gazdar, 1979) - at
least with respect to a model. The remaining ambivalence rests on that to which truth