Only connect.

ABSTRACT

The "misplacement" of only illustrated by She only travels on foot, where there is a perceived discrepancy between the placement of only, before the verb, and its scope, the adjunct on foot, has long been controversial among the prescriptively minded; compare the approved-of juxtaposition in She travels only on foot. But the "misplacement" is also a well-established feature of Present-day English usage, which most speakers find more natural than the alternative. Here I look at the syntax of "misplacement" within the framework set forth in Anderson (1997), as well as at other aspects of the syntax of only. Traditionally, grammarians and lexicographers have distinguished an "adverbial" and an "attributive" only. I suggest here, however, that in both uses only is a specifier in terms of the adopted framework, an element whose major function is to seek to modify a head. In its "adverbial" use it is a very general specifier which allows or requires, under certain conditions (particularly the category of the specified), what I call here vicarious ("misplaced") specification. In terms of this last, elements can come to share their specifier with a governing predicator, or, in one case, a governed predicator. The attempt to clarify the syntax of vicarious specification is the major motivation behind the present investigation. In its "attributive" use only is a specifier of a quantifier category, either a periphrastic quantifier or one incorporated into the internal structure of a noun.

0. Introduction

It is a familiar observation concerning current (non-pedantic) English usage that the word only in (1a) can be interpreted as having the same scope as in (b) or (c):

1) a. Bill only enjoys cheese in France.

b. Bill enjoys cheese only in France.

c. Bill enjoys only cheese in France.

On neither of these readings for (1a) is only adjacent to the elements within its scope, unlike in (b) and (c). This usage is by now well established, despite much disparagement, under the rubric of "misplacement of only" (see Vallins 1956: 138-9 for a brief survey of some attitudes to this usage, and Nevalainen 1991: 8, 133-4 for further illustration and references); and the reason(s) for its currency is part of my concern here.

Sentence (1a) has another two readings where the scope is more directly reflected in the positioning of only. On one of these, enjoys alone is in the scope of only, and it bears the tonic or sentence accent; and this version of (1a) might be continued as in (2):

2) He only enjoys cheese in France, he doesn't worship it.

Such a reading is most transparent if it involves an item not at an extreme point on some conceivable scale. On the other reading, enjoys cheese in France is in the scope of only, and this interpretation excludes any other verb phrase which might be pragmatically appropriate; and in this case (1a) might be continued as in (3), for example:

3) Bill only enjoys cheese in France, that doesn't mean he goes in for culinary orgies.

(On these distinctions, see, e.g., the brief discussion by Horn 1969: 100-1). In each of these latter two cases the only immediately precedes the sequence in its scope, as it also does in (1b) and (c). We might roughly paraphrase these various readings for (1a) as in (4):

4) a. Bill doesn't enjoy cheese other than in France. (= (1b))

b. Bill doesn't enjoy anything in France other than cheese. (= (1c))

c. Bill doesn't do anything with respect to cheese in France other/more than enjoy it. (cf. (2))

d. Bill doesn't do anything (in France) more worth remarking on than enjoy cheese (in France). (cf. (3))

The bracketings in (4 d) discriminate a further distinction concerning whether or not in France is in the scope of only, whether its scope includes all of the "outer verbal phrase" or just the "inner".

These observations are part of the abundant evidence that there is a "slot" in sentence structure in which the scope-imposing elements that have been referred to as "focusing adverb(ial)s" (Nevalainen 1991: [section] 1.1.2.) can occur in a position that neutralises the expression of scope differences. I am concerned here with the syntax of this phenomenon as exemplified by only, while recognising that not all such elements share exactly the same syntactic possibilities, even all those among the "focusing adverb(ial)s" which, like only, just and merely, have been described as "exclusive adverb(ial)s" (cf. again Nevalainen 1991). In [section] 2.I shall be looking in detail at the syntactic properties of the preverbal "slot" illustrated in the previous paragraph. [section] 3.1. considers the role of this "slot" in the syntax of nominal constructions, particularly those involving quantification; while [section] 3.2. looks briefly at another aspect of nominal syntax, the role of vicarious specification with respect to nominal attributives. [section] 4. looks at the character of a familiar distinction, involving the existence of an apparently "attributive" role for only; in this function only does not show vicarious specification. At this point, however, I want to outline how the more straightforward behaviour of only and similar "adverbial" forms is to be accommodated within the syntactic framework developed in Anderson (1997).

1. The status of "adverbial" only

In its "exclusive adverbial" use only is clearly one of those elements that do not themselves take complements and are themselves inherently non-complements; Anderson (1997: [section] 2.8.) calls them specifiers. Basic syntactic structure, in that framework, is determined by the pattern of categorisation and subcategorisation associated with the elements, words, that go together to form a potential sentence: complements are dependent on their complementees, and basic linearisation reflects head-dependent relations. Thus, all of the non-verbal constructions in (5) are complements of the verbal:

5) John sent it to Edinburgh.

The verb is subcategorised as in (6):

6) send {P;N/{erg} {abs} {loc} }

That is, it is a non-auxiliary verb ({P:N}) which takes an ergative (agentive), absolutive (neutral) and a locative as complements, with this complementation being specified to the right of the slash in (6). (I ignore here the unexpressed ablative argument associated with directional predicators). Let me now comment on these categories.

The system of primary categories exemplified by {P;N} is built on the combination of the two notional features P (predicability) and N (referentiality), which combine in various proportions to identify different categories. {P}, i.e. the category represented as P on its own, characterises "operators" (finite auxiliaries), whose presence guarantees sentencehood to a combination of words (provided, of course, the sub-combinations are well-formed); and {N} is associated with referentials, such as names, pronouns and determiners (with these last taking nouns as complements); verbs, nouns and adjectives involve positive combinations of P and N: as in (6), the non-auxiliary verbal is represented as {P;N}, i.e. with P, on the left of the semi-colon, preponderating over N, whereas nouns are {N;P}, i.e. with preponderant N, while adjectives involve "mutual preponderance", represented as {P:N} (or {N:P}). I shall refer to any category containing P as a predicator; categories that combine P with N are lexical predicators. {P} and {P;N} are verbal (predicators). {N} and {N;P} are nominal. Erg, abs and loc are secondary categories of the functor category, the latter being represented { }, i.e. as involving absence of both features: each of {erg}, {abs} and {loc} in (6) indicates a complement requirement for the verb, involving a functor category specified as to its secondary category.

Like the other simplex combinations {P} and {N}, functor, { }, is a functional category, and may be expressed as an independent word or morphologically or by some syntactic means. In English, whereas {loc} is typically expressed by a preposition, like to in (5), {erg} and {abs} are reflected by the syntax, specifically by the fact that, in terms of the subject-selection hierarchy (Anderson 1997: [section] 3.1.1.), in the unmarked case {erg} outranks {abs} as potential subject, as actualised in (5). Each functor takes a referential as complement, by a general syntactic redundancy, functor complementation:

7) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

And each referential is complement to a functor, whether the functor is a separate word, as with to in (5), or, failing this, it is present by virtue of the general lexical redundancy, secondary functorhood, given in (8a), which allows referentials to incorporate a functor which governs them:

8) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

The vertical in (8a) represents a dependency relation that is word-internal and does not involve a difference in precedence between the upper (head) and the lower (dependent) element. Lexically, referentials thus have both of the specifications given in (8b). And the complex referential can be associated with any secondary functor category (erg, abs, etc.) whose unmarked realisation is not as a separate word, though, depending on the referential (and any dependent noun--I am ignoring this aspect here), some associations of functoral secondary category and referential will be pragmatically more likely.

Thus, the lexical representations for John, it and Edinburgh in (9) will be available, among others, in the lexicon:

9) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

10) to {loc}

These and the other relevant lexical representations determine both the viability of a sentence combining these words and the basic linearity relationships. Given the set of words in (6), (9) and (10), we have a compatible set of subcategorisational requirements; indeed, we have several sets, in principle; but some combinations result in pragmatically less likely sentences than (5), such as, perhaps, It sent John to Edinburgh, or Edinburgh sent John to it. Let us focus on the derivation of (5).

Syntactic structure is assigned to the categorisations underlying (5) as follows. Edinburgh is most obviously available as a referential to satisfy the complementation of to, expressed in (10). This means that, given possibilities available via functor incorporation, we have three functors to satisfy the requirements of read, (6), with John, via secondary functorhood, as a plausible {erg} referential, and it as an {abs }. If the various complements are syntactically made dependents of the elements whose subcategorisation they satisfy, we can derive the unlinearised syntactic structure in (11) from the collection of words in (6), (9) and (10):

11) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Each category projects a node in the tree in (11), and the nodes are connected by continuous lines representing the dependency relation projected by the complementation relationships, with the lower node associated with a complement being dependent on the higher node to which it is connected; a complementee projects a head to govern its complement(s). The non-vertical dependency lines signal the adjunctive dependency holding between individual words, and the verticals are projected by the word-internal categorial dependencies involved--in these instances, functor-referential complexes. The discontinuous lines associate the items and their categories and the categories and the nodes they project.

English is a centrifugal language: the default is for adjoined dependents to be linearised to the right of their heads. This accounts for the sequence of sent and it and sent and to, and to and Edinburgh. {abs} precedes {loc} by general rule in English and other centrifugal languages. In these respects the sequencing is (after all) as shown in the left-to-right ordering of (11). In the case of sent and John something else supervenes.

The sentence structure in ( 11 ) is incomplete (even apart from the missing linearity specifications just acknowledged). Missing is the finiteness element. In (12) this is provided by has, a finite auxiliary, or operator:

12) John has sent it to Edinburgh.

Has is of category {P} and by a general syntactic redundancy, operator complementation, it takes a {P;N} complement:

13) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

(This is part of a more general redundancy affecting all auxiliaries, all words that can function as operators). In (12) the subcategorisation is satisfied by sent, and in the syntactic structure sent will be represented as a dependent of has and (in accordance with centrifugality) as following has, as realised in (12). However, this {P} also introduces a further structurally relevant component, based on a general principle.

For, just as in some other frameworks "subject" is taken to be universally present in sentences, so, in Anderson (1997: 166-77), it is assumed that every predication must contain an abs, the unmarked semantic relation; and if the predicator, unlike send, is not subcategorised for abs, an unsubcategorised-for abs is introduced. Let us refer to this as a free abs. This free {abs} dependent may remain "empty" and lacking a dependent referential, and thus be realised by an expletive, as in It rained or We hot-footed it out of there or It seems that the biscuit tin is" empty. But it may alternatively "gain content" by virtue of sharing the argument of another functor. This is what happens in raising. Has is a predicator which receives a free {abs}, and this "empty" {abs} shares, by raising, the subject argument of the non-finite verb dependent on it. The resultant syntactic structure is as shown in (14):

14) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

John is shared between the {erg} of sent and the {abs} of has, and in its case centrifugality (with respect to sent) is overruled by subject formation, which places the non-verbal dependent of {P} on its left. In English this is, as we shall see, the specifier position, in this case with respect to {P}; and Anderson (1997: 224-36) suggests that the subject is a "syntactically derived specifier". (See also, for a discussion of different principles of linearisation, Anderson 1997: [sub section] 3.3.4., 3.7.2. I comment below, at the end of [section] 2., on the non-projectivity (tangling) associated with such representations).

(5) lacks the operator of (12). In its case the finiteness element is incorporated into the word which also realises the lexical verb, sent. (Recall that finiteness, like functor, is a functional category, and so is non necessarily realised as a separate word). This is allowed for by the optional lexical redundancy, secondary finiteness, of (15), available to lexical verbs:

15) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Sent in (12/14) is non-finite, {P;N}; sent in (5) is finite, specified as on the right in (15). This means that we should substitute for the incomplete structure in (11) the similarly unlinearised structure in (16), which incorporates a {P} element and its free {abs}, which latter shares its argument with the hierarchically highest functor of the {P;N} predicator that shares its realization with {P}:

16) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

The subcategorisation of {P} is satisfied internally, by {P;N}. And again there is a "raising" relationship involving, in this case, two predicators which are realized by a single word. Recall that (16) is unlinearised; as with (14), linearity is imposed on the basis of the information in the (sequence-free) syntactic tree, it is not changed. Nor are attachments to nodes destroyed.

(16) leads to the linearised structure in (17), where again subject-formation overrules the status of John as a complement of {P;N}:

17) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

Basic to the erection of (17), the dependencies therein and the linearisation, is the interaction of complements and complementees. There are other elements, however, which belong to neither class. Among these are the specifiers.

As a specifier, only is not subcategorised-for; it is not a potential complement. Nor is it a complement-taker: it is an optional element whose basic syntax is dependent on that of the construction within its scope. Accordingly we can, following Anderson (1997: [section] 2.8.) on specifiers, characterise only as an element that necessarily seeks a head to depend on. Lexically such an element can be represented schematically as in (18):

18) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

The back slash indicates what we might call "retro-complementation": the representation to the right of the slash in this case indicates the category of the head that the element specifies, and thus depends on. X is a variable over the various categorial possibilities. Sometimes the categorial specification of what the specifier specifies is narrow, as with very in English, which me might represent as in (19a):

19) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

That is, very requires to attach to an adjective (on the assumption that manner adverbs share this {P:N} categorisation--so we get very quickly). Syntactically, the specifier is made dependent on a node superjoined to the syntactic node projected by the element it "retro-complements", or specifies, giving a structure like that in (19b). The construction remains adjectival, headed by an adjective. Only appears to be categorially more promiscuous in its choice of head: only large men, only slightly late, only later, the only survivors, only takes a minute, only on Sundays. But it entails the same structural addition:

20) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

Only is represented here as a completely general specifier: determining limitations on this is one of our concerns here. Return now to the examples involving only given initially:

1) a. Bill only enjoys cheese in France.

b. Bill enjoys cheese only in France.

c. Bill enjoys only cheese in France.

We can represent the structure of (1a), on the variant of interpretation (4d) in which in France is within the scope of only ("All Bill does is enjoy cheese in France"), as in (21), which ignores, e.g., the internal categorisation of cheese:

21) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

With in France outwith the scope of only ("All Bill does in France is enjoy cheese") we have (22):

22) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

Bill is the {erg,loc} ("experiencer") argument of the predicator enjoys, which also takes an {abs}. In France is a circumstantial: like a specifier, it is not a complement, but rather seeks a category to modify; here I've taken it to be {P;N}, which thereby has another node (apart from that triggered by only) added above the one it projects in the syntactic tree. Unlike basic specifiers, circumstantials are potential complements. And their serialisation is different: in the present case in France shows the default linearisation: it follows its head. We are not concerned here with the internal structure of the locative {N}. In the syntactic tree only adds a node above the head of the construction that falls within its scope. The two interpretations correlate with whether only is attached above or below the circumstantial.

Likewise, with respect to (1b) and (c), the specifier inserts a node above the head of the following construction, which lies within its scope, as in (23) and (24), respectively:

23) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

24) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

Only thus specifies functor phrases as well as verbal, and, as a specifier, creates the structure in (21).

I shall address below the question of the representation associated with reading (4c) of (1a), wherein only the predicator is within the scope of only. This involves distinct issues to do with the character of the specification of non-maximal projections.

All of the present representations, (21) through (24), conform to the generalisation that the construction to their immediate right, above whose head they insert a syntactic node, is within their scope. But what is for most speakers the most salient interpretation of (1a), namely that it shares with (1b), violates this generalization, as does the interpretation shared with (1c): its scope in both instances is a non-adjacent subpart of the construction that follows. This discrepancy is the concern of the section that follows. In it I do not attempt to survey the attempts that have been made within various frameworks to account for the apparent discrepancy in linear placement associated with the most salient interpretation of (1a) and the like. See Nevalainen (1991: [sub section] 2.2.1-2.), however, for a short review of the various abortive attempts at analysis emanating from the generativist tradition in the 70s.

2. The preverbal slot for only

In (1a), interpreted as is (1b) or (c), the specifier comes as usual to the left of the construction in its scope, but not imlnediately to its left; rather, to the immediate left of the governing predicator, in the position normally adopted by the specifier of the latter. That this vicarious specifier slot is specifically that associated with {P;N}, a lexical verbal predicator, is confirmed by the dubious character of (25b), in the sense of (c):

25) a. He might only eat the cheese.

b. ?* He only might eat the cheese.

c. He might eat only the cheese.

This is in contrast with the syntax of circumstantials like usually, which can precede either {P}, the operator, or {P;N}:

26) a. He usually (?* only) will (only) eat the cheese.

b. He (?* only) will usually (only) eat the cheese.

The vicarious slot can be associated with any {P;N} in the clause:

27) a. He may have only liked the cheese.

b. He may only have liked the cheese.

c. ?* He only may have liked the cheese.

d. He may have liked only the cheese.

28) a. He might have been only deceiving Molly.

b. He might have only been deceiving Molly.

c. He might only have been deceiving Molly.

d. ?* He only might have been deceiving Molly.

e. He might have been deceiving only Molly.

Non-operator auxiliaries are {P;N} rather than {P}: thus, modals in English are always {P}; the other auxiliaries may be {P} (as in (11)) or {P;N} (as in (27/28)), and are distinguished lexically from other verbals by precisely this dual categorisation.

We can accommodate this phenomenon if we associate with a non-operator verbal predicator the capacity to provide a vicarious specifier node to any element dependent on that predicator, as in (1a), (25a), (27a) and (28a). In (27b), (28b) and (28c), this provision occurs recursively, so that only comes to be associated not just with the predicator on which the cheese depends, but also on the predicator that governs that predicator; and, in the case of (28c), only comes to be associated with three successive predicators. The application of this vicarious verbal specification is illustrated by the representations for (27a) and (27b) given in (29), with the latter showing two instances of such vicarious specifiers:

29) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]

As with subject-formation, it is the position of the upper, "derived" specifier that determines the placement of the shared element, in this case only (Anderson 1997: 312). And, again, there is no movement: linearity is established on the basis of the categories and dependencies established in the syntactic tree.

We can formulate vicarious verbal lexical predicator specification as the optional syntactic redundancy in (30):

30) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

which permits only to seek additional specifier positions to that for which it is categorized, specifically verbal specifier positions which are attached to predicators which govern {X}, the category which only is lexically marked as specifying, and which in this particular instance "{X[}.sub.i]" it is specifying. (30) is strictly local. Here I stipulate this (the locality imposed by the dependency relation), by deploying the "i-subscript", for transparency; it is, however, the unmarked assumption, and need not be included in the rule itself. Given this locality restriction, we can also say that (30) applies only to an only which specifies a functor or a predicator, since (according to Anderson 1997: [section]3.6., for example) only functors and predicators depend directly on verbal predicators, as required by (34).

We also find examples of vicarious specification involving the non-vicarious specifiers of non-verbal predicators as inputs, as in (31):

31) a. So far he has only been jealous--cf. So far he has been only jealous.

b. So far he has only been a nuisance--cf. So far he has been only a nuisance.

But, again, the operator ({P}) does not allow vicarious specification:

32) a. He (?* only) is (only) jealous.

b. He (?* only) is (only) a nuisance.

Moreover, non-verbal predicators can themselves also provide a vicarious slot:

33) a. He is only afraid of spiders.

b. He is only a friend of the rich.

This is not so common as with verbal predicators, given the more limited argument structures typical of adjectives and (particularly) nouns.

We can thus sharpen up (30) a little as (34), vicarious lexical predicator specification:

34) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

"< >" indicates optionality; and "{P,N}" characterizes the class of lexical predicators, where "," generalizes over ";" and ":". That is, "{P,N}" indicates any combination involving both P and N--thus verbs, nouns and adjectives. "{<P,N>}" therefore includes all combinations of both P and N and neither P nor N--verbs, nouns, adjectives and functors.

The specifiers of subjects are excluded from acquiring vicarious specification; (35a) and (b) are not equivalent on any interpretation:

35) a. Only John has learnt about it.

b. John has only learnt about it.

By virtue of subject-formation, only John in (35a) comes to depend (as in (16/17)) on a component predicator that, contrary to the requirements of (34), governs the predicator which would provide the point of attachment for vicarious specification. It is tempting to see this as the basis for the exclusion. However, there are indications in earlier texts that subjects were once eligible for vicarious specification:

36) a. The oldest son shall only inherit his father.

b. You have no clause before you, only the word, "declare", that word is only moved.

(see OED: entry for only; (b) is noted by Nevalainen 1991: 133). This suggests that avoidance of "inversion" of the dependency relation shown in (34) is not a necessary part of the conditions governing the rule. At best such avoidance is a parochial property of the Present-day English treatment of this construction. This is supported by consideration of another apparent exclusion from vicarious specification.

(34) also excludes {P} both as a source of vicarious specification and an undergoer of the process. We shall have to look more carefully at this latter aspect of (34). We should register, first of all, that only can have a construction headed by {P} within its scope; it can specify {P}. As well as there being readings of lesser scope for only in (37a), for example, on one interpretation, {P} is within the scope of only, with roughly the meaning of (b):

37) a. John only abandoned Mary.

b. All that happened was that John abandoned Mary.

But, even though (38) permits a similar interpretation, only occurs in a position that is clearly that of the specifier of {P;N} rather than {P}, whereas its position in (37a) could be said to be ambivalent:

38) John has only abandoned Mary.

That is, it looks as if only in (38) necessarily occupies a vicarious specifier slot, and in (37a) it could do. We should perhaps generalise that, whereas vicarious specification (34) is optional, its equivalent applies to {P} obligatorily. So that even (37a) involves a representation like that in (39):

39) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

(Treatment of Mary here as a simple {abs} is no doubt a simplification, but this need nor concern us in the present context). Representation (40) for (38), where again the whole construction with {P} as its head is in the scope of only, makes overt the operation of vicarious specification:

40) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Here, with application of vicarious lexical predicator specification of {P}, the predicator of the vicarious specifier is governed by that of the basic specifier, not vice versa, as in (34):

41) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

But again the dependency requirement ensures that the structure addition is determined strictly locally. And it is once more the attachment of the "derived" specifier that determines placement.

What seems to be involved in vicarious specification is at least in part a manifestation of the results of a diachronic tendency for (particularly verbal) predicators to attract to themselves "light" elements which may cliticise or indeed morphologise: elements associated with the dynamic character of verbal predicators (tense, aspect, stativity), with their "mode of existence" (modality, negation), with the status of their arguments (transitivity, voice) or the character of their arguments (concord): cf., e.g., Anderson (1985: [section] 2.2.), who remarks that "in most languages, a large part of the complexity of word formation (and inflexion in particular) concerns the verb". In this case, the predicator attracts to itself the specifier of its dependent functor or predicator (34), or the specifier of the {P} that governs it (41). These observations remain informal, but the weight of evidence for the "tendency" is formidable. Moreover, in English this position is already an established circumstantial adverbial slot. And, on the other hand, immediately post-verbal only is in a linear position disfavoured by circumstantials, particularly if an object follows (* She ate slowly the spaghetti).

There is perhaps also a particular motivation in the present instance for the "usurping" of the {P;N} specifier position: For semantic reasons only-specification of an argument or of the predicator alone is commoner than only-specification of the {P;N} construction as a whole. Consider again (1a):

1) a. Bill only enjoys cheese in France.

The readings on which only takes enjoys cheese (in France) in its scope is much less salient than either the vicarious readings, equivalent to (1b) and (c), or the reading with only enjoys in the scope of only, i.e. (4d). This is partly because in these latter cases the paradigmatic alternatives to the elements in the scope of only are more obvious: one has to hunt around for what is excluded by "enjoys cheese (in France)" (cf. Horn 1969). Thus sentences like (1a) seldom in practice present us with much ambiguity: the vicarious readings are the normal ones, unless the predicator bears the tonic, signaling that it alone is within the scope of only, as indicated in (4d). Moreover, as Fowler reminds us, there are often rhetorical reasons for placing an only earlier than the element in its scope: "the orthodox It would be safe to prophesy success to this heroic enterprise only if reward & merit always corresponded cries out to have its only put early after would, & unless that is done the hearer or reader is led astray" (1926: 405). And compare here Nevalainen (1991: 134-5).

Vicarious specification involves a shared argument, and often non-projectivity (tangling), as in, say, (29b), where the two lower arcs terminating in a node that is associated with only cross (other) association lines. These involve departures from the unmarked assumption, that syntactic representations are proper, projective trees. However, the present proposal does not add significantly to the repertoire of violations theoretically permitted. Subject-formation creates analogous configurations, as again illustrated by (29b), where once again the two lower arcs terminating in nodes associated with he intercept association lines. The "derived" specifier, subject, involves an "empty" {abs} serialised to the left of {P}. The "misplaced" only occupies an empty "derived" specifier position to the left of {P,N}, and particularly {P;N}.

Let us turn finally in this section to the representation of those only constructions in which the predicator is signaled by the intonation to be alone in the scope of the only. Take a sentence like (42a):

42) a Bill only likes Fifi these days.

b Bill only likes Fifi these days.

c Bill only likes Fifi these days.

Here only the predicator and not the {P;N} construction as a whole is within the scope of only. We can represent this as in (43):

43) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

In this case the scope of only is sub-syntactic: only "retro-complements" individual lexical items ("I" in the representation for only in (43)) rather than categories. Thus it introduces a node below the syntactic node projected by {P;N}. I shall refer to this kind of representation as characterising a lexical reading. These characteristically involve items that can be placed on a scale with respect to which they are not the strongest item: He only likes Fifi--he's not in love with her (Horn 1969: 101-2).

Lexical readings are not categorially restricted, and vicarious specification is very generally available, as illustrated by (42b). Subjects are again excluded from vicarious specification: (42c) does not involve vicarious specification of Bill by only. With lexical readings only may even have sub-lexical-item elements as its scope, as in (44), which also illustrates, in (a), vicarious specification:

44) a. I'm only advocating prosecution, not persecution.

b. Helen only was a spy.

(Horn 1989: 435; Nevalainen 1991: 47; Ross and Cooper 1979: 413). In (44 b) the scope of only is apparently the tensing. I do not pursue this aspect of the behaviour of only here, recognising that it is a massive topic in itself.

3. Nominals and "adverbial" only

Given the formulation in (34), referentials ({N}) are also excluded from vicarious predicator specification. They are not dependents of predicators, given that a functor always intervenes between predicator and referential (recall the discussion following (6) above); and they do not involve a combination of both or neither P and N. There is, moreover, empirically no need to appeal to (45a) as the non-vicarious version of (45c), given the acceptability of (45b), which, indeed, some speakers prefer to (45a):

45) a. Hetty is fond of only Bert.

b. Hetty is fond only of Bert.

c. Hetty is only fond of Bert.

d. Hetty is only fond of Bert.

((45d) is a lexical reading.) And not regarding it as such enables us to maintain the more restrictive version of vicarious predicator specification. Moreover, the putative version with putative simple referential specification is often much worse than (45a):

46) a. ?* She practices with only professionals.

b. She practises only with professionals.

c. She only practices with professionals.

d. She only practices with professionals.

((46d) is gain lexical). Referential specification by only seems to be very marked.

It looks as if we are arriving at a hierarchy of eligibility for (categorical, as opposed to item) specification by only, and for vicarious specification. Only-specification of functor phrases, headed by { }, is very generally feasible, though vicarious specification often yields more "natural" sentences, as with (45c) vs. (b). Specification of {P} by only is likewise general, but vicarious specification now appears to be obligatory, as in (37a), on the appropriate reading of that sentence (i.e. as in (37b)):

37) a. John only abandoned Mary.

b. All that happened was that John abandoned Mary.

Such a reading, however, is often less salient than one with vicarious specification of a functor. Only-specification of lexical predicators, and particularly verbal predicators, is common, but predominantly as manifestations of vicarious specification, particularly of a functor phrase; simple specification of {P,N} is unusual, and often highly contextualised, as with interpretations (4d) for (1a):

1) a. Bill only enjoys cheese in France.

4) d. Bill doesn't do anything (in France) more worth remarking on than enjoy cheese (in France).

We are now finding, cf. (46), that specification of (N} by only is often very awkward, not just difficult to contextualise, as illustrated by (46a), or (47a):

47) a. ?*She listens to only Bert.

b. She listens only to Bert.

c. She only listens to Bert.

I am therefore assuming that (47c) is the vicarious congener of (47b), not of (47a), and that in (25c) only specifies the incorporated funetor category rather than the referential:

25) a. He might only eat the cheese.

c. He might eat only the cheese.

(25a) is the vicarious equivalent of the functor-specifying (25c). And vicarious specification does not apply naturally to referentials. But we must consider now some apparent exceptions.

Before proceeding to these, I present in the table a summary of some tentative conclusions concerning only-specification.

Table 1. Distribution of only-specification and vicarious specification

        Only-specification        Vicarious
{ }     [check] [check] [check]   optional
{P}     [check] [check]           obligatory
{P,N}   [check]                   (target)
{N}     ?*                           --

3.1. Quantification and vicarious specification

Quantifier referentials, including numerals, are exceptional as {N}s in both the respects just mentioned. They are freely specified by only, as in (48a), beside which (48b) is rather stilted:

48) a. She listens to only some of his advice.

b. She listens only to some of his advice.

c. She only listens to some of his advice.

And there seems to be no reason, unlike with (46), not to regard (48c) as the vicarious congener of (48a) rather than the functoral (48b), particularly given that many putative functoral equivalents are low in acceptability, as illustrated by the pattern in (49):

49) a. He travels with only two of them.

b. ?*He travels only with two of them.

c. He only travels with two of them.

But in that case vicarious specification would in this instance not involve a dependent of {P;N}, as required by (34). Both these exceptional aspects of quantifiers {N}s can be related to a facet of quantifier syntax that, after all, will enable us to maintain the view of vicarious specification that we have built up--as resisting referential inputs.

Quantifiers are subordinate not just to the lexical predicator whose subcategorisation is satisfied by the functor phrase they occur in but also simultaneously to an existential predicator superordinate to that predicator (cf. Anderson 1997: 307-13); they are shared arguments. This existential is introduced by the lexical redundancy in (50):

50) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

which optionally introduces an existential extension above any verbal predicator. The {[P.sup.E]} may be expressed analytically, as a form of be, as in (51a):

51) a. There are two of them he travels with.

b. He travels with two of them.

But in (51b) it is incorporated as part of a complex predicator, as shown in (52):

52) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

By virtue of quantifier-raising, the functor phrase containing the quantifier shares its argument, the quantifier, with the free abs dependent on the existential predicator; the basic verb is an agentive intransitive ({"/{abs,erg}"), with a circumstantial containing a quantifier; prt is the partitive functor (see further [section] 4.). Existential sentences lack subject formation in accordance with the subject-selection hierarchy; even in the overtly expressed existential (51a), the subject is expletive (cf. again Anderson 1997: 308-13).

This means that in (49a) only can be interpreted as specifying not the {N} but the existential {abs} that governs it; after all, quantifiers are not exceptional with respect to the awkwardness of only-specification with referentials. We can represent the structure of (49a) as in (53):

53) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

We thus can maintain the generalization that only-specification is least compatible with referentials, in accordance with the hierarchy sketched out above. The apparent exceptions with quantifier-headed phrases involve specification of the existential functor phrase with which the quantifier is associated.

What of the vicarious version (49c)?

49) c. He only travels with two of them.

As formulated, vicarious specification applies to dependents of {P;N}, or, more generally of {P,N} (any lexical predicator). The construction affected in (49c) is not a dependent of {P;N}, but of {P}, in particular {[P.sup.E]}. Suppose, however, we generalize vicarious specification to a dependent of any category which includes P in its categorization, thus any predicator; i.e. we substitute for (34) a formulation of vicarious specification such as that in (54):

54) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

(where "{P,}" allows for the null combination, and "X" conforms to the hierarchy in the table). This allows the existential {P} to provide a vicarious specifier slot for only. But specifiers of {P} obligatorily require vicarious specification by a {P;N}, as formulated in (42), vicarious lexical predicator specification of {P}. This is repeated here, and somewhat generalised, as (55):

55) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

(55), what I shall refer to as predicator specification of {P}, requires that any only-specified {P} has a vicarious specifier associated with any dependent {P,}, predicator. This can apply recursively, as we shall see, and only terminates when the vicarious specifier is dependent on a lexical predicator. At this point the conditions for (55) will not be met: the input is not "{ \{P/{P, }}}".

We can thus represent (49c) as in (56):

56) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Here only specifies the existential {abs}, the existential predicator itself (again, unlike in (51a), not expressed overtly), the {P} below that and the most inclusive construction headed by {P;N}. The topmost syntactic node above two is that required by the specifier of two. The nodes above travels are respectively, from the top, the head node introduced by the specifier of the existential predicater, the node projected by the existential predicator, the head node required by the specification of the lower {P}, the node projected by the lower{P}itself, the head node introduced by the specifier of {P;N}, the head node introduced by the circumstantial with two of them and the node projected by {P;N}.

Apparent only-specification of quantifiers and vicarious specification thereof involve rather specification of the {abs} of the existential predicator which is subcategorisationally satisfied by the quantifier phrase. Referentials themselves in general are not "happily" specified by only, as embodied in the formulations in (34) and (54).

3.2. Attributives and vicarious specification

Only-specification of referential phrases varies from the "awkward" to the clearly unacceptable, and there seems to be no need to allow for vicarious specification of them. But we must apparently provide for vicarious specification of subparts of referential phrases, specifically attributives. Consider here (57a):

57) a. He only enjoys flowers with a scent.

b. * He enjoys flowers only with a scent.

c. He enjoys only flowers with a scent.

d. Only flowers with a scent please him.

On one reading of (57a) with a scent is in the scope of only. Indeed, here vicarious specification is obligatory, as shown by the badness of (57b). But notice also that although (57c) shows simple specification of the functor phrase on one reading, on another only with a scent is in the scope of only. The same structural properties as with quantifiers provide for vicarious specification. That is, we have alternative vicarious attachments: one is the standard one to {P;N}, illustrated by (one interpretation of) (57a); the other is a sort of "halfway-house", where the modifier of flowers is specified from pre-functor position, as instantiated by (one interpretation of) (57c), or of (57d).

With a scent should not be accessible to standard vicarious specification, as formulated in (54), repeated here:

54) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Its relationship with the predicator is not strictly local; it is not a direct dependent. This suggests that the vicarious structure associated with (57a) may be fed by that associated with (57c). (57c) instantiates a new species of vicarious specification which makes available the pre-functor slot to nominal modifiers. Appeal to this species also seems to be necessary to account for (the relevant interpretation of) (57d).

If on one reading (57c) and (57d) show vicarious specification of with a scent, the appropriate partial representation would be as in (58):

58) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

As an attributive, thus circumstantial to a nominal, with a scent introduces a node above the node projected by the {N} of the complex nominal realised as flowers. Here {N} is incorporated as part of the word flowers, by virtue of an analogue to secondary finiteness (15), secondary referentiality. As a functional category, {N} can also be expressed periphrastically, as a quantifier, for example, or as in a scent--though I am not concerned at this point with the internal structure of these nominal phrases: we return to them in [section] 4. (and see further Anderson 1997: [section] 3.7.), and in particular the role of the semantic relation prt, partitive.

Vicarious functor specification then takes the form of (59):

59) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Thus, an only specifying a "retro-complement" of a {N} comes, obligatorily, also to specify the functor on which the {N} is dependent. Here too the relationships invoked are local, involving a shared {N}.

The output from (59), in this case (58), can then feed vicarious specification by a predicator, (54), if (58) is not in subject position: i.e. it occurs as in (57c) not (57d). Application of (54) gives the structure associated with (the relevant interpretation of) (57a), which we can represent as in (60):

60) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Here we have two vicarious specifications, one the result of (59), which applies obligatorily, the other, higher one resulting from (54). (57d), with only flowers with a scent in subject position, has, as expected, no equivalent with vicarious specification of a predicator, unlike (57c).

4. "Attributive" only

Only has, and has had, other senses, and categorisations, than the "adverbial" ones that have been our focus so far. It behoves me to note the most salient of these, the "attributive" use, a distinction long recognised by lexicographers (cf., e.g., Johnson 1843), without, however, its character necessarily being agreed on (cf., e.g., Rissanen 1985: 222, note 8; Nevalainen 1991: 130). This use is illustrated by (61a), for which the sentence in (b) does not constitute a vicarious congener:

61) a. She landed on only two engines. [not equal]

b. She only landed on two engines.

(61b) is perhaps most obviously to be interpreted as showing a simple specification of {P;N}. In (62) we find both "adverbial" and "attributive" only:

62) a. She only landed on only two engines.

b. She only wants to land on only two engines.

c. She wants to land only on only two engines.

(62b) neutralises the expression of several potential specifier interpretations one of which (foolish though it may be) is associated with (c).

"Attributive" only characteristically precedes an overt quantifier (as in (61/62)) or follows the (or a possessive). Compare with (61a) the examples in (63) with preceding the:

63) a. He had met the only survivors.

b. She avoided the only infants with blond hair.

c. The only three infants with blond hair have disappeared.

63c) shows both a the and a quantifier. Following the such an only is clearly nominal-phrase internal, and even when preceding a quantifier it is clearly bound to the latter, though it may be postposed to the nominal phrase:

64) She landed on two engines only.

In neither case, pre-quantifier or post-the, is vicarious specification possible.

I have suggested that the traditional "adverbial" use of only is specifically a "specifier" use. We must now consider whether the present use, or set of uses, is appropriately referred to as "attributive".

As observed in [section] 3.2., attributives, like specifiers and verbal circumstantials, introduce a syntactic node above the node projected by the element they "retro-complement", as shown in the partial structure in (65):

65) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

As with with a scent in (60), the attributive, here red, "retro-complements" the head of an incorporated quantifier (partitive) structure, and the latter complements the. Let us look a little more carefully at what I have called the "partitive" structure.

I am assuming that there is a general optional lexical redundancy of secondary referentiality available to nouns, of the form of (66):

66) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

This partitive {N} may be of different types, reflecting dimensions and sub-dimensions such as count/mass, singular/plural; and its character is reflected in the number etc. of nouns and aspects of their syntax. The {N} in (66) is optional: the partitive configuration may be attached to an overt quantifier, as in (71) below, and the {prt} may be overtly expressed, as in some of them, as represented in (52). In these instances we have periphrastic expression, as an overt quantifier, of this functional category.

Red is categorised as an adjective and as such can be attributive, as made explicit in (65); in this case it "retro-complements" the {N/{prt}} of dress. This departs markedly from the proposals made in Anderson (1997: [section] 3.7.3.); but this is not the place to show that expressivity has not been lost; what is being proposed here is indeed rather more conventional.

Like verbal circumstantials, and unlike specifiers, attributives are potential complements; the typical attributive in (64/66a) can also be predicative, given its basic {P:N} categorisation, complementing the copula, as in (67a):

67) a. The dress is red (cf. the red dress).

b. * The survivors are only (cf. the only survivors).

This is not the case with the only of (67b). There are a few well-known "attributive adjectives" that apparently may not be predicative. Contrast (68a) and (b):

68) a. The present/late/mere vice-president was ignored.

b. The vice president was present/late/ * mere.

c. She is a present/late/mere vice-president.

Present and late in (68b) do not bear the same senses as in (the most obvious readings of) (68a), and mere is simply unacceptable. But they may function as part of a predicative phrase, as in (68c). Whatever one makes of this (Anderson (1997: 304-5) suggests that they involve a transitive construction), it is not a distribution shared with only. The most obvious interpretation of (69) involves only specifying a predicator, and any interpretation parallel to (68c) is forced, perhaps based on the model of the idiomatic only child:

69) They are only survivors.

Likewise (70a) is much more likely than (b), confirming the specifier-of-predicator rather than 'attributive' status of only in predicative phrases:

70) a. I'm only a survivor.

b. I'm an only survivor.

What then is the status of only in (61) and (63)?

61) a. She landed on only two engines.

63) a. He had met the only survivors.

b. She avoided the only infants with blond hair.

c. The only three infants with blond hair have disappeared.

Let us look first at (61a).

Here only is plausibly the specifier of the quantifier: it is "{ \{N/{prt}}"; so that we might represent the second nominal in (61a) as in (71):

71) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Only is categorised in the same way as attributives, except that, crucially, it is not specified for primary category, and so lacks the predicative possibility associated with, say, red in (65).

We can paraphrase the only that specifies quantifiers as "no more than": (61a) = "She landed on no more than two engines". The more general specifier we have previously been looking at invites other paraphrases (though the paraphrases, admittedly, do overlap), such as '"(did) nothing other than"; so that one sense of (70a) = "I am nothing other than a survivor". Recall too the initial discussion of the different senses of (1), as illustrated in (2)-(4).

Sentences with vicarious specification of a quantifier, such as (49c), and their non-vicarious congener, (49a), allow either kind of paraphrase:

49) a. He travels with only two of them.

c. He only travels with two of them.

Thus: "He travels with no more than two of them"/"He travels with none other than two of them". This raises the question of what is the difference between (49c), which allows vicarious specification and the "other than" interpretation, and (61a) which allows neither. I suggest that though ((49a/c) are associated with an existential predicator, as represented in (53), (61a) is not existential (cf. Anderson 1997: 311). This is revealed by the contrasting periphrases corresponding to the salient interpretations of these sentences:

72) a. There are only two of them he travels with.

b. It was on only two engines that she landed.

With (49a/c) and (72a) the quantifier is associated with the free {abs} of an existential predicator, and the vicarious specification of (49c) is possible. In (61a) the quantifier phrase is not attached to an existential {abs}, but is dependent solely on the {loc} functor; and no vicarious specification of the quantifier is possible.

Let us turn now to (63). The definite article therein, in combination with only, insists on "uniqueness": the "three infants with blond hair" of (63c) constitute a unique set. So that we might paraphrase (63c) as "The unique three-membered set of infants with blond hair has disappeared". A representation such as (73) seems to be appropriate:

73) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

The attributive with blond hair introduces a syntactic node above three, which is itself categorised as, a quantifier, a {N} that takes a partitive {N} as its complement. Again, the only specifies the "{N/{prt}}". (We are not concerned with the internal structure of blond hair).

In cases like (63a), with no periphrastic quantifier, only specifies the "{N/{prt}}" component, the incorporated quantifier, of the noun, as in (74):

74) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

We might paraphrase (63a), again rather cumbersomely as "that group more than whom didn't survive". The presence of the allows the only to specify not just a periphrastic quantifier but also, as here, an internalised quantifier category. It seems, on the basis of what we've looked at in the present subsection, that we should distinguish a specifically quantifier-specifying only, thus one with an attributive-like distribution, from the general specifier discussed in [section] 2. and [section] 3.1., which latter attaches to a range of categories and allows vicarious specification. It is with this distinction that we can associate the different "shades of meaning" with temporals that Poutsma (1928: 457) draws our attention to:

   When denoting a relation in time, it occurs in two shades of
   meaning, viz.: [alpha]) no longer ago than (Dutch nog), [beta]) not
   before (Dutch eerst). In the first its ordinary place is before the
   adverbial adjunct, in the second before the verb. Compare I saw him
   only yesterday with I only saw him yesterday.

But we must approach this indirectly, via some related observations.

It has been observed more recently (Jorgensen 1974; Taglicht 1984: 153-4; Nevalainen 1991: 42-3) that the ambiguity of (75a) is resolved by (in terms of the present proposals) the vicarious specification in (b) (albeit (b) also introduces other, verbal readings):

75) a. They received the telegram only two hours later.

b. They only received the telegram two hours later.

(75a) and (b) share a "not before" reading; (75a) but not (b) has a "no longer ago than" or "no more than" reading. The latter is what we would associate with specification of the quantifier. The former is associated with specification of the functor phrase containing the quantifier, which may have vicarious specification apply to it. This is confirmed by a comparison of examples with sentence-initial temporals such as those in (76):

76) a. Only after two hours did they receive the telegram.

b. After only two hours they received the telegram.

(76a) has a "not before" reading, and the specifier is appropriately placed for having the functor phrase in its scope; and preposing of the negatively specified functor phrase triggers inversion. (76b) has a "no more than" reading associated with specification of the quantifier and not the functor phrase as a whole, whose preposing thus fails to trigger inversion.

Konig (1981: 119) notes a similar ambiguity in:

77) Only twenty pounds would solve my problems.

The "no more than' interpretation is associated with specification of the quantifier: "as little as twenty pounds would solve my problem". But there is also a "nothing other than" interpretation, which reflects only-specification of the subject functor. Accordingly, only this latter, negatively-specified-functor reading survives in the "non-assertive" context of (78):

78) Only twenty pounds will be any help.

Only is not a typical "negative-polarity" item (Nevalainen 1991: [section] 2.3.3.), but in this respect it conforms to the expected pattern, provided that it is accessible to the non-assertive item--that is, under the "nothing other than" reading, as specifier of the functor rather than the functor-phrase-internal quantifier. Some consequences of the distinction between quantifier-specific only and the general specifier can be observed here as well as in the temporal cases.

However, the account offered here is still incomplete. And one aspect of the incompleteness returns us to Poutsma's observations, quoted above. For notice that a "no more than" interpretation is possible not just with overt quantifier phrases but also with strongly scalar items, like that in (one interpretation of) Poutsma's (1928: 457) example replicated as (79a):

79) a. I saw him only yesterday.

b. I only saw him yesterday.

or in (80a):

80) a. He retired only last year.

b. He only retired last year.

Arguably, such temporals in such a use include a quantificational structure, though: "only yesterday" = "only one day before the present one", and "only last year" = "only one year before the present one". So I shall continue to distinguish the more restricted specifier as a specifier of quantifiers, overt and incorporated.

But there is another problem. Poutsma (1928: 457) describes the placements illustrated in (79) as representing the "ordinary" places for only when associated with respectively a "no longer ago than" and a "not before" interpretation. But he immediately goes on to concede that "this nicety of word-order is not, however, always observed". Indeed, contrary to what was observed in relation to (75), the vicarious variants in (79b) and (80b), as well as the non-vicarious (79a) and (80a), are ambiguous between a "no longer ago than" and a "not before" reading. Only a "not before" reading is available to the vicarious (75b). I associated this with the unavailability for vicarious specification of the quantifier construction ("no longer ago than') as opposed to the functor ("not before"): only the functor and not the quantifier directly depends on the predicator which provides the vicarious slot. But the vicarious versions of (79b) and (80b) can be associated with a "no longer ago than" reading.

The resolution of this discrepancy is along similar lines to those described in [section] 3.1 in relation to accounting for the exceptional status of quantifiers with respect to specification by "adverbial" only, involving a superordinate existential predicator. Here we are concerned with focus rather than existence. What I suggest is this: only in (79) and (80) specifies, on the "no longer ago than" reading, specifies the {abs} of a superordinate focus predicator rather than directly the quantifier-including configuration. In (79) and (80), on a "no longer ago than" reading, the quantifier phrase is focused; the quantifer is associated with the {abs} of a focus predicator, and so available for vicarious specification. In (75) it is the whole phrase two hours later that is in focus, not just the quantifier subpart, which is thus not associated with the focus {abs} and so not eligible for vicarious specification. Compare the periphrastic variants in (81):

81) a. It was only two hours later that they received the telegram.

b. It was only yesterday that I saw him.

In (81a) it is clear that the whole phrase only two hours later is in focus, and that only is associated with only a part of that phrase, which is thus not accessible to vicarious specification. In (81b) the quantifier-incorporating noun itself is associated with the {abs}of the focus predicator, and the specifier of this {abs} is available for vicarious specification, as shown in (82):

82) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

Here, to avoid complications, see is treated as a simple experiencer verb; its precise character is not germane to our present concerns. The internal structure of yesterday remains speculative, and incomplete.

5. Conclusion and a brief retrospect

I have distinguished here two major uses of only, largely along rather traditional lines, whereby there have been differentiated an "adverbial" and an "attributive" only. I have suggested, however, that in both uses only is a specifier. In its "adverbial" use it is a very general specifier which allows or requires what I have called vicarious specification. In terms of this last, elements come to share their specifier with a governing predicator, or, in the case of {P}, with a governed predicator, or, in the case of attributives, with a governing functor. In each case the relationships involved are strictly local. The attempt to clarify the syntax of vicarious specification was the major motivation behind the present investigation.

In its "attributive" use only is a specifier of a quantifier category, either a periphrastic quantifier or one incorporated into the internal structure of a noun. Various observations concerning the syntax of phrases containing only are readily relatable to the "adverbial" vs. "attributive" distinction.

The growing currency of only over the Modern English period, largely at the expense of but, has been thoroughly documented by Nevalainen (1991). The form goes back only to early Middle English, but can be seen as a continuation of the "adverbial" use of ane (Rissanen 1985). This latter, according to Rissanen, represents the "adverbialisation' of postposed "attributive" an 'one', once the morphology failed to reflect the agreement of an with the noun to which it was attributive. In Old English, the forms of an are only "attributive". Pre-modifying an is preceded always by a definite form; and post-modifying an agrees with the preceding noun. The Old English pre-modifying usage is the ancestor of the Modern English one instantiated by (62/74). In one respect, Present-day English is more restrictive, in that demonstratives do not license pre-modifying only (though Old English, of course, did not formally distinguish article and (distal) demonstrative); compare the sentence from Addison cited by Rissanen (1985: 254):

83) She was turned into a Man, and by that only Means avoided the Danger.

But the availability to only of specification of a quantifier illustrated by (60/71) extends the range of pre-nominal possibilities. The "adverbialisation" of postmodifying an, confirmed by its adoption of an immediately preceding specifier position, establishes an/only as an "exclusive adverbial", a grammatical class not well represented in Old English. And overt-quantifier-specifying only retains the postposing possibility, as illustrated by (64), repeated here:

64) She landed on two engines only.

The development of vicarious specification made available a more flexible syntax for only, one that could satisfy rhetorical demands, for instance. Its dominance as an "exclusive" over the late Modern English period depends on a number of factors, however (cf. again Nevalainen 1991).

Note

This paper has benefited from the helpful comments of Fran Colman and Graeme Trousdale. It was substantially completed while on sabbatical leave from the University of Edinburgh, third term 2000-2001.

REFERENCES

Anderson, John M. 1997 A notional theory of syntactic categories. Cambridge: Cambridge University Press.

Anderson, Stephen R. 1985 "Inflectional morphology", in: Timothy Shopen (ed.), 150-201.

Binnick, Robert I.--Alice Davison--Georgia M. Green--Jerry L. Morgan (eds.) 1969 Papers from the fifth regional meeting of the Chicago Linguistic Society. Chicago, 1L.: Chicago Linguistic Society.

Cooper, William E.--Edward C. T. Walker (eds.) 1979 Sentence processing: Psycholinguistic studies presented to Merrill Garrett. Hillsdale, N J: Lawrence Erlbaum Associates.

Eaton, Roger--Olga Fischer--Willem Koopman--Frederike van der Leek (eds.) 1985 Papers from the Fourth International Conference on English Historical Linguistics. Amsterdam: Benjamins.

Eikmeyer, Hans-Jurgen--Hannes Rieser (eds.) 1981 Words, worlds and contexts: New approaches to word semantics. Berlin: Walter de Gruyter.

Fowler, H. W. 1926 A dictionary of modern English usage. Oxford: Clarendon Press.

Horn, Laurence R. 1969 "A presuppositional analysis of onIy and even", in: Robert I. Binnick--Alice Davison--Georgia M. Green--Jerry L. Morgan (eds.), 98-107.

Horn, Laurence R. 1989 A natural history of negation. Chicago: University of Chicago Press.

Johnson, Samuel 1756 A dictionary of the English language. (Edited by Alexander Chalmers). London: [No indication of publisher]. [1843] [1994] [Reprinted. London: Studio Editions].

Jorgensen, Erik 1974 "'Only' with a temporal value", English Studies 55: 247-60.

Konig, Ekkehard 1981 "The meaning of scalar particles in German", in: Hans-Jurgen Eikmeyer--Hannes Rieser (eds.), 107-32.

Nevalainen, Terttu 1991 But, only, just: Focusing adverbial change in Modern English 1500-1900. (Memoires de la Societe Neophilologique de Helsinki LI).

OED = The OxJbrd English Dictionary.

Poutsma, Hessel 1928 A grammar of Late Modern English. Vol. I: The sentence, 1. (2) Groningen: P. Noordhoff.

Rissanen, Matti 1985 "Expression of exclusiveness in Old English and the development of the adverb only", in: Roger Eaton--Olga Fischer--Willem Koopman--Frederike van der Leek (cds.), 253-67.

Ross, John Robert--William E. Cooper 1979 "Like syntax", in: William E. Cooper--Edward C. T. Walker (eds.) 343-418.

Shopen, Timothy (ed.) 1985 Language typology and syntactic description. Vol. III: Grammatical categories and the lexicon. Cambridge: Cambridge University Press.

Taglicht, Josef 1984 Message and emphasis: On focus and scope in English. London: Longman.

Vallins, G. H. 1956 The pattern of English. London: Andre Deutsch.

JOHN ANDERSON Methoni, Greece