Dictionaries and grammars uniformly classify the once and twice of such examples as I only met her once or twice as adverbs. Huddleston & Pullum et al. (Reference Huddleston and Pullum2002, henceforth CGEL) is no exception. However, a closer investigation of the distribution of these words leads to the proposal of a new analysis which reflects the relationship between them and the expressions one time and two times. We take this relationship to be comparable to that between somebody and some person, so that, at the phrasal level, once and twice in the above example are noun phrases (henceforth NPs).
In section 1 we outline and develop relevant aspects of the analysis of NP structure presented in CGEL. We begin with the analysis of fused relative constructions and conclude with the analysis of expressions like somebody as compound determinatives which function simultaneously as the determiner and head of NP. These analyses distinguish carefully between functions like ‘determiner of’ and categories like ‘determinative’, and permit, under restricted circumstances which we here explicitly define, the fusion of functions (henceforth FF).
In section 2 we then argue in detail for an analysis of once (in all but one of its uses) which invokes FF and in all essential respects parallels the analysis of somebody. Analogous analyses for twice and the somewhat archaic thrice follow straightforwardly in section 3. We then in section 4 discuss certain non-standard uses of once which lend further support to the FF analysis.
Finally, in section 5 we show that the constructions which we analyse using FF have properties which clearly demarcate them from superficially similar constructions such as incorporation, where a new grammatical word is formed by the incorporation of one category into another, and hybridization, where a head projects to a phrase of a distinct category. Hybrid constructions are found only marginally in English, but typically occur in event nominalizations in a variety of other languages. In such cases it can be argued that nominalizations (of category NP) are ultimately headed by a verb rather than a noun.
The new analysis therefore provides further empirical evidence for the existence of FF as a distinct syntactic mechanism, and hence lends support to descriptions, such as those found in CGEL, which postulate functions as independent constructs.
1. Fusion of functions
The basic analysis of NP structure we assume in this paper is that presented in Payne & Huddleston Reference Payne, Huddleston, Huddleston and Pullum2002 (chapter 5 of CGEL). The first point to note is the systematic distinction in CGEL between functions and categories. At the topmost level, the basic distinction employed with respect to functions is that between head and dependent. In canonical constructions, dependents can then be subdivided into the more specialized subtypes of complement (Comp), modifier (Mod) and determiner (Det), this latter function being restricted in its occurrence to NPs only.Footnote 2 A further relevant case of the head–dependent relation, employed in non-canonical constructions involving fronting, is that between nucleus and prenucleus. The prenucleus is the (dependent) function of a fronted constituent such as a relative phrase, and the nucleus is the (head) function of the clause containing the corresponding gap. In the case of fronted relative phrases, the prenucleus function corresponds approximately to specifier of CP in other frameworks, although ‘specifier’ is generally construed as a structural position rather than as a function in the sense intended here.
On the other hand, the main categories assumed for English are the lexical categories of noun (N), verb (V), adjective (Adj), adverb (Adv), preposition (Prep), determinative (D), and a corresponding set of phrase-level projections. While Adj, Adv, Prep and D project simply to AdjP, AdvP, PP and DP respectively, N projects via the intermediate-level phrasal node Nom (=nominal) to NP, and V projects via the intermediate-level VP to Clause. The category N will then, in canonical NPs, function as what we call the ultimate head of NP. The category Nom corresponds approximately to the N′ category in X-bar theory, but differs from this in that Nom can consist of a head noun and a modifier as well as of a head noun and a complement. Modifiers and complements in the CGEL framework are not defined in structural terms as dependents of different categories of head (e.g. complements as sisters of N and modifiers as sisters of N′). Nom therefore represents the first level of the phrasal expansion of N, whether this be by a complement or by a modifier.
An important instance of the distinction between function and category is the distinction between the function determiner and the word-category determinative.Footnote 3 Although the determiner is characteristically realized by a determinative such as the, a, this, some, any, one, two, etc., it can also be realized by NPs – usually genitive, as in this guy's shoes, but sometimes plain, as in what size shoes. Conversely, while all determinatives can function as determiner, most can also appear in other functions too. Thus the demonstrative determinative this is determiner in this size but modifier in AdjP structure in It's about this big. Similarly, the cardinal numeral determinatives are determiners in one time, two times, but modifiers in that one time, the two times I tried it myself, where they are part of the head nominal determined by that and the.
Structural diagrams in CGEL follow the convention of labelling each node with its function (upper line followed by a colon) and category (lower line). Thus the CGEL representation of the noun phrase the two times I tried it myself would be as follows (ignoring for simplicity the internal structure of the relative clause):
(1)
Here then NP consists of a head of category Nom and a determiner of category D (the). The noun times is the ultimate head, and there are two modifiers (two and I tried it myself). Note also that, as in CGEL, we conventionally simplify diagrams by representing lexical categories in basic dependent functions which can in principle project to higher phrasal levels, but in the given example fail to do so, simply by their lexical category label. The modifier two, for example, is conventionally labelled D rather than DP.
The grammatical framework used in CGEL also allows for the fusion of two functions (FF), i.e. for two functions which are realized canonically by discrete expressions to be realized jointly by a single expression in certain constructions. There are two cases: a simple case and a complex case. In the simple case, FF involves the fusion of a head function and an immediate dependent function (by which we mean a determiner, complement or modifier of that head or of its phrasal projections). This case is schematically represented with head-dependent order in (2a) and dependent-head order in (2b). The fused functions are indicated by writing a hyphen between them (e.g. ‘Head-Dep’):
(2)
The path above the fused node from Head to XP may contain any normally licensed intermediate heads, and we indicate this by representing this head projection as a bold dashed line. For example, in the case where XP=NP, a Nom head may intervene between the fused head and NP in (2a) and (2b), and similarly between the head N and NP in the counterpart constructions (2a′) and (2b′). As can be seen, at the fused node, the category which realizes the fused functions need not belong to the expected category X (and quite typically does not). Instead, it may be a lexical category Y or any licensed phrasal projection of Y. The ultimate head of XP will therefore be Y rather than X. The fused node can thus be thought of as a point in the normal path of headedness at which there can be a change of category (a property of FF which is shared with hybrid constructions, as shown in section 5 below). There will, however, still be an unbroken path of headedness from Y to XP, and in this sense FF constructions, like hybrid constructions, might be considered endocentric rather than exocentric.
It should be noted that we regard constructions with FF as constructions in their own right, licensed by independent rules governing well-formed structures. Nevertheless, for the majority of the constructions discussed in this paper, there exists a canonical counterpart construction without FF which is structurally similar in that (i) it is dominated by the same category XP, where XP is in principle any phrasal category including the intermediate category Nom, and (ii) the category of the fused node is the same as that of the dependent in the counterpart (see (2a′) and (2b′) above). In order to highlight these correspondences in the structural diagrams in this paper, fused functions and their category will be shown in bold face, as will be the corresponding functions and the category of the dependent in the counterpart construction.Footnote 4 It should be noted, however, that, although the FF and counterpart constructions may share these structural similarities, they frequently differ in other respects, e.g. distributionally or semantically, and in some cases no counterpart even exists. We do not therefore wish to construe FF constructions as related to their counterparts in derivational terms.
In the complex case, FF involves not a head and its immediate dependent, but rather a head and an immediate dependent of an immediate dependent (IDOID). This case is represented schematically in (3a) and (3b), with the corresponding counterpart constructions in (3a′) and (3b′):
(3)
It is particularly obvious from the complex case that the structural diagrams required for FF represent in one sense a violation of what Sampson (Reference Sampson1975) calls the single mother condition. Although each fused function in principle has its own mother (in the simple case the two mothers happen to coincide), the single category which realizes them will in effect have two mothers. This is the major departure from canonical principles of structure which FF entails. The second departure is of course the possibility noted above that a phrasal constituent of category XP may have a category other than X as its ultimate head. As already noted, this second departure will be independently required in the case of hybrid constructions. Otherwise, standard structural conditions obtain. In particular, there is no weakening of any prohibition against crossing branches, and this will have the useful consequence that fused functions are automatically adjacent.Footnote 5
In sum, therefore, we associate the following general properties with FF:
(i) In FF, a single lexical or phrasal category Y(P) simultaneously realizes two functions.
(ii) FF is permitted in a category XP only between the head of XP and either an immediate dependent of XP or an immediate dependent of the immediate dependent of XP.
(iii) The fused functions are adjacent.
(iv) The category Y(P) is typically not identical to XP.
Comparison with the counterpart constructions suggests two further properties:
(v) The category XP is a projection of the category of the ultimate head in any counterpart construction.
(vi) The category Y(P) is the category of the dependent in any counterpart construction.
In practice, we expect FF to be restricted in its application within a given language to a small set of constructions. In English, for example, FF appears to be largely restricted to the internal structure of NPs, with XP being either NP itself or Nom (the fused relative construction can, however, also involve other categories, see section 1.1 below). The subtypes of dependent involved are determiner, modifier and, in fused relative clauses, subject or prenucleus. However, we envisage the likelihood that FF constructions might involve further categories and further subtypes of dependent in other languages.Footnote 6
1.1 Fused relatives
The most obvious example of an FF construction in English is found, we claim, in what we call the fused relative construction (in other frameworks this is commonly known as the ‘free’, ‘headless’ or, where the category involved is NP, ‘nominal’ relative construction). A standard NP example is given in (4):Footnote 7
(4) I'll only eat whatever meals you can get without additives. (NP)
Here the underlined expression in (4) is an NP in which we propose that the relativized NP whatever meals functions as prenucleus of the relative clause modifier, but also simultaneously as head of the larger NP.Footnote 8
In the case of the NP fused relative construction in (4) there is a counterpart non-fused construction as in (5):
(5) I'll only eat meals which you can get without additives.
Diagrammatically, CGEL represents the non-fused relative in meals which you can get without additives as in (6a), and the fused relative whatever meals you can get without additives as in (6b):
(6)
In the non-fused relative in (6a) the NP is headed by a Nom constituent which itself contains the ultimate head noun meals and the modifying relative clause. This relative clause in turn consists of the prenucleus (the relativized NP consisting of the relative pronoun which), and the nucleus (a clause containing a gap in object position). The parallel with the fused relative in (6b) is clear and systematic: in the fused relative the prenucleus function fuses with the head function, and this fused combination of functions is realized by a single constituent (the NP whatever meals).
The analysis in (6b) illustrates all of the distinctive properties we attribute to FF constructions:
(i) In FF, a single lexical or phrasal category Y(P) simultaneously realizes two functions.
A single category, in this case the phrasal category NP, simultaneously realizes two functions: head of Nom (therefore head, but in this case not ultimate head, of the whole NP) and prenucleus in the relative clause.
(ii) FF is permitted in a category XP only between the head of XP and either an immediate dependent of XP or an immediate dependent of the immediate dependent of XP.
The prenucleus–nucleus relation is a special case of the dependent–head relation, and so we have here fusion between the head of Nom and an immediate dependent (the prenucleus) within an immediate dependent (the relative clause modifier of the head). This is then the complex case of FF schematically shown in (3a).
(iii) The fused functions are adjacent.
The prenucleus in the fused relative construction must be adjacent to the head to which it is fused. Predictably, no modifiers can then intervene between the head and the relative clause. Note that this is not an immediate consequence of analyses in which the head is empty and the relative phrase realizes (in our terms) solely the prenucleus function.
(iv) The category Y(P) is typically not identical to XP.
The category NP of whatever meals is not identical to the mother category Nom, although of course NP and Nom are both legitimate projections of N. Fused relatives in English have the matching requirement that YP and XP are indeed both legitimate projections of the same lexical category. However, this is not a universal property of fused relatives, and not something which is required by the definition of FF.Footnote 9
Furthermore, if we compare the fused relative in (6b) with the counterpart standard relative in (6a), there is conformity to the correspondence rules between FF and counterpart constructions:
(v) The category XP is a projection of the category of the ultimate head in any counterpart construction.
The category XP is Nom, a projection of the category noun which functions as the ultimate head of the standard relative construction.
(vi) The category Y(P) is the category of the dependent in any counterpart construction.
Taking which as a relative pronoun heading an NP, the category of the fused constituent (NP) is that of the dependent (the prenucleus) in the standard relative construction.
The FF analysis of fused relatives extends naturally to cases in which the fused constituent belongs to a category other than NP:
(7)
Following the category-matching constraint of fused relatives in English, the category of the underlined expression in (7a) will be a PP in which the PP in whichever city functions both as prenucleus of the relative clause and as head of the larger PP. Analogously, the underlined expression in (7b) will be an AdjP containing an AdjP (however responsive) as fused head-prenucleus, and the underlined expression in (7c) will be an AdvP containing an AdvP (however correctly) as fused head-prenucleus. The diagrammatic representation of the PP fused relative in (7a) will be (8):
(8)
The properties of PP, AP and AdvP fused relatives are clearly similar to those of NP fused relatives, though it should be noted that no counterpart construction exists in this case: standard relative clauses are modifiers only within Nom.Footnote 10
An immediate advantage of the FF analysis follows from the automatic adjacency of the fused functions. Under an alternative approach in which the head and prenucleus are separate functions (or in other frameworks structural positions), there is no obvious reason why it should not be possible to insert any other post-head dependent between the head and the prenucleus:
- (9)
(a) [Head people] [ClauseREL who [you despise __ ]]
(b) [Head Ø] [ClauseREL whoever you despise __ ]
(c) [Head people] [Mod present] [ClauseREL who [you despise __ ]]
(d) *[Head Ø] [Mod present] [ClauseREL whoever you despise __ ]
Consider then a potential non-fused structure such as (9b) in which the head is null and the relative phrase is solely the prenucleus of the relative clause.Footnote 11 Some additional stipulation would seem to be required to prevent the modifier present from intervening between this null head and the relative clause, as in the ungrammatical (9d).
A second useful property of the FF analysis is that it provides a natural account of the formal differences between ordinary and fused relatives. As is well-known, the series of wh-words which can be used in fused relatives differs considerably from those used in ordinary relatives; in particular, the non-personal form is what rather than which, and the personal form who is highly restricted in its use. What is more, the fused relative construction permits forms in -ever whereas ordinary relatives do not. For a full account of such differences see GGEL, pp. 1074f. These differences follow immediately from the notion that formal realizations of the same basic category can vary depending on the function it performs. The NPs which realize the fused head-prenucleus (or head-subject) function can differ formally from those which realize just the prenucleus (or subject) function. It is, however, not immediately clear why these formal differences should exist in an analysis which treats fused relatives as ordinary relatives modifying a null head.
The most striking advantage of the FF analysis, however, is that it straightforwardly accounts for the observation that optimal examples with the case-marked relative pronouns who or whoever are those in which the case of the pronoun (nominative or accusative) is proper to both the function of the fused relative itself and the function of the relativized NP. This is a special case of the potential for formal distinctions between categories realizing fused and non-fused functions. Consider the following examples (see CGEL: 1074):
- (10)
(a) [Whoever is responsible for the damage] must pay for it.
(b) He will criticize [whomever she brings home].
(c) ?[Whomever he marries] will have to be very tolerant.
(d) ?She lunches with [whomever is going her way].
In (10a), the fused relative and the relativized NP are both subjects, and the nominative case of whoever satisfies the requirements of both functions. Similarly in (10b) the fused relative and the relativized NP are both objects and the accusative form whomever is fully acceptable in those varieties of English which permit accusative case in object relatives.Footnote 12 Where there is a conflict of function, however, as in (10c) and (10d), the accusative form whomever is marginal.
Quite simply, an NP like whomever in (10b) will be assigned two cases by pre-existing mechanisms in the grammar. One case (let us call this the ‘internal’ case because it originates properly from within the relative clause) is assigned by virtue of the object function of the relative clause gap. This case will be inherited by the prenucleus using exactly the same mechanism as is required for the case of the prenucleus in a standard relative clause. The second case (let us call this the ‘external case’) is derived by standard mother-head percolation mechanisms from the top NP, and reflects the object function of this NP. We will represent this double case assignment by a hyphen, viz. ‘external case-internal case’. In (10b), then, the cases assigned are acc-acc, as illustrated in (11):
(11)
The realization of ‘external case - internal case’ is then straightforwardly a matter for morphological resolution rules. In English, we essentially have a matching requirement: nom-nom is realized as nominative case, and acc-acc as accusative case. When the rules which assign the cases conflict, we have an instance of what we might call a ‘syntactic quandary’. For nom-acc or acc-nom there is no fully acceptable resolution. However, many languages do allow case conflicts to be resolved (for an overview see in particular Vogel Reference Vogel and Alexander2001, Reference Vogel, Gisbert and Caroline2002). One strategy is that resolution is always possible and always in favour of the external case, as in Icelandic. Another is that resolution is always possible but depends on a hierarchy of obliqueness: nom<acc<obl.Footnote 13 In Gothic, for example, resolution is always in favour of the more oblique case, whether this be external or internal. It is also possible, as in German, that resolution is sometimes possible and sometimes not, more oblique cases being generally harder to suppress. What is most important is that in principle either the external case or the internal case can ‘win out’. There is therefore no reason on grounds of inflectional form to assume that either the head function or the prenucleus function must predominate in these constructions.
We regard it as a striking advantage of the FF approach to these constructions that no special mechanisms are required beyond morphological resolution rules to account for the matching effect in languages like English and for case conflict resolution in those languages where this is a possibility. Analyses without FF have to assume either that the relativized phrase functions as the head of the whole NP and there is no fronted element in the relative clause (this is the analysis of Bresnan & Grimshaw 1978, now largely rejected), or that the head of the whole NP is null and the relativized phrase occupies, in our terms, prenucleus position at the front of the relative clause (Groos & van Riemsdijk Reference Groos, van Riemsdijk, Adriana, Brandi and Rizzi1981 and most subsequent work). One of the main motivations of the original Bresnan & Grimshaw approach was the matching effect, since the relativized phrase as head would automatically bear the case of the whole NP. However, it is not clear how this analysis could plausibly handle case resolution where the relativized phrase bears the internal case.Footnote 14 On the other hand, the Groos and von Riemsdijk approach requires the postulation of an ad hoc ‘Comp Accessibility Parameter’, which allows the matrix case to percolate from the top NP to the prenucleus position inside the relative clause.Footnote 15
The FF analysis of fused relative constructions thus has the significant advantage of not requiring additional ad hoc stipulations to account for either the adjacency of the head and relative clause, or the observed formal differences between fused and ordinary relatives. In particular, it provides a straightforward account of case matching and case resolution phenomena. For other properties of the analysis see CGEL (1068–1079).
1.2 Fusion of determiner or modifier within NP
The concept of fusion is also invoked to handle examples like two of the apples, where two is fused determiner-head, and the biggest of the apples or the very hard of hearing, where biggest and very hard of hearing are fused modifier-head and head-modifier respectively (CGEL: 410–422).Footnote 16
Compare the proposed analysis of the very hard of hearing (12b) and the analysis of non-fused the people very hard of hearing (12a):
(12)
In (12a), the adjective phrase very hard of hearing is a post-head modifier of the head noun people. In (12b) the adjective phrase modifier is fused with the head. The adjectival status of the fused constituent is clearly demonstrated by the presence of the adverb modifier very. Note that in this case the fused and non-fused constructions are not semantically equivalent. Examples like the people very hard of hearing cannot be interpreted generically. This is an a priori argument for thinking that the fused construction is not related to its counterpart by simple omission of the head.
This FF construction conforms to the principles we propose:
(i) In FF, a single lexical or phrasal category Y(P) simultaneously realizes two functions.
A single category, in this case the phrasal category AdjP, simultaneously realizes two functions: head of Nom (therefore head, but not ultimate head, of the whole NP) and modifier of that head.
(ii) FF is permitted in a category XP only between the head of XP and either an immediate dependent of XP or an immediate dependent of the immediate dependent of XP.
The modifier relation is a special case of the dependent–head relation, and we have here fusion between the head of Nom and an immediate dependent of Nom. This is the simple case of FF schematically shown in (2a).
(iii) The fused functions are adjacent.
The modifier must be adjacent to the head to which it is fused. Predictably, then, no further modifiers can intervene between the head and the fused modifier. Note that this is not an immediate consequence of analyses in which the head is null and the adjective phrase realizes (in our terms) solely the modifier function.Footnote 17
(iv) The category Y(P) is typically not identical to XP.
The category AdjP of very hard of hearing is not identical to the mother category Nom.
Furthermore, if we compare the fused construction in (12b) with its counterpart construction in (12a), there is conformity to the correspondence rules between FF and counterpart constructions:
(v) The category XP is a projection of the category of the ultimate head in any counterpart construction.
The category XP is Nom, a projection of the category noun which functions as the ultimate head of the non-fused construction.
(vi) The category Y(P) is the category of the dependent in any counterpart construction.
The category of the fused constituent (AdjP) is that of the modifier in the non-fused construction headed by people.
A detailed rationale for the FF analysis of such examples is given in CGEL (410–422), and we will not repeat it here. However, it can be noted that the potentiality for formal differences between categories realizing fused and non-fused functions is observed here just as it is in the fused relative construction. For example, the determinative every cannot occur as fused determiner-head (*every of the students), and the determinative no has a specialized morphological variant none which is restricted to the fused function. The transparent relation between form and function in our approach, together with the independently motivated lexical distinction between determinatives that allow FF and those that do not, provides for a very natural description of such facts.
1.3 Compound determinatives
The special case of fusion most relevant to the analysis of once is that illustrated by somebody, everything, anyone, etc.: these are compound determinatives functioning as fused determiner-head (CGEL: 423f.). The primary motivation for analysing these compounds in this way (rather than as pronouns, as in traditional grammar) is that they take exactly the same pre-head modifiers as the determinative bases they contain: compare not everything and not every house, hardly anyone present and hardly any writer present, and so on. By contrast, the post-head modifiers are those, like the adjective present, which are appropriate as dependents within NP.
The CGEL representations of any writer, anyone, hardly any writer present and hardly anyone present are given in (13):Footnote 18
(13)
In (13a) we have an NP consisting of the determinative any functioning as determiner and the head noun writer. In (13b) we see the parallel analysis of the compound determinative anyone: here the functions of determiner and head are fused and the exponent of the fused functions is the compound determinative anyone. Example (13c) shows the determinative any premodified by the adverb hardly; together these form a DP (Determinative Phrase) which functions as determiner (see footnote 2). The most complex example is (13d), in which anyone functions as the head of the DP, allowing modification by hardly, but also simultaneously as the head of the whole NP in which there is an adjectival modifier. This adjectival modifier fulfils a specialized modifier function, termed the ‘restrictor’ function; this function is non-recursive and can only be realized by constituents in post-head position, but in addition to normally permitted post-head adjectives and adjective phrases, it also allows simple adjectives which would normally occur pre-nominally. It is a particular strength of the FF analysis, we argue, that it provides an explanation for the janus-like behaviour of forms such as anyone which have adverbs as premodifiers but adjectives and adjective phrases as postmodifiers. Also, the assignment of determinative category status to these forms follows from the nature of the pre-head and post-head modifiers. The pre-head Adv is an appropriate modifier within the DP which fulfils the fused Det-Head function, and this DP must have a D as head.
Null-headed analogues of the CGEL analysis have subsequently appeared, in particular Larson & Marušič (Reference Larson and Marušič2004) and following them Leu (Reference Leu2004), and these can be directly compared.Footnote 19Larson & Marušič (Reference Larson and Marušič2004) provide an interesting and novel set of arguments that adjectival modifiers which follow (in our terms) compound determinatives pattern with post-nominal adjectival modifiers generally, and not with pre-nominal ones, in the sense that specifically pre-nominal forms or pre-nominal interpretations of general adjectival forms are excluded.Footnote 20 For example, Sadler & Arnold (Reference Sadler and Arnold1994) note that measure adjectives in pre-nominal position are inflectionless and more word-like (a 23-inch-long rope) while in post-nominal position the modifier is more phrase-like and the measure itself is marked plural (a rope 23 inches long). With compound determinatives, the post-nominal position and form are required (anything 23 inches long/*23-inch-long anything/*anything 23-inch-long). Observations such as these are of course consistent with CGEL's claims that the modifier in the compound determinative construction is essentially post-head, and that certain adjectives are restricted to pre-head positions.
Larson & Marušič (Reference Larson and Marušič2004) very briefly sketch two null-headed analyses, one of which is a direct null-headed analogue of the analysis in CGEL.Footnote 21 In it, forms such as anyone are morphologically complex Ds in the following configuration:
(14) [DP D (*AdjP) [NP Ø ] AdjP]
As Larson & Marušič themselves point out, a significant problem with this account is that it requires a stipulation that pre-head AdjPs are not permitted when the head is null. This is of course precisely the kind of stipulation which is not required when the determiner and head functions are fused. The impossibility of an intervening pre-head AP follows automatically from the adjacency requirement in the FF construction, and this itself ultimately follows from the more fundamental constraint against crossing branches.
2. Once
We distinguish five uses of once. The examples in (15) illustrate:Footnote 22
(15)
Numerical once indicates ‘one occasion’, and is directly equivalent to the analytic expression one time with an overt occurrence of the determinative one.Footnote 23 Just as the numerical use of determinative one invites a contrast with other cardinal numerals, so numerical once contrasts with twice and archaic thrice, indicating ‘two occasions’ and ‘three occasions’ respectively.Footnote 24
The uses of once in (15b) and (15c) both correspond to what are called ‘singulative’ uses of the determinative one (CGEL: 386f.), where the contrast with higher cardinals is backgrounded. In (15b) we have the case where once is modified by not; this not can similarly modify one, but is for the most part not compatible with the higher cardinals. Equivalently, we do not have here a contrast between once and twice. Past once can be thought of as a specialized singulative use with the meaning ‘at some time in the past’, where the ‘one’ component is similar in meaning to some and again has no numerical import. An obvious parallel is with the analytic expression one day, which employs a similar singulative use of one but has the special meaning ‘at some time in the past/future’.Footnote 25
The term-of-office use of once in (15d) appears related to the past use, but in fact is restricted in its application to terms of office, with a meaning something like ‘for one (past) term’. Typical uses are therefore as attributes to names of offices (mayor, president, prime minister, governor, etc.). We could not naturally say *my once job as a glass-blower, where the adjectives former or onetime seem more appropriate. This term-of-office use is for many speakers singulative, so that the numerical import is absent and once indeed has a meaning very close to onetime. However, there are also many speakers who allow the use of higher forms like twice in a similar function with the meaning ‘for two (past) terms’.
Finally, sequential once in (15e) relates two events sequentially in time, and is similar in meaning to after. Any relationship with the determinative one in this case seems absent.
2.1 Numerical once
From a syntactic point of view, numerical once has a host of distributional characteristics which are inconsistent with an adverb analysis. They are illustrated in (16):Footnote 26
- (16)
(a) I admit that I did try jogging once, but only the once.
(b) I am determined to choose, just this last once, my own destiny.
(c) It only happens every once in a blue moon.
(d) Every once in a while we come up with an offer that is too good to miss.
(e) Pam took him [a guidedog] over to the vet when we got back to campus; the recommendation was to try split feeding, that is, twice a day instead of the once that we're doing now.
(f) If you study the pulleys you will see that the little one is going round four times for every once that the large pulley goes round.
(g) Once is enough.
(h) I can foresee at least once when I won't have to do it.
Example (16a) illustrates a common use of numerical once in which it is preceded by the definite article. Typically this use also involves a preceding peripheral focussing adverb such as only, just or even. Example (16b) illustrates that instead of the definite article we can have a demonstrative, and also the adjectival modifier last. The expression once in a blue moon in (16c) is an idiom meaning ‘very rarely’, but clearly includes an instance of numerical once as shown by frequently attested examples of the derivative expression twice in a blue moon, as in (17):
(17) Please EMAIL ME and indicate any of the current software they [use/have used/may use] [regularily[sic]/sometimes/once or twice in a blue moon].
Most significantly, however, example (16c) shows that numerical once can be determined by the determinative every. The idiom once in a while, ‘sometimes, but not often’, in (16d) is similar in allowing every as a determiner.Footnote 27 Here too we seem, on the basis of meaning, to have an occurrence of numerical once; the derivative twice in a while is again attested, though less frequently than twice in a blue moon. In (16e) and (16f) once is modified by a post-head relative clause, with the or every again as determiner;Footnote 28 the numerical meaning is evident from the overt contrast with twice and four times. And finally, numerical once can, in an admittedly limited way, function in major argument positions, as in (16g), where it is subject, or (16h), where it is object.
A fair number of dictionaries and grammars note some of these constructions and analyse once as a noun when it appears in these contexts. There are, however, two objections to this way of handling the data.
In the first place, it is an unnecessary complication to assign once to different categories in such examples as (18a), ‘adverb’, and (18b), ‘noun’:
- (18)
(a) I only met her once.
(b) I only met her that once.
At the phrasal level, that once in (18b) is uncontroversially an NP, and there is no reason to exclude an NP analysis of once in (18a). To be sure, the function of once in (18a) is that of adjunct, but there is no general exclusion of NPs from adjunct function; a limited range of NPs can appear as adjuncts of various types, as in She arrived last week, Do it this way, I found it just the other side of the path, and so on. In the specific case of a frequency adjunct we have that once but also four times and other expressions headed by times. There is a tacit but quite unjustified assumption in traditional grammar that nouns can appear in adjunct function only when accompanied by an explicit dependent. Thus while Tuesday is a noun in I met her last Tuesday, where it has last as dependent, the yesterday of I met her yesterday is analysed as an adverb – though both are taken as nouns in We spent last Tuesday/yesterday in Rome. But we don't need a dual-category analysis of yesterday: it is a noun in both cases, the difference being simply a matter of function, adjunct vs. object (see CGEL: 564f.). We likewise don't need a word-category distinction to handle the difference between the once of (18a) and that of (18b) – or (16g).
A second objection is that while the underlined expressions in (16) and (18) are NPs, it does not follow that once has to be analysed here as a noun. A more explanatory account of the data can be given if we include numerical once in the category of compound determinatives that appear in the fused-head construction. The compound determinatives given in CGEL – somebody, someone, something, somewhere, etc. – consist of a determinative base (some‧, every‧, any‧ or no‧) and a nominal base (‧body, ‧one, ‧thing and ‧where/place).Footnote 29 The morphological compounding in this case reflects the fusion of two syntactic functions within NP structure, that of determiner and that of head. Numerical once and likewise twice and archaic thrice can similarly be thought of as compounds of a determinative base (numerical on‧, twi‧, thri‧) and a noun base ‧ce (meaning ‘time’).Footnote 30
An immediate advantage of this conception is that we obtain an explanation for why numerical once can occur both with and without a determiner, a property which is not that of a normal singular count noun. While the determinatives some, every, any and no can function solely as determiners and therefore never permit a preceding determiner, the numerical one can function both as determiner, as in one time, on‧ce, and as a modifier with a preceding determiner, as in examples like this one time, this on‧ce). The structures we propose are thus as in (19):
(19)
Further properties of numerical once follow straightforwardly. Since once in (19b) represents a fusion of determiner and head, there is no possibility of an attributive adjective preceding the determiner: *last once is excluded for the same reason as *last one time. On the other hand, it is perfectly possible for the attributive adjective to follow the determiner in (19d) and (19c), giving this last once and this last one time. These properties as before follow automatically from the adjacency requirement on FF. Note in particular that a potential analysis without FF in which once is D and is followed by a null head will not, without stipulation, be able to block examples in which an adjective intervenes, for example *(this) once last Ø.
As a determinative, numerical once permits the same pre-head modifiers as the determinative one:
- (20)
(a) You should really see this film at least once.
(b) You should really see this film at least one time.
The analysis of (20a) is analogous to that of (13d): once is the head of a DP in which it is modified by at least, and this DP functions as the fused determiner-head of the whole NP.
Also, the determinative one can occur embedded within semantically numerical DPs such as more than one and PPs such as up to one. In these cases the numeral is not the head of the DP or PP (see CGEL: 432f.); the respective heads are the determinative more and the preposition up. Rather, one functions as the complement of the prepositions than or to. As can be seen in (21), the behaviour of once is again exactly parallel.
- (21)
(a) I have been to Archangel more than once.
(b) I have been to Archangel more than one time.
The analysis of (21a) has to be slightly more complex than that of (20a), but is essentially the same. The function of once is complement of than within a DP which functions as fused determiner-head of the whole NP. Representations of the underlined NPs in (20a) and (21a) are given in (22):Footnote 31
(22)
Predictably, DPs of this more complex type can also function, preceded by a determiner, as fused modifier-heads. An attested example is (23):
(23) Similarly, 1 Timothy 3 insists that a bishop must have been married only one time, not the more than once which would have been permitted upon the death of a wife.
The analysis of (23) follows just as straightforwardly from the general framework, and is given in (24):
(24)
In (24), the DP more than once functions simultaneously as the modifier within the higher nominal, and the head of the nominal which contains the relative clause.
A final parallel between numerical once and in particular the existential compound determinatives (something, anything, etc.) is the ability to take more as a special post-head modifier, as in once more, something more, anything more.
2.2 Singulative once
The most productive use of singulative once is following the negator not, just like singulative one.Footnote 32 However, we can also identify two further uses as belonging to this type: the once which occurs following conditional if, and the once which occurs in the fixed expressions for once and once and for all.Footnote 33 Examples with not and if are given in (25) and (26):
- (25)
(a) Not once will you hear anything like this again.
(a′) *Not twice will you hear anything like this again.
(b) Not one answer satisfied him.
(b′) *Not two answers satisfied him.
(26) If you once allow the dog onto furniture or into the bedroom, it will be difficult to reverse such action if you subsequently decide that it was not a good decision.
The negator not is not generally found with higher cardinals, and this holds in (25a′) just as it does in (25b′). The form once in (25a) therefore stands as the singulative counterpart of numerical once, just as singulative one stands in relation to numerical one (CGEL: 386). Likewise, the use of once following if in (26) parallels the use of one in the phrase at one time or another; this one is also singulative, as can be seen from the ungrammaticality of *at two times or another. Just as there is no reason to think of singulative one and numerical one as belonging to distinct categories (they are both manifestations of determinative one), so it seems reasonable to consider singulative once as belonging to the same compound determinative category as numerical once.
One apparent difficulty with this conclusion which must be addressed is the fact that singulative once nevertheless lacks many of the distributional characteristics of numerical once. In particular, singulative once is incompatible with determiners and attributive adjectives. However, it is also a characteristic of singulative one that it always functions as a determiner, not as a modifier, and this will naturally account for these distributional properties. A further determiner is inadmissible because NP structure does not permit more than one determiner, and an attributive adjective is impossible because here the determiner function is fused with the head function. It should be noted however that post-head relative clauses are indeed still possible, including ones with when and while as relative words:
- (27)
(a) Not once that I can think of have I ever seen a mess like this!
(b) Not once when I reported my uncle did I use the word ‘rape’.
(c) Not once while I have used this software has it caused a computer to crash.
2.3 Past once
Past once occurs typically as a clause-level adjunct, as in (28a), and can also be identified as the once which occurs in the phrase once upon a time. It can take post-head relative clauses with when and while, as in (28b) and (28c):
- (28)
(a) I was once a teenage werewolf.
(b) Once when the opera Aida was about to be performed a singer was missing.
(c) Once while I was in Australia I went to see the penguins on Phillip Island.
In all environments, past once has the meaning ‘at some time in the past’. The sense of ‘one’ that occurs in this use is the same as occurs in one sense of the expression at one time, when this is interpreted as ‘at some time in the past’, or in the expression one day, meaning ‘at some time in the past/future’. Both of these involve singulative uses of one. Past once would seem therefore to be best considered as a further, somewhat specialized, use of the compound determinative with a singulative determinative component. The impossibility of further determiners and attributive adjectives follows for the same reasons as with singulative one in section 2.2.
2.4 Term-of-office once
Term-of-office once occurs solely as an attributive modifier referring to a term of office in the past. It does contrast with higher forms like twice in this sense, but such higher forms are rare, so that the salient interpretation is singulative, somewhat akin to that of onetime or former in the same function. Two attested examples are given in (29):
- (29)
(a) The once mayor of New York David Dinkins spoke of ‘the gorgeous mosaic’ that has presumably replaced the earlier image of the American melting pot.
(b) We were lucky enough to have Grant (a local) bring his car out and run us around. He took us up to Monticello, once home of the once president.Footnote 34
The boundaries of this use of once are murky, but the overwhelming majority of attestations occur as attributes to names of offices such as mayor, governor, senator, prime minister, president, captain, etc.Footnote 35 Occasional uses are found with terms like partner, wife, and husband, e.g. ?my once partner, though these seem somewhat less natural. The contrast with past once can be seen in (30a) and (30b):
- (30)
(a) *I am a once gardener at Kew.
(b) I was once a gardener at Kew.
From a syntactic point of view, term-of-office once might simply be treated as an adjective. However, there would seem to be no compelling reason to proliferate the number of categories attributable to once in this way (especially as a similar move would have to be made for twice and thrice in the same function). Term-of-office once can again plausibly be identified as a specialized use of the compound determinative, though it should be noted that, as an attribute, it will act as a fused modifier-head forming a nominal rather than forming a full NP.
2.5 Sequential once
Unlike all the previous uses of once, sequential once does not transparently contain any sense of ‘one’ as part of its interpretation:
- (31)
(a) Once you have finished, report to me!
(b) At the time you have finished, report to me!
Once here conveys only that the event of reporting should directly follow the event of finishing; an analytic paraphrase involves the expression at the time, rather than at one time.Footnote 36 The closest parallel in this case is with temporal after, as in After you have finished, report to me!: both take complements having the form of a declarative clause that excludes the subordinator that. We argue therefore that sequential once should be handled in the same way as this after.
Dictionaries and traditional grammars likewise assign this once to a different word-category than those covered in earlier sections. They analyse it – again like after in the use just illustrated – as a subordinating conjunction. We take the view, however, essentially following Jespersen (Reference Jespersen1924: 89) and numerous works in formal grammar such as Emonds (Reference Emonds1976), that most of traditional grammar's subordinating conjunctions should be analysed as prepositions, thus allowing prepositions, like verbs, nouns and adjectives, to take a range of different categories of complement, including finite subordinate clauses (see CGEL: 1011–1014; Huddleston & Pullum Reference Huddleston and Pullum2005: 128–130). After will thus be a preposition not only in after dinner but also in after you have finished, and this is the category to which we assign sequential once.Footnote 37
3. Twice (and thrice)
We distinguish three uses of twice and archaic thrice. Two of these correspond to uses of once: the numerical use in (32a) and the term-of-office use in (32b). Counterparts to the singulative, past and sequential uses of once are self-evidently lacking. However, we can also identify a further multiplier use for twice (and thrice) which is absent for once. This is illustrated in (32c):
(32)
3.1 Numerical twice (and thrice)
Numerical twice has exactly the same distributional characteristics that were described above to motivate the compound determinative analysis of numerical once. Attested examples include the occurrence with a determiner in (33a), an attributive adjective in (33b), a post-head relative clause in (33c), and pre-head modifiers such as at least in (33d), and the ability to function as subject in (33e):
- (33)
(a) I want to confess: I drank Jeffrey Archer's champagne. In mitigation, it was just the twice, and many years ago, at the parties he used to throw in his hotel suite at Tory conferences. I never touched so much as a forkful of his shepherd's pie.
(b) BBC2 is repeating the war walks series on a Tuesday so keep an eye out for the Bosworth episode, I've missed it the last twice.
(c) I have given way generously and I do not propose to do so again, unless it is to the Minister, to whom I am prepared to give way more than the twice that I promised.
(d) I stuck my finger into an outlet at least twice that I can remember.
(e) Violations of privacy through history prove twice is enough.
The analysis of numerical twice therefore exactly parallels that of numerical once. When a determiner is absent it is a compound determinative functioning as a fused determiner-head within NP structure; when a determiner is present, it is a compound determinative functioning as a fused modifier-head within a nominal.
3.2 Term-of-office twice (and thrice)
Attested examples of twice or thrice as an attribute referring to the number of times a particular office has been held are not as common as simple examples of term-of-office once, and many speakers find them unnatural. However, this usage is illustrated for twice from a contemporary source in (34a), and in (34b) for thrice in a source dating to the mid-19th-century:
- (34)
(a) Grover Cleveland was the twice president of the United States (1885–1889, 1893–1897), and spent his retirement years in Princeton.
(b) Between this spot and Highgate Hill was the Whittington stone, now replaced by another, where the thrice Lord Mayor of London heard the sound of Bow bells.
As with term-of-office once, this use can be analysed as a compound determinative in fused modifier-head function, forming a nominal functioning as attributive modifier.
3.3 Multiplier twice (and thrice)
Multiplier twice is seen in examples such as those in (35):
- (35)
(a) They won twice the amount we did.
(b) They were travelling twice as fast as we were.
(c) Twice three is six.
Evidently twice here corresponds to and can be replaced by the analytic expression two times, where times has a multiplicative rather than an ‘occasion’ sense. In (35a) twice is functioning as a predeterminer modifier in NP structure, preceding a definite determiner. In (35b) it is functioning in the structure of an AdvP, modifying a comparison of equality.Footnote 38 Example (35c) illustrates its use in multiplication tables. The absence of once in these constructions follows straightforwardly from the fact that once corresponds to ‘one time’ rather than ‘one times’. In the multiplication table example we would have One times three is three as opposed to *Once three is three.Footnote 39
Some modern grammars and dictionaries analyse the twice of (35a) as a predeterminer – see, for example, Quirk, Greenbaum, Leech & Svartvik (1985: 257), Sinclair (1987: 1575), Proctor (1995: 1572), Biber, Johansson, Leech, Conrad & Finegan (1999: 258). In these works, however, ‘predeterminer’ is the name of a category (or subcategory within the determiner category), rather than of a function, as in our own usage. For these works twice belongs to a different category in (35a) than (35b), where it is taken to be an adverb, but this is an unnecessary complication. The difference is simply one of function: there is no more reason to make a category distinction with the single word twice than there is with the two-word expression two times, which can replace it in both functions (predeterminer modifier and modifier within AdvP), and is uncontroversially an NP in both cases.Footnote 40
4. Non-standard uses
In this section we illustrate uses of once which we would consider non-standard (and mark as such with the annotation ‘!’), but which are consistent with the compound determinative analysis.
Consider first the attested examples in (36):
- (36)
(a) !I'm talking about a speeding ticket here. Since almost everyone has been stopped by a cop once or another, that's a good place to start.
(b) !You have received this email because at once or another you entered the weekly draw at one of our portals or FFA sites.
In these examples, we have a past use of once which would more standardly be replaced by the analytic expression one time. The speakers here are allowing the internal ‘one’ element of once to be accessed by or another, which is a natural accompaniment of singulative one. There is no reason why this construction should not exist; compound determinatives allow the determinative base some‧ to be accessed in just this way (somebody or another, something or another, somewhere or another).Footnote 41 Particularly significant is (36b), which allows once to be preceded by the preposition at and would be totally incompatible with an adverb analysis.
Secondly, in (37) we have an interesting blend construction:
- (37)
(a) !This is the once and only time you will use the User Name and password supplied by your instructor.
Here once would naturally be regarded as the first element of a coordinate attribute with only as the second element; however, *the once time is ungrammatical. The construction can therefore only have the grammatical structure the [[once] [and [only time]]], where once, a nominal with the form of a compound determinative, is coordinated with the nominal only time. However, the influence of the construction the one and only time is clearly visible here.
Finally, as a non-standard American rather than a British usage, we note that it is possible (at least for some speakers) for multiplier twice as predeterminer to be itself preceded by the definite article. Two examples are given in (38):
- (38)
(a) !As some piano strings are rigid and thick, the string dividing points at which no vibration occurs have a certain length. If vibrations occur between a string dividing point and string end, vibration will occur over a length little [sic] shorter than the half of the original string length. This causes the second harmonic's frequency to be slightly higher than the twice that of the fundamental.
(b) !The package offers the twice the amount of web space as our ‘Standard’ hosting package for businesses, and more than twice the number of e-mails, all at half the price.
This usage is obviously consistent with an analysis of multiplier twice as a compound determinative; here twice functions as fused modifier-head in an NP with the as determiner. The construction is totally incompatible with the traditional analysis of multiplier twice as an adverb.Footnote 42
5. Distinctive properties of FF
In section 1, we demonstrated the distinctive properties of FF and its analytical advantages in a variety of English constructions, concluding with the analysis of compound determinatives of the somebody type. In sections 2–4, we argued in detail that once (in all but its sequential use) and twice should also be analysed as compound determinatives. The constructions in which they occur exhibit all the characteristic properties of FF, which we recapitulate in (39):
- (39)
(i) In FF, a single lexical or phrasal category Y(P) simultaneously realizes two functions.
(ii) FF is permitted in a category XP only between the head of XP and either an immediate dependent of XP or an immediate dependent of the immediate dependent of XP.
(iii) The fused functions are adjacent.
(iv) The category Y(P) is typically not identical to XP.
(v) The category XP is a projection of the category of the ultimate head in any counterpart construction.
(vi) The category Y(P) is the category of the dependent in any counterpart construction.
In simple cases, constructions involving FF can look superficially similar to other marked construction types, but the above properties clearly demarcate FF from these. We will conclude, therefore, by briefly comparing FF with (a) incorporation, and (b) hybridization.Footnote 43
Firstly, the compound nature of on‧ce and twi‧ce might suggest an alternative analysis in which the noun base ‧ce is merged with the determinative bases on‧ and twi‧ by incorporation. Incorporation, however, has only one property which resembles the properties in (39). To begin with, incorporation, whether treated as a syntactic phenomenon (Baker Reference Baker1988) or as morphological compounding (Rosen Reference Rosen1989, Spencer Reference Spencer1995, Anderson Reference Anderson2005: chapter 9), is correctly described as the merger of two zero-level categories in a single lexical item which has a unique function.Footnote 44 In the classic examples of incorporation, a theme or patient noun is incorporated into a verb to yield an output which has the regular functions of any intransitive verb, e.g. ‘fish‧kill’. In corresponding non-incorporated constructions, in which the noun is expressed separately as an object, the noun involved in incorporation is the head of the object (not any dependent). A schematic example would be incorporated ‘father fish‧kill son’ (=‘father killed son's fish’), compared with unincorporated ‘father kill son fish’. Notably, the incorporated noun is not necessarily adjacent to the verb, and the category of the merged word is verb (not noun). What is more, incorporation is not permitted to incorporate a phrasal category: we never find examples like ‘*father son‧fish‧kill’. The only property whereby incorporation bears any resemblance to FF is therefore property (v): after incorporation, the result of merging a noun into a verb is a verb which, as is normal, projects a VP or clause.
Finally, FF has some affinities with hybridization, a phenomenon characteristically found in nominalizations where the top of the construction is nominal and the bottom verbal. Hybrid constructions are not typical of modern English, but can be found in examples such as there's [no stopping him] (see CGEL: 1189 for further examples), where the whole construction is an NP containing the determiner no, but the head is a verb (as shown by the direct object him). In CGEL terms, the analysis of hybrid constructions is straightforward: at some point a verbal category occurs in a function normally filled by a nominal one.Footnote 45 The analysis of no stopping him is then:
(40)
The similarity with FF is that at a particular node, an unexpected category appears as head (here a VP as head of an NP). However, there is no fusion of functions in this case.
In conclusion, then, fusion of functions is a natural feature of the CGEL framework, in which function and category are systematically distinguished. Fusion of functions has clear conceptual and empirical advantages over analogous null-head analyses for a variety of English construction types. Constructions involving fusion of functions are clearly demarcated from other types of construction in which categories are merged or which have unusual heads. The detailed analysis we have given of once, twice and thrice provides further empirical evidence for the existence of FF as a distinctive syntactic construction, and implicitly supports syntactic frameworks which treat functions and categories as distinct primitives.
APPENDIX
URLs or printed sources of attested examples (websites accessed DATE)
(16a) http://www.ciao.co.uk/Alphabet_Diets_The_Y_Diet__Review_65405/SortOrder/2
(16b) http://pressforword.tripod.com/adm/interstitial/remote.html
(16e) http://www.cwrl.utexas.edu/~slatin/dillon/dillon_102398.htm.
(16f) http://www.btinternet.com/~hognosesam/bitsofstuff/bitsofstuff/pulleys.htm
(17) http://www.dsse.ecs.soton.ac.uk/newsletter/issues/2.html
(23) http://divinity.library.vanderbilt.edu/burns/2701_01/lectures/15marriage.html.
(27c) http://www4.gvsu.edu/toppera/Reviews/Keefe_Software_Reveiw.htm
(29a) http://www.firstthings.com/ftissues/ft0010/public.html.
(29b) http://www.trishandphill.optiic.com/virginia/virginia_may11.htm
(33a) Andrew Rawnsley, The Observer, Sunday July 22, 2001
(33c) Tom Clarke MP, Hansard, UK Parliament 24 Jan 1995
(33d) http://www.whatdoiknow.org/mt/mt-comments.cgi?entry_id=431
(33e) http://www.duluthsuperior.com/mld/duluthtribune/news/opinion/5636478.htm
(34b) http://homepages.goldsmiths.ac.uk/genuki/MDX/Islington/HollowayHistory.html
(36b) http://mail.nl.linux.org/spamfilter/2002-02/msg00000.html
fn.34 http://www.gurteen.com/gurteen/gurteen.nsf/0/E2D26C73DC79DDF980256C2D00475B5F/
