OF NATURAL LANGUAGE
or the ‘Algebra of Meaning’
p The task of semantic theory, from the standpoint of the transformational-generative conception of natural language, dominant in contemporary formal linguistics, is a systematic explication of the intuition of speakers of that language, which consists in their capacity to understand and produce any new sentence.
p The appearance of a new doctrine of natural language in regard to neopositivism and linguistic philosophy is linked with this theory, which is free of the shortcomings of the former schools, namely an exclusive orientation to the physical reality of the language, in the one case, and to asystematicity in the other.^^1^^ On the methodological plane the proponents of this conception link it with ’the rationalist universal grammar’ whose principles had already been formulated in the seventeenth century by the philosophers and grammarians of the Port Royal school who had suggested that the meaning of a sentence, or its ’logical form’, was not identical with a given perceptible ’surface structure’. The meaning of a sentence is directly linked with its ’deep’ structure, which has an abstract character that is explained by the immanent properties of mind. The terms ’deep structure’ and ’surface structure’ themselves express the difference, on the philosophical plane, between ’ essence’ and ’phenomenon’. In the modern interpretation that signifies the following: if an artificial language of logic is so constructed that the logical form of expressions of this language is clearly represented in the expressions themselves, the logical form of expressions of natural language will be contained in them in a concealed way, implicitly, not on the surface level. The transformational-generative theory thus differs as a philosophy of natural language from the above-mentioned philosophical doctrines primar- 35 ily in explaining the surface structure of linguistic expressions accessible to observation by deep, abstract logical structures not accessible to observation. Bearing in mind the significance attached to the concept ’logical form’ in this theory, let us examine it in more detail.
p In the transformational-generative theory of language, the logical form of an expression is understood as that which is necessary for drawing a correct conclusion, as that on which the deductive relations of linguistic expressions depend. The classical understanding of logical form, or structure (accepted by Russell, Wittgenstein, and Quine). is regarded, moreover, as that which is constituted exclusively by ’logical words’, or ’logical constants’ (’not’, ’and’, ’or’, ’if, ’then’, ’all’, ’some’), and presupposes division of the vocabulary of natural language into logical and extralogical, or descriptive parts. Doubts about the validity of this delimitation were expressed by Tarski, who noted that
no objective grounds are known to me which permit us to draw a sharp boundary between the two groups of terms. It seems to be possible to include among logical terms some which arc usually regarded by logicians as extralogical without running into consequences which stand in sharp contrast to ordinary usage. In the extreme case we could regard all terms of the language as logical. The concept of formal consequence would then coincide with that of material = consequence.^^2^^
p It is thus proposed to regard as relations of consequence not only those which depend on semantics as relations of consequence (to the extent that semantics is disclosed in the classical logical calculi which were then considered as theories of the logical truths of natural language) of the expressions ’no’, ’and’, ’or’, ’if, ’then’, ’all’, ’some’, and other expressions of natural language similar to them in the aspect being considered.^^3^^ The point also concerns the relations constituted by the meaning of descriptive expressions belonging to grammatical categories of noun, verb, adjective, adverb (as, for example, in the argument ’There are bachelors, consequently there are men’. ’I had a nightmare, so I had a dream’). Just as it has been necessary, in modern logic, to introduce new rules of inference adequate 36 for speakers’ intuition to formalise the transition from, for example, sentences like ’Every horse is an animal’ to the sentence ’Every horse’s head is the head of an animal’ it is natural to expect the introduction of rules ’embracing’ a corresponding link in the transition, for example, from ’All nightmares have a Freudian significance’ to ’All dreams have a Freudian significance’, and so on.
p The complexity of the problem of logical form in natural language is, however, that the logical form of expressions of natural language, in contrast to the language of logic, is not given on the surface, and that behind identification of the surface grammatical form of expressions there are concealed fundamental logical differences. That, for example, can explain why the argument ’Today I ate what I bought last week. Last week I bought expensive fish. Consequently, I ate expensive fish today’ is correct, while the argument ’Today I ate what I bought last week. Last week I bought fresh fish. Consequently I ate fresh fish today’ is incorrect.
p Deep structure, according to the interpretative conception of transformational-generative grammar, is a formal syntactic object, recursively generated by certain rules of grammar (the ’direct formation rules’, or ’phrase structure rules’), and interpreted by ’projection rules’ (the semantic rules of the model) in a set of universals, i.e. of ’ semantic markers’, not dependent on the specific nature of the natural language (see below). The concept of the recursive generation, clearly, has been borrowed from mathematical logic in which it signifies the producibility of an infinite set of expressions from a finite set of initial elements by application of a finite set of rules to them.
p All that is treated as agreeing with intuition that the speaker of a natural language understands a sentence by understanding the components of it, and the way these elements are employed in it. It is thus presupposed to theoretically reconstruct a native’s capacity to make and understand an infinite set of sentences of natural language through a finite vocabulary and a finite set of rules.
p In the theory being considered semantic markers are theoretical constructs, ’atoms of meaning’ independent of the 37 concrete natural language, i.e. the minimal semantic units from which the meaning of the dictionary (lexical) units, finite components of the deep structure, is constructed. Syntactical markers (for example, ’noun’, ’verb’, ’adjective’) express the syntactical relations of lexical units, whereas semantic markers (for example, ’human’, ’live’, ’physical’) express their semantic relations. The meaning of linguistic expressions also contains certain ’selection restrictions’ in addition to semantic markers, as formal restrictions on the constructing of possible semantic objects, necessarily postulated in view of the stated aim of formalising the concept of the meaningfulness of linguistic expressions (for example, in order to exclude such structures as ’My pain weighs three pounds’).
p The semantic component of grammar thus includes vocabulary, which in turn contains semantic representations for each lexical meaning of a definite lexeme and ’ projection rules’ linking these representations. Together with the vocabulary (dictionary) and ’projection rules’ semantic representations are ascribed to sentences which corresponds to understanding the meaning of a linguistic expressions as a function of its component meanings (’Frege’s principle’) and their deep syntactic organisation. The deep level is regarded as that at which, logically, or semantically significant characteristics of expressions of natural language are given: knowledge of it presupposes knowledge of the deep structure of its expressions.
p The physical reality of natural language is a sequence of signals, whereas the set of semantic, syntactic, and even phonetic characteristics ascribed by speakers to the expressions of natural language do not have a physical correlate, and so are not functions of the physical signal. The transformational-generative theory is therefore opposed to those conceptions of natural language (in particular taxinomic and behaviourist) according to which all the characteristics needed to explain the phenomena of this language are contained in its surface structure. It is a fact that speakers of a natural language, on perceiving an acoustic signal, a priori meet it with a much more complicated model than that which could be inferred from the signal itself.38
p Just as the difference (identity) of the meaning of some lexical units is determined by the difference (identity) of the semantic markers that constitute these meanings, so the difference (identity) of sentence meanings is determined by the difference (identity) of the possible semantic structure formed by means of the projection rules from the different meanings of the dictionary units that form the sentence. In any case the semantic properties (for example, meaningfulness) and relations (for example, consequence, synonymy) of linguistic expressions are established formally according to this conception—through the presence in them of some or other semantic structure, and according to the definitions of these properties and relations contained in the theory and set out in terms of formal characteristics that the semantic structures of the linguistic expressions being examined should correspond to. An expression is considered meaningless, for instance, if no semantic structure can be ascribed to it (’My pain weighs three pounds’), i.e. if it does not get any semantic representation. An expression is said to be ambiguous if two or more semantic representations are assigned to it. Two expressions are considered synonymous if the same semantic representation is related to them.
p The question of the choice of markers included in the meaning of a given dictionary unit W is decided as follows: if there are two markers RI arid R.,, then the choice of RI as entry into the meaning of the unit W is determined by the fact that RI presents a possibility of forecasting semantic properties and relations of the appropriate linguistic expressions that include W, which is impossible with choice of R2. If, moreover, the information represented in RI and R.,, and which constitutes the sole formal difference between RI and R2. plays a role in forecasting the semantic properties and relations of the corresponding linguistic expressions, then this information relates to the vocabulary of the natural language. It is its semantic information one is only concerned with in the interpretative conception of semantics. If one can simplify the dictionary unit W without including symbols in its meaning that differentiate RI from R2 and, moreover, without damage to the predictive 39 potential of the grammar, then ’this information is not dictionary information but encyclopedia information, factual information about the referent of W’.^^4^^
p Semantic markers, being theoretical constructs, are regarded as concepts in an,informal interpretation. By the latter are understood not some individual representations or ideas, not ’elements in the subjective process of thinking’,^^5^^ (called ’cognitions’ in the theory, which constitute a concrete individual’s conscious experience, similar to individual feelings and emotions, sensations, experiences, recollections, and hallucinations), but some abstract entity in the spirit of Frege’s ’thoughts’ (Gedanken) as the objective content of the thought process, which can be passed from one individual to another as something common to all or most speakers of natural language. The entities that the linguistic expressions are related by are those that the following forms, for example, express on the interlanguage level: Lithuanian •m.’.n salta’, English ’/ am cold’, French ’j’ai jroid’, Spanish ’tengo frio’, etc.,
p Frege’s idea of ’thought’ is very close in general to the methodology of deep structures.
Nowadays [he wrote] people seem inclined to exaggerate the scope of the statement that different linguistic expressions are never completely equivalent, that a word can never be exactly translated into another language. One might perhaps go even further, and say that the same word is never taken in quite the same way even by men who share a language. I will not inquire as to the measure of truth in these statements; I would only emphasize that nevertheless different expressions quite often have something in common which I call the sense, or, in the special case of sentences, the thought. In other words, we must not fail to recognize that the same sense, the same thought may be variously expressed; thus the difference does not here concern the sense, but only the apprehension, shading, or colouring of the thought, and is irrelevant for logic. It is possible for one sentence to give no more and no less information than another and, for all the multiplicity of languages, mankind has a common stock of thoughts. If all transformations of the expression were forbidden on the plea that this would alter the content as well, logic would simply be crippled, for the task of logic can hardly be performed without 40 trying to recognize the thought in its manifold guises. Moreover, all definitions would then have to be rejected as false.
[My italics— R.P.]^^6^^
p As to the ontological status of concepts the theory does not provide a final answer so as not to be linked with such a strong assumption as the existence of universals. It is supposed that first approximations to an answer to questions of that kind can only be expressed when researchers have more knowledge at their disposal about the semantic structure of expressions. Therefore, the required explanations of an ontological order about the concept ’concept’, although possibly extremely desirable, cannot be treated as preliminary conditions of the successful construction of a theory and a theoretical explication of the semantic structure of linguistic expressions. The reliability of the definition of a finite set of elementary semantic markers, and equally substantiation of its delimitation from the infinite set of complex semantic markers determined by it, are dependent on the success of this explication.
p Semantic universals, like the other fundamental characteristics of transformational-generative grammar, are presumed to be components of the universal basis of any natural language. This grammar is treated as a theory explaining what is common for natural languages in terms of the concept ’naural language’; it is a theory of natural language in general, and one speaks of universals as having the status of innate principles.
p The rationalistic theory of these principles constitutes the philosophical stock of transformational-generative grammar; in that sense the theory can be regarded as a formal reconstruction of the hypothesis of ’innate ideas’. According to this theory a person can very early on, and independent of how intelligent he is, assimilate the concrete grammar and most intricate constructions of a natural language on the basis of these principles, while dealing mainly in his linguistic practice with fragmentary and defective material.
p Affirmation of the universal character of the semantic aspect of the correlation ’linguistic expression—meaning’ 41 also gives the theory philosophical significance and, as its supporters claim, is the basis for interlinguistic postulates of meaning as an invariant of intralanguage and interlanguage transformations that make it possible to construct a hypothesis of the relation of natural language and logic more definitely than before, i.e. of the relation between the logical form of an expression and its linguistic formulation.
p In this interpretation acceptance of the methodology of deep structures means that all natural languages have, at a sufficiently abstract level, one and the same grammatical structure and semantics.^^7^^ The existence of meanings specific for separate natural languages is explained, in turn, by the universal semantic components’ being able to form semantic combinations specific for the given language, fixed in lexical units in its vocabulary. The ’possible’ concepts are thus defined in terms of universal semantic components that also constitute the initial semantic basis of any natural language. From that it follows that all languages have identical expressive possibilities and are all intertranslatable in the sense that for every sentence in a language there is at least one sentence in any other natural language that expresses the same meaning.
p Even allowance for the hypothesis of linguistic relativity cannot bring proponents of the thesis of natural language’s universal expressive possibilities to reject it, but only to modify it: namely, as follows. If one admits, in accordance with the hypothesis, that natural language predetermines the categories and forms of the thoughts of people who acquire it, then the fact that mastery of a natural language limits its speakers’ expressive possibilities will equally be evidence that each thought of the speakers of a natural language is expressed in that language.
p The possibility of understanding any new expression of a natural language is ensured (as follows from what has been said above) by this expression’s being contained, in view of the recursive character of the basic rules of grammar, in an infinite set of expressions generated by these rules. Finally, the transformational-generative theory is not a theory of the use (’performance’) of a natural language by its real speakers, but a theory of linguistic competence, i.e. 42 a theory that models the linguistic intuition of an ideal speaker of a natural language, whose ’performance’ of it is not limited by any physical and psychological factors like volume of memory, attention, and motivation, which are assumed to be ‘grammatically irrelevant’.^^8^^
p It is consequently a matter of a theory that models what the ideal speaker of a natural language knows about the grammatical structure of his language (which consists of phonological, syntactical, and semantic components). This constitutes the implicit knowledge of a real speaker of the language, through which he communicates with other speakers of the language. Linguistic ’performance’ is how linguistic competence is employed in actual speech situations by speakers of a natural language. Thus, in the theory of transformational-generative grammar as a theory of linguistic competence, not only are the functions of language in the business of cognising the world abstracted from the irregularities, distortions, and mistakes peculiar to the use of natural language, but are also abstracted—because of exclusion of the factor of the use of linguistic expressions from the field of review—from examination of the fundamental link of language and the world. It pursues the aim of systematising only those aspects of language that reflect its knowledge as a system, i.e. the ideal speaker’s knowledge of all the semantic relations available in the system of the language provided by the postulated absolute set of meanings. This knowledge is determined irrespective of the speaker’s knowledge of the world. The concept of linguistic competence, on the contrary, is defined through the concept of an ideal speaker of the language as the basis of the linguistic ’performance’ of real speakers of the language, the basis of its use. In other words, it is a theory of the competence of a speaker of the language whose linguistic knowledge is wholly determined by the possibilities of explication contained in the theory itself.
p The attempts of proponents of this conception to justify treatment of knowledge of natural language unrelated to its speakers’ knowledge of the world (in essence, counterposed to this knowledge) is backed up with references to the idea that the task of the linguist engaged in constructing 43 a semantic theory of natural language does not include discover)’ of which of its expressions denote one and the same object of the world (as, for example, in the case of the ’morning star’ and the ’evening star’, which designate the planet Venus) or which of its sentences express truth and which falsehood (for example, that the sentences ’The morning star is the evening star’, or ’All cats have tails’ express something true or false is considered the business of the astronomer or zoologist, and not of the linguist). Ignoring of the extralinguistic circumstances of the use of natural language determines in advance the closed character of the formal model concerned. Meaning is then determined exclusively through combinatorial analysis of initial, universal, ’semantic atoms’.
p According to this conception everything that can be said about meaning can be said in terms of finite sets of elements, having in mind the generation of an infinite set of possible finite objects (meanings and the expressions of the language related with them) from a finite set of ’ semantic atoms’ through a finite number of applications of certain rules. As David Lewis has remarked, in such a case,there is no risk of alarming the ontologically parsimonious. But it is just this pleasing finitude that prevents Markerese semantics from dealing with the relations between symbols and the world of non-symbols—that is. with genuinely semantic relations.^^9^^One can therefore only say—that the interpretative conception, in proposing such a universal algebra of meaning, ’models’ the meaning of linguistic expressions in that understanding, abstracted from analysis of the relations of natural language and the world. Only in that sense, divorced from examination of the relation of natural language and the world, can one speak as well of the degree of.the theory’s adequacy to the object investigated by it.
p The aims of interpretative theory, the claims about its empirical verifiability, and at the same time about its necessary intralinguistically closed character, inevitably come into conflict. That indicates the unsoundness of the opposing of linguistic knowledge (knowledge about language) to 44 extralinguistic knowledge (speakers’ knowledge of the world). Such an opposing of these varieties of knowledge in the theory under consideration is not simply the usual theoretical ploy employed for a more adequate description of the object of investigation, in this case language, but a direct consequence of the methodologically mistaken precept initially adopted in treating language divorced from its links with reality, and from the information about the world that its speakers have at their disposal.
p The concentration of semantic theory exclusively on description of ’linguistic knowledge’, which is itself a theoretical construct, and an abstraction from the real knowledge of real speakers of the language, leaves open not only how the language is related to reality and how it is possible to employ it to describe and understand reality but also how it is possible in general to understand this reality-describing language.
p The claims of this theory to explain the intuition of understanding language in such a situation simply cannot be realised. It cannot then also be a matter, of course, of explaining the functions of language in cognition. The theory’s incapacity to embrace what is called speakers’ ’ extralinguistic knowledge’, and in spite of that, its claim to be a theory of understanding of language, condition acceptance of its abstract ’semantics of language’ devoid of links with reality, and of the hypothesis of ’innate knowledge’ in order to justify its universal character, which even further deepens doubts in its methodological foundations.
At the same time the importance of a number of the theoretical ideas and constructs put forward by this theory cannot be denied. The determination of two levels of analysis of linguistic expressions—deep and surface—must be included in these constructive principles, which provides the basis for a broad comparison of logical and linguistic structures and for bringing out the relation between them. The theory’s proposition to broaden the very concept of the logical form of linguistic expressions is also positive; consequently, the possibilities of determining the logical correctness of arguments carried on through natural language are substantially enhanced.* * *