RUSSELL, CHOMSKY and NOMINALIZATION



Steven R. Bayne



Bertrand Russell held, at one time, that all propositions of pure mathematics are of the form "p implies q" (1903:3). But, when he set out to state the meaning of this, he encountered a problem he would describe as "inevitable," a problem he professed to be unable to solve. What I propose to do is show how, working within the framework of Russell's early platonism, a solution to this problem based on Chomsky's (1970) views on nominalization can be made available. Others have commented on the significance of Russell's use of nominalization (Cocchiarella 1980; Landini 1998), but here the larger objective is to demonstrate how tools supplied by theoretical linguistics may be applied in resolving subtle philosophical problems of consequence.

Russell's Problem

What bothered Russell was a semantic contrast between (1) and (2).

1. Socrates is a man therefore Socrates is mortal
2. Socrates is a man implies 2+2=4

(1) states a relation which obtains only between asserted propositions, whereas (2) asserts a relation that may hold between any two propositions such that the first is false or the second is true; although neither proposition is itself asserted by (2). Problematic is the fact that if assertion modifies a proposition then "no proposition which can possibly in any context be unasserted could be true, since when asserted it would become a different proposition" (Russell 1903:35). Russell observes that the distinction between asserted and unasserted propositions is expressed grammatically in terms of sentential and nominal constructions: the difference, he says, between (3) and (4) is that (3) is asserted, whereas the proposition represented by (4) is merely considered, insofar as it is a "complex concept."

3. A is greater than B
4. A's being greater than B

Russell argued that we must be able to make propositions logical subjects. Although he formulated his argument in terms of adjectives (1903:46), his reasoning can be clearly stated in the case of sentences: suppose we were unable to nominalize some proposition, p, then we could not say "P is F," since we must make a subject (argument) of the proposition; but neither could we say "P is not F." Conjoining these two facts, double negation, applying in the second instance, leads to a violation of the principle of non-contradiction. Russell was acutely aware, however, that nominalization was a grammatical move that might have logical consequences. In one important passage he remarks:

The question is: What logical difference is expressed by the difference of grammatical form?... By transforming the verb, as it occurs in a proposition, into a verbal noun, the whole proposition can be turned into a single logical subject, no longer asserted, and no longer containing in itself truth or falsehood. But here too, there seems to be no possibility of maintaining that the logical subject which results is a different entity from the proposition... Yet it is quite plain that the latter ["the truth of Caesar's death" - SRB] at any rate is never equivalent to "Caesar died"...thus the contradiction of an entity which cannot be made a logical subject appears to here become inevitable. (1903: 48)


Central to Russell's problem is that nominalization may result in an expression indicating some entity other than the entity expressed by the associated sentence. If assertion is an essential feature of sentences, then since nominalization dispenses with the feature of assertion, there arises a reasonable doubt whether the nominalization can be used to indicate the same entity expressed by the sentence. If it cannot, then not only is it doubtful that propositions can be represented as logical subjects; it also becomes a matter of controversy whether the truth conditions for material implication can even be stated. Material implication could become as ineffable as logical form in Wittgenstein's Tractatus. In his remarks on Meinong (1904:339), Russell sides with Meinong, and against Frege, in maintaining that sentences and associated derived nominals refer to the same things.

Russell's Problem and Chomsky's (1970) Views on Nominalization

It will prove significant that the nominals Russell employs for the purpose of indicating "complex concepts" are frequently gerundive nominals and not, following Chomsky's typology, "derived nominals" (Chomsky 1970). Aside from relations of material implication (and recall that for Russell material implication is fundamental to understanding formal implication) , there are other constructions besides material implication where the problem posed by Russell arises. Here I have in mind sentences involving propositional attitudes such as (5) - (6).

5. John believes that Ortcutt is a spy
6. Mary fears that Olatunji may play Howard's drums

The attitude verbs, "believes" and "fears," appear to take a proposition indicated by the expression following the complementizer "that" as relatum. Such examples have been pursued with considerable persistence by philosophers. What I will attempt here is to examine sentences, such as (6) and (7), incorporating verbs of propositional attitude that take nominal objects which themselves are related to verbs of propositional attitude.

6. John believes Bill's assertion that Mary left
7. Mary doubts Bill's denial that John left

We will discover that Russell thought that he could speak consistently of complex propositional concepts instead of asserted propositions possessing a truth value by nominalizing the verb and thereby cancelling the assertion. Nevertheless, he oversimplified, treating all nominalizations as essentially the same.

Russell typically located the assertion element of a proposition in the v erb (1903:44); more specifically, he relied on the notion that the assertion element of the verb is always tied to a predicate, so that in the proposition "John left home" the assertion "left home" would combine with the subject in forming that proposition. What Russell paid little attention to were differences among the various sorts of nominals. He would move freely from gerundive nominals such as "Socrates' being dead" to derived nominals such as "the death of Socrates" (1903:43). No doubt his only concern was to detach the assertion element using any stylistically acceptable nominal that suited him.

Russell's effort to clarify material implication by using nominals to refer to propositions minus the assertion feature opens the door to consideration of other forms of entailment and of the role of assertion. Hopefully, this will be made evident once we have laid out some linguistic data, but for now a general point needs to be made, and that is that Russell in moving back and forth between gerundive and derived nominals failed to appreciate the fact that what sentences are asserted by an utterance in at least one sense of "assert" depends on the syntax of the particular nominal being used. Perhaps Russell's neglect of the details of nominalization in his treatment of propositional attitudes can be attributed to confining his concept of implication to relations of the propositional calculus.

Chomsky and the Variety of Nominals

Chomsky (1970) identifies three sorts of nominals: gerundive (8), derived (9), and "mixed" (10).

8. John's refusing the offer
9. John's refusal of the offer
10. John's refusing of the offer

His primary focus is on gerundive and derived nominals. For our purposes what distinguishes them are two facts: first, in the case of gerundives there is a uniformity of meaning between them and corresponding subject-predicate sentences; and second, our inability to substitute determiners for the "subject" of the nominal suggests that it lacks the internal structure of an NP and is in fact arrived at by transformation from a sentence. There is a third difference between gerundive and derived nominals (Chomsky 1970: 264-265), productivity, but for the time being I will concentrate on these two features alone.

Chomsky (1970): The Transformationalist and Lexicalist Positions

With respect to gerundive and derived nominals there are at least two positions to take: the lexicalist position and the transformationalist position. The lexicalist position (Chomsky 1970: 265) is that in order to accommodate such nominals we must extend the base, consisting now of both lexical and phrase structure rules (Chomsky 1965: 84). It's competitor, the transformationalist position, would require additions to the transformational component of the grammar. In the case of derived nominals, for example, where internal structure "mirrors" (Chomksy 1970:282) that of the corresponding sentence, we may take advantage of the fact - on the lexicalist position - that the lexical entry for the head is free to take the categorial feature of noun or verb. No transformation is involved in accounting for the difference, just a difference in the categorial feature. Much later Chomsky is quite explicit in maintaining that this "correspondence" involves possessing "essentially the same internal properties" (Chomsky 1986:63), a position also strongly encouraged by Abney (1987:110).

Chomsky ultimately concludes that gerundive nominals are best treated by the transformationalist theory while derived nominals belong in the lexicalist camp. He is less certain about the mixed cases, which will not concern us. Now that we have some of the elements of Chomsky's theory, let us return to Russell and see what philosophical ground can be gained by Chomsky's analysis.

Russell's "indifference" Re-examined

Because Russell locates both assertion and having a truth value in the verb, he tends to treat them as one and the same. This may not always be obvious, as when he maintains that there is a sense of logical assertion tantamount to truth itself (1903:49), and which is to be distinguished from "psychological assertion." We will benefit at this point by entertaining the possibility that assertion and truth are to be distinguished. Doing so will allow us to state a proposal that avoids violating the principle of non-contradiction, which Russell felt was unavoidable. Before making explicit how this proposal is to be implemented an emendation of Chomsky's 1970 views on nominalization will provide a suitable opportunity to consider linguistic data that would have considerably benefitted Russell's position.

Selectional Properties and Chomsky's Nominalizations

Chomsky concludes that some complex nominals, gerundives, are best treated transformationally, while other nominals require a lexicalist approach. Besides this there is at least the implicit suggestion that there are two kinds of syntactic nominals: those which are transforms and those which follow from the freedom of lexical entries to select a feature, either +verb or +noun. There is another proposal which has yet to be considered: it may be the case that a derived nominal may be arrived at in either of these two ways, and that which sort a nominal must be taken to be depends on certain selectional properties of the matrix verbs that take such nominals as syntactic complements. Here I depart, somewhat, from Chomsky's 1970 analysis.

It is quite natural to suppose that if a nominal is reached by tranformation of the associated sentence that the properties of that sentence will be retained by, or at least recoverable from, the transform. In this case the prediction will be that substituting the sentence for that nominal in context will have no effect on the relevant semantic properties of the sentence in which it occurs. This prediction appears to be born out in the case where we substitute without change in meaning the associated sentence for the derived nominal in (11), yielding us (12).

11. Mary regrets Bill's belief that p
12. Mary regrets Bill believes that p

There is a further prediction: if we accept Chomsky's view that gerundives are best accounted for on the transformationalist theory, then such gerundive nominals will be substitutable for derived nominals in those cases. The prediction is confirmed by a comparison of (11) and (13), just as in (12) no change in meaning is apparent.

13. Mary regrets Bill's believing that p

Suppose, however, that the very same nominal, viz. "Bill's belief that p" in sentences like (14) can also be arrived at assuming lexicalist principles. In this case, we would have two predictions: first, that the associated sentence would not be substitutable for the nominal, and second, that the gerundive would not be substitutable for the derived nominal. (15) and (16), respectively, confirm these predictions.

14. Mary doubts Bill's belief that p =
15. Mary doubts Bill believes that p
16. *Mary doubts Bill's believing that p

The substitution resulting in (15) is unacceptable because the meaning may be that Mary doubts the belief that p; she may not doubt that Bill believes that p. There is yet another prediction based on the idea that the derived nominal in (11) is not an NP in its internal structure, while the one in (14) is: If (11) did possess this internal structure then, following Chomsky, we would expect a definite article to be capable of occupying the SPEC position of the purported NP. But this is not the case, as (17) shows.

17. *Mary regrets the belief that p

The article is, however, capable of occupying this position in (14). It is fundamental, then, that the matrix verbs have different selectional properties. The conjecture is that in constructions of the type under investigation verbs like "regret" select nominals with a lexicalist explanation, while verbs like "believe" select gerundives with transformational histories. It is not obvious that gerundives receive any other treatment than that offered by the lexicalist. With these linguistic considerations behind us we are now prepared to solve Russell's dilemma regarding assertion and nominalization.

A Solution to Russell's Problem

Russell takes two seemingly unwarranted positions. First that in some sense truth and assertion are the same, and second that there is no significant logical difference between gerundive and derived nominals. In solving Russell's problem, I will show that the difficulty arises only in relation to assertion, not truth; next, it will be shown that taking a "mixed" view with respect to Chomsky's competing theories of lexicalist and transformational approaches to nominalization overcomes the challenge presented by assertion.

Since gerundives are actually derived from their associated sentences, the suggestion that what the gerundive indicates lacks the property of having a truth value, a property possessed by its associated sentence, is without sufficient warrant. In the case of derived nominals a similar point can be made. There is simply no reason to believe that a nominal expression if it indicates any proposition will fail to represent a proposition having a truth value. This still the problem presented under the guise of assertion.

Regarding an attitude verb as taking as its semantic object the proposition expressed by its syntactic complement is simplistic. While the syntactic object of (18) is a nominal its semantic object on the relevant reading is in fact the proposition p.

18. Mary accepts Bill's belief that p

The associated sentence of the derived nominal in (18), "Bill believes that p," is not asserted, but it is asserted in (19), the verb being factive in (19) but not in (18). This difference determines what the sentences can be used to assert. Russell to this extent was quite right: What is asserted depends on the verb, and the verb can be treated logically as an operator much like material implication or entailment.

19. Mary regrets Bill's belief that p

Notice that the speaker in the case of (18) may mean either that Mary accepts the fact that Bill believes that p or that she accepts p as true. This attests the possibility that a derived nominal may be lexically or transformationally accounted for. Although it raises certain issues related to the degree, if any, to which syntax is semantically driven it contributes a syntactic solution to a semantic problem. It is to be observed that there is a certain ambiguity contra (19) in what is being asserted. If the nominal is construed transformationally, and, therefore, as capable of assimilation with gerundives, then what is asserted is the first interpretation given, otherwise what is asserted is the second interpretation.

Commonplace examples of opaque verbs, such as "believes" in (5), show that sentential objects need not be asserted. How, then, does "believes" differ from "implies"? At least part of the problem is not limited to nominalization, but by pursuing nominalization, in particular along Chomsky's lines, we may take some first steps towards a solution.

The proposal, then, is this: Whether a nominal can be used to indicate the same entity as that expressed by the associated sentence taken in isolation depends on whether the nominal is arrived at by tranformation or is base generated, that is, amenable to explanation along lexicalist lines. Nominals which are gerundives, following Chomsky (1970), or derived - albeit substitutable salva veritate for gerundives - retain the feature of assertion in Russell's sense of assertion. When the nominal is base generated, the feature is not retained, and the entity, if any, indicated by the nominal differs from the proposition expressed by the associated sentence. Some matrix verbs, as in (19), allow only a reading in which the complex nominal object is either gerundive or derived but substitutable for a gerundive; other matrix verbs, as in (18), allow for the possibility of at least one other reading: the object is base generated and what is indicated does not retain the feature of assertion. In the former instance, "Bill's belief that p" indicates the proposition Bill believes that p and the assertion element is retained. In the latter case, and on the relevant reading, Mary may have never heard of Bill. Here what is indicated is the proposition expressed by "p." The solution to Russell's problem entails saying that sometimes nominalization preserves the assertional feature and sometimes not.

It is possible that following Russell's latter approach (Russell 1905) that a nominalization may not be a referring term at all. If so it may under no circumstance be used to refer to a proposition. On the present proposal this would apply only to nominalizations like derived nominals with true NP internal structure. Nominalizations would then be definite descriptions, supportable as such by the fact that like such derived nominals a definite article would be permitted in SPEC position.

If we flank the operator "therefore" with gerundive nominals, the verb forces a transformationalist understanding of those nominals: whereas, "implies" forces a lexicalist interpretation. In the case of "regret," the properties are as they are in the case of "therefore," while in that of "accepts" there is parallelism with "implies." One might ask: Is it being suggested here that a derived nominal cannot be used to refer to what is expressed by its associated sentence in "The mortality of Socrates implies 2+2= 4"? No, only that this sentence is ill formed, stating a relation between one of Socrates' properties and a proposition. The intention is fulfilled by using instead "Socrates' being mortal implies 2+2= 4" or on a transformationalist understanding "Socrates' mortality implies 2+2= 4."

Russell uses nominalizations in making propositions logical subjects; but he is faced with a contradiction which he takes to be unavoidable, a contradiction created by a presumed shift in reference which accrues to nominalization. It has been argued here that there is equivocation on the sense of "nominalization." By maintaining distinctions first drawn by Chomsky and by applying a "mixed" view of nominals derived from Chomsky's proposals, the contradiction is averted and hopefully the solution becomes generalized to related problems of propositional attitude. It is to be admitted that how a nominal is to be arrived at by transformation has not been given. Limitations of space prevent this, but a number of possibilities come to mind.

Postscript

On this the occasion of Noam's seventieth year, setting technicalities aside, we are moved to reflect on how linguistics, as he has come to have us understand it, has been undertaken in the service of what Kant regarded as the "dignity" of humankind; in matters of science and human nature, we consider that by virtue of his intellect he has inspired both courage and curiosity; and, for those of us who have benefitted, it is to be recognized that while the intellect may inspire courage rarely does courage inspire the intellect, amounting to what philosophers in search of the "cardinal virtues" might describe along with linguists as a "curious asymmetry."



BIBLIOGRAPHY


Abney, S. 1987. The English noun phrase in its sentential aspect diss. MIT.

Chomsky, N. "Remarks on Nominalization." (1970). Reprinted in Davidson and Harman. pp. 262-289.

Chomsky, N. Aspects of the Theory of Syntax 1965.MIT Press.

Chomsky, N. Knowledge of Language 1986. Praeger.

Cocchiarella, N. 1980."The Development of the Theory of Logical Types and the Notion of Logical Subject in Russell's Early Philosophy" Synthese 45. 71-115

Davidson, D and G. Harman. The Logic of Grammar (1975). Dickenson.

Landini, G. "'On Denoting' against Denoting". 1998. In RUSSELL: the Journal of the Bertrand Russell Archives. 18. 43-79.

Russell, B. The Principles of Mathematics 1903. 2nd. ed. W. W. Norton. 1964.

Russell, B. "Meinong's Theory of Objects and Complexes." 1904. Mind336-354