2 Specification as a relation of class inclusion

Among functional approaches, the concept of specificational meaning is perhaps most commonly characterised as a kind of value-variable relation (see Higgins Reference Higgins1979; Declerck Reference Declerck1988; Davidse Reference Davidse2000). For Higgins (Reference Higgins1979), this captures the observation that such sentences function like lists: the precopular noun phrase acts as the heading of the list and the postcopular elements serve as the items on that list. He notes, ‘the heading of a list provides a “variable”, thereby delimiting a certain domain, to which the items on the list conform as “values” of that variable’ (Higgins Reference Higgins1979: 155). Consequently, the specificational copular sentence in (7) can be paraphrased as the list given in (8) (see Higgins Reference Higgins1979: 154).

1. (7) The waitresses were Diane and Carla.

2. (8) The waitresses were the following: Diane and Carla.

While this characterisation is widely accepted, it is not clear from Higgins’ (Reference Higgins1979) account how specificational meaning relates to more well-defined concepts.

For Croft (Reference Croft1991), there are several possible semantic relations that can hold between two noun phrases; that is, between a predicate nominal and its argument. These include expressions of identity, class inclusion, naming and individualising, among others. The question then is which of these concepts (if any) applies to specificational copular sentences, and best captures their ‘value-variable’ characterisation. Here, I suggest that the meaning relation involved in acts of specification is one of class inclusion, ‘the relation of token to its subsuming type’ (Croft Reference Croft1991: 69). More specifically, I claim that while a meaning relation of instantiation does not entail a specifying function, the act of specification can only occur where a class inclusion relation is present.

My analysis begins with an assumption: that sentences containing definite and indefinite NP predicates ‘are equivalent in all essential respects’ and differ only to the extent that definiteness differs from indefiniteness (see Langacker Reference Langacker1991: 67). Most authors agree that postcopular indefinite noun phrases are like predicative adjectives, predicating a property of the subject referent (see, for example, Taylor Reference Taylor2002: 362). Thus, indefinite predicative NPs have an interpretation equivalent to a bare common noun interpretation. On this analysis, (9) contains a precopular referring expression (Carla) and a postcopular predicative NP which denotes the set of individuals that have the relevant property (of being waitresses). From a cognitive grammar perspective, (9) represents an instantiation relation between an instance (i.e. token, Carla) and a type specification (i.e. category, waitress).

1. (9) Carla is a waitress.

2. (10) Diane and Carla were the waitresses.

Although cognitive accounts often view definite NP predicates as inducing a distinct semantic relation from indefinite NPs, namely ‘token–token identity’ (see, for example, Croft Reference Croft1991: 69; Evans & Green Reference Evans and Green2006: 599), I assume that this relation of class inclusion is also present in sentences containing postcopular definite NPs, such as (10).Footnote ² Here too, the predicate nominal relates to a set or category, albeit one restricted by the pragmatic context: Diane and Carla are both instances of the type specification waitresses (in a particular bar).

These basic assumptions establish the view of nominal predication as an expression of set (or class) membership, as is found in set theoretic accounts. Through this relation of class inclusion, such sentences function to ascribe a property to an individual; that is, they tell us ‘something about the referent of the subject’ (Mikkelsen Reference Mikkelsen2005: 1, emphasis original). However, in (10), what I suggest is the same class inclusion relation has yet another function. If the subject is placed in focus (marked by intonation in (11)), the sentence acquires a specificational reading: the speaker is listing the membership of the category of waitresses.

1. (11) Diane and Carla were the waitresses.

Such examples are sometimes referred to as reverse specificational sentences, since they share the same specifying function as sentences like (7) above, but not the characteristic inverse word order.Footnote ³

We can account for this specifying function straightforwardly once we examine the concept of definiteness, as was argued in Patten (Reference Patten2012). As noted above, definite descriptions are understood in relation to the pragmatic context. For Hawkins (Reference Hawkins1991), they are interpreted in relation to a shared set or pragmatic (P)-set which is manifest in the speech participants’ mutual cognitive environment. From this, it follows that the category designated by a definite predicative NP will inevitably be restricted in size, with a limited membership: waitresses (in a particular bar). Indeed, for singular nouns, there is only a single member: manager (of a particular establishment).

In addition, definiteness is associated with inclusiveness. This means that the definite NP applies to all entities that satisfy the description within the P-set (see also Hawkins Reference Hawkins1978). It follows from this that when a definite noun phrase is predicated of a referring expression, the description will uniquely characterise the individual(s) referred to (see Declerck Reference Declerck1986: 30). Consequently, the referring expression will be understood to comprise a complete list of entities that make up the restricted set or type (Patten Reference Patten2012: 37). This is specification, i.e. our ‘value-variable’ relation. For example, in (11), Diane and Carla are taken to represent an exhaustive list of waitresses (in a particular bar). The properties of P-membership and inclusiveness (or uniqueness) therefore enable a specificational reading for sentences containing definite predicative NPs.

This specifying, as opposed to ascriptive, function arises whenever the token (or the instance) is in focus, rather than the type specification (or category). This can occur via intonation, as in (11), or through (what I suggest is) an inversion construction, shown in (12) and (13), repeated from (7) and (2) respectively.

1. (12) The waitresses were Diane and Carla.

2. (13) The manager is Rebecca Howe.

This inversion construction has a fixed information structure and so focus is consistently placed on the postcopular element. Once we add class inclusion semantics, we can begin to understand why this sentence type is intrinsically specificational in meaning: when focus is not on the predicative NP, the function of the sentence is no longer to ascribe a property to the referent, but the relation of class inclusion, or class-membership, still holds.

So, in specificational sentences like (12) and (13), the initial NP provides the type, or class, and the postcopular NP provides (or rather specifies) the tokens or instances that make up this category. In essence, specificational meaning is conceived of here in much the same way as by Higgins (Reference Higgins1979), as the listing of entities (despite the fact that he chooses to discuss specification separately from the concept of predication (p. 214)).

3 Inversion restrictions and non-classifying predication relations

In section 2, I set out an analysis of specificational sentences, developed in Patten (Reference Patten2012), which draws largely from the inverse tradition; that is, it involves an inverse predication relation, in which the predicative, set-denoting NP occurs in precopular position, while the NP that refers to an individual, or individuals, is situated in postcopular position (see section 1). The analysis differs from earlier formalist accounts in that it assumes that this sentence type is a construction associated with an exclusively specifying function, rather than being the result of a syntactic movement operation.Footnote ⁴ Nevertheless, it is still necessary to account for the restrictions on predicative inversion; or (more appropriately, assuming a monostratal model of language), to explain why some predicative inversion sentences are unacceptable. Indeed, since I explicitly associate the specifying function with the concept of definiteness, the unacceptability of definite NP inversions like (6), repeated here as (15), could be claimed to be especially problematic for my analysis.

1. (14) John is the one thing I have always wanted a man to be (that is, he's honest).

2. (15) *The one thing I have always wanted a man to be is John.

   (examples from Heycock & Kroch Reference Heycock and Kroch1999: 379–80)

In example (14), the postcopular definite noun phrase the one thing I have always wanted a man to be is clearly property-denoting, in that it corresponds to the adjective phrase honest. However, as shown in (15), the referential and predicative NPs of this sentence cannot be inverted. As Heller (Reference Heller2005: 187) observes, this is unexpected assuming an inverse account, as it ‘wrongly predicts them [(this class of examples)] to be acceptable’. This prediction follows especially from the assumptions of formalist theories. On formal inverse accounts, predication is treated uniformly as an interaction between expressions of type e and <e,t>; that is, between individuals and sets of individuals (or properties). Since the one thing I have always wanted a man to be is predicative (and so of type <e,t>), it should be subject to any movement rule that can apply to predicative elements. As Heycock & Kroch (Reference Heycock and Kroch1999: 380) note, the unacceptable outcome in (15) ‘rules out any analysis under which the “inverse” construction really involves movement of the predicate’.

However, on a monostratal model of language, movement is not invoked as a means of deriving specificational inversion structures. Instead, I claim that the formal characteristics of the NP inversion construction are governed by its specifying function. On adopting the analysis of specificational meaning outlined in section 2, the unacceptability of sentence (15) becomes unsurprising, and is in fact anticipated. On this account, the particular kind of predication relation that is relevant to acts of specification is class inclusion, the relation between token and its subsuming type. Only sentences that involve an instantiation relation can enable a specificational interpretation and, consequently, are expected to correspond to acceptable instances of the specificational inversion construction.

As Partee ([Reference Partee, Partee, Groenedijk, de Jong and Stokhof1986] Reference Partee, Partee, Groenedijk, de Jong and Stokhof2004: 220) observes, predicative <e,t> uses of a noun phrase can involve a different kind of predication relation from class-membership. She considers the idiosyncratic behaviour of sentence (16); here, ‘the entities in the extension of color are colors: blue, red, green etc. – not shirts’ (p. 220). Semantically, (16) amounts to the combination of the two distinct classifying relationships found in (17), albeit with the colour unspecified.

1. (16) This shirt is a nice color.

2. (17)

   1. (a) This shirt is blue.

   2. (b) Blue is a nice color.      (examples from Partee Reference Partee, Partee, Groenedijk, de Jong and Stokhof2004: 220)

While in (17a) the adjective blue is predicated of this shirt, in (17b) it functions as a type e referring expression – an ‘individualised’, or nominalised property – which is predicated of by the NP a nice color. Partee (Reference Partee, Partee, Groenedijk, de Jong and Stokhof2004: 221) labels color an ‘attribute noun’; as such, it expresses a property of properties.

The example in (14) appears to represent a definite noun phrase counterpart to the predication relation observed by Partee (Reference Partee, Partee, Groenedijk, de Jong and Stokhof2004). This sentence asserts that John has a desirable quality. However, it does not classify John as a member of some set because the animate John cannot be classified by the inanimate thing. Instead, the one thing I have always wanted a man to be describes a property that John has. That is, while this definite description expresses a property, the entities in its extension are also properties (honest, reliable, passionate etc.) rather than objects or animate beings. As Partee (Reference Partee, Partee, Groenedijk, de Jong and Stokhof2004: 221) suggests, ‘the combining stem -thing can function as [an] attribute noun’. The relation that exists between the subject and the predicate nominal in (14) is therefore quite distinct semantically from ordinary cases of class inclusion. Here, the one thing I have always wanted a man to be classifies (or has as its extension) a particular property of John; it does not classify John directly. Since example (14) does not involve a classifying relationship, we would not expect it to correspond to an acceptable specificational inversion sentence, as shown in (15).

This account offers an explanation for the different acceptability of sentence (19) as compared to (15). In (18), honest replaces the proper name John as predicated of by the postcopular definite NP. When occurring in subject position of the predicate nominal construction (a slot that must be filled by a referring expression), honest refers to an ‘individualised’ property; that is, a non-individual which is nevertheless presented as a discrete entity (see Szabolcsi Reference Szabolcsi1983; É. Kiss Reference É. Kiss1998). Unlike the animate John, the subject honest can be properly classified as an inanimate thing: honest is contrasted with other properties as uniquely matching the description thing I have always wanted a man to be, and so comprises the sole instance of this single member category. Thus, (19) makes for an acceptable specificational inversion sentence.

1. (18) Honest is the one thing I have always wanted a man to be.

2. (19) The one thing I have always wanted a man to be is honest.

   (Heycock & Kroch Reference Heycock and Kroch1999: 379)

It is therefore possible to account for the problematic data presented by Heycock & Kroch (Reference Heycock and Kroch1999), despite maintaining what is essentially an inverse analysis. Specifically, only relations of class inclusion or instantiation, as in (18), can invoke a specificational reading. Predicates that do not directly classify the referent do not enable a specifying function and so are not expected to occur in the specificational inversion construction. On this account, the unacceptability of sentence (15) is entirely predictable.

What is special then about the account presented here is that it assumes, with Mikkelsen (Reference Mikkelsen2005) and many other formalists, that the specificational sentence structure is a form of NP inversion, with the semantic predicate preceding its argument; however, rather than looking for categorial constraints on the movement operation of predicate raising to account for the range of acceptable specificational sentences, it suggests that these sentences form a particular inversion construction, which has its own construction-specific properties and stipulations: namely, an exclusively specifying function. As Heycock & Kroch (Reference Heycock and Kroch1999) and Heller (Reference Heller2005) note, there is little difference in the category, complexity or discourse status of the predicates in (14) and (18); in both, the one thing I have always wanted a man to be is a definite noun phrase, is property-denoting, and designates a single member category. Instead, where the difference lies for these two sentences is in the particular kind of relationship that holds between subject and predicate nominal, and whether a specificational interpretation can result from it. Here, the focus is brought back to the concept of specificational meaning, as it was for Higgins (Reference Higgins1979), with the restrictions on inversion viewed in terms of what makes an acceptable specificational sentence.

Essentially, the unacceptability of the predicative inversion in (15) boils down to an issue of matching between value and variable (for us, between instance and type specification). According to Higgins (Reference Higgins1979: 155), this represents the most basic kind of connectivity effect apparent in specificational sentences; he states that ‘a well-formed list contains items that conform to the heading at the top of that list’. The reason is, as Higgins (Reference Higgins1979: 150) puts it, that in a specificational sentence, the postcopular element ‘says what constitutes or makes up’ the denotation of the precopular NP. Higgins gets very close to the concept of class inclusion here; he cites Akmajian (Reference Akmaijan1970: 19), who, in talking about clefts, notes that ‘the focus item must belong to the appropriate semantic class, i.e. the class represented by the variable’ (my emphasis).

4 Discounting alternative semantic relations

As was noted above in section 2, Croft (Reference Croft1991: 69) recognises several possible semantic relations that can hold between two noun phrases, including expressions of identity, class inclusion, naming and individualising, which ‘does not exhaust the semantic possibilities’ (p. 70). In section 3, I argued that the relevant relation to acts of specification is class inclusion (the relation between instance and type specification). Copular sentences which do not involve a directly classifying relation, specifically those involving what Partee (Reference Partee, Partee, Groenedijk, de Jong and Stokhof2004: 221) labels as a ‘special predicative use of attribute NPs’, were found not to allow a specifying interpretation. It follows from this account that indefinite NP predicates that ‘characterize. . .without subsuming’ the logical subject (Croft Reference Croft1991:70), labelled individualising by Bolinger (Reference Bolinger1980), will also be excluded from the specificational inversion construction, as I go on to suggest in section 5. However, first I consider, and discount, two alternative proposals, on which specificational meaning is characterised as naming and identity, respectively. The latter warrants special consideration, since the association of specification with identity forms the basis of the equative approach to specificational sentences, which is set up in opposition to inverse analyses (see section 1).

5 Inversion restrictions and indefinite NP predication

In this section, I turn to the second set of data that has presented a longstanding problem for traditional inverse accounts: namely, the restrictions on specificational sentences introduced by indefinite NPs. As shown in (45), repeated from (4) above, predicative NPs marked as indefinite often cannot occur in precopular position.

1. (44) Rebecca Howe is a manager.      [predicational]

2. (45) #A manager is Rebecca Howe.      [specificational]

In (44), repeated from (3) above, the postcopular NP is clearly predicative, providing a description of the subject referent (rather than referring to a distinct individual). Furthermore, this predication relation is one of class inclusion: Rebecca Howe is a member of the class or category of managers. Consequently, the unacceptability of the inversion in (45) cannot be accounted for by appealing to the insight that specification is a type of classifying relation; it is clear that while specificational meaning is dependent upon the relation of class inclusion, class inclusion does not entail a specificational interpretation.

Nor can we say that because specification is tied to the concept of definiteness (as was argued in section 2 above), it follows that indefinite NP predicates cannot occur in the specificational inversion construction. This is not simply a categorial distinction; we do find acceptable examples introduced by indefinite noun phrases, including (46).

1. (46) A philosopher who seems to share the Kiparskys’ intuitions on some factive predicates is Unger (1972), who argues. . . (Delacruz Reference Delacruz and Partee1976: 195 [Mikkelsen Reference Mikkelsen2005: 155])

Nevertheless, in the following discussion, we will see that the restrictions on indefinite specificational sentences can be accounted for by the combination of these properties: specification requires a class inclusion relation, but also prefers a relation involving definite NP predicates. The defining characteristic of acceptable indefinite noun phrases therefore lies in their degree of closeness to definite noun phrases.

This suggestion – that the behaviour of indefinite specificational sentences rests upon the distinction between definiteness and indefiniteness – is not a new one. However, such arguments usually revolve around their differing discourse status. For example, Mikkelsen (Reference Mikkelsen2005) claims that inversion (which for her is a syntactic movement operation) is governed by a discourse requirement: the predicative NP will only be raised to subject position if it contains discourse-old information. On this account, ‘we would expect indefinite specificational subjects to be rare’ (Mikkelsen Reference Mikkelsen2005: 154): it is well known that while the definite article is associated with familiarity, indicating that the entity can be located in the previous discourse, the indefinite article instead signals that the entity is new to the discourse. Acceptable inversions like (46) contain indefinite NPs that are somehow linked to the preceding text. (46) is from an article on factive predicates, in which the Kiparskys figure prominently. Although the existence of a philosopher who shares the Kiparskys’ intuitions is new information, the embedded phrase the Kiparskys’ intuitions on some factive predicates connects this footnote to the previous discussion.

On the assumption that specificational copular sentences are instances of NP inversion, it is not surprising that they are sensitive to discourse status: from her corpus of English inversion sentences,Footnote ⁸ Birner (Reference Birner1994, Reference Birner1996) observes that they are subject to a robust pragmatic constraint, such that ‘the preposed element . . . must not be newer in the discourse than the postposed element’ (Birner Reference Birner1996: 90). However, discourse status cannot explain all of the distributional facts surrounding indefinite specificational sentences. Using examples similar to (45), Mikkelsen (Reference Mikkelsen2005: 159) shows that their unacceptability prevails regardless of the discourse context; she concludes that ‘there are still other factors that seem to play a role in determining the felicity of indefinite specificational subjects’.

In Patten (Reference Patten2012), I claimed that we can better account for the data if we attend also to the semantic distinction between inclusiveness and exclusiveness, which further differentiates definiteness from indefiniteness. This follows from the interpretation of specification as a class inclusion relation. Definite NP predicates are well suited to acts of specification, which involves listing the entities (or instances) that comprise the membership of a particular category (see section 2). The characteristics of inclusiveness and P-set membership mean that the individual(s) referred to will be taken to form an exhaustive list of entities that satisfy the description. This creates an ideal scenario for an act of specification, on the principle that a well-formed list should ideally be complete (see also Declerck Reference Declerck1988: 30). For example, in (47), repeated from (10), Diane and Carla together make up the set of waitresses (in a particular bar). Likewise, in (48), repeated from (1), Rebecca is the only manager (of this establishment).

1. (47) Diane and Carla were the waitresses.

2. (48) Rebecca Howe is the manager.

The characteristic properties of indefinite descriptions, however, are less suited to the specifying function.Footnote ⁹ The use of the indefinite article conversationally implicates exclusiveness, or non-uniqueness (see Declerck Reference Declerck1987; Hawkins Reference Hawkins1991). Thus, (49) implies that Carla is not the only waitress; it is ‘possible (and in fact suggested) that the property can be ascribed to other people as well’ (Declerck Reference Declerck1986: 30). Furthermore, indefinite NPs are not always understood in relation to a pragmatically given context, and so often denote general categories (see Hawkins Reference Hawkins1991: 419). Since, in (49), the indefinite NP is unmodified, we can interpret the sentence as including Carla within the unrestricted set of waitresses that exist. It follows that, while (49) tells us something about Carla (providing an ascriptive assertion), it does little to inform us about the membership of the category waitresses. This explains why inversions involving unmodified indefinite noun phrases, like (50) and (45), are deemed unacceptable.

1. (49) Carla is a waitress.

2. (50) #A waitress is Carla.

In contrast, the indefinite noun phrases that occur in the initial position of acceptable specificational inversion sentences do not exhibit these characteristic properties of indefiniteness, and so are closer in interpretation to definite noun phrases. Example (46) contains an establishing relative clause. As well as anchoring the indefinite description to the speech event, the restrictive modification serves to further specify the type designated by the indefinite noun phrase. This limits the number of possible instances or tokens of the type category. Furthermore, as Hawkins (Reference Hawkins1978: 225) points out, ‘the exclusiveness condition is no longer operative’ in indefinite NPs containing establishing relative clauses. Consequently, there is no implication that there is more than one philosopher who shares the Kiparskys’ intuitions on some factive predicates.

What characterises acceptable indefinite noun phrases, then, is that ‘they bridge the gap between definiteness and indefiniteness’ (Patten Reference Patten2012: 54). This is true not only of their discourse status (in that they establish a brand-new type description, which is nevertheless grounded to the speech event) but also in the fact that they designate restricted categories and allow us to be non-committal with respect to exclusiveness and inclusiveness. Specificational sentences containing indefinite descriptions therefore have a unique and useful function, which separates them from those containing definite NPs. A similar point is made by Declerck (Reference Declerck1988: 20), who claims that their use is ‘motivated by the semantic difference between these NPs and the corresponding definites’. However, while Declerck associates these examples with the characteristics of indefiniteness, namely exclusiveness (as indicating that there are other entities that satisfy the description provided), we can see that the form of indefinite specificational clauses is governed instead by the degree of closeness to definiteness – a concept better suited to acts of specification.

One advantage to this analysis is that it anticipates the gradient acceptability judgements associated with such sentences. As Donka Farkas observes (in Mikkelsen Reference Mikkelsen2005: 159), the content and form of modification seems to influence the degree of felicity of indefinite specificational subjects. Such behaviour is difficult to account for on inverse analyses drawing from the formalist tradition, which seeks to identify general constraints on syntactic movement operations.

For example, Patten (Reference Patten2012: 54–5) compares pairs of examples like (51) and (52). Here, the initial description is anchored through the discourse-old head noun waitress. This carries with it an implicature of P-membership, and so is understood within the more restricted context of waitresses in the Boston Barmaid contest. The restricted type is further specified by adjectival modification, which also ensures that the description is novel to the discourse. Nevertheless, despite clear similarities in form and discourse status, the two sentences differ in felicitousness.

1. (51) There are several waitresses in the Boston Barmaid contest. An especially efficient waitress is Diane Chambers.

2. (52) There are several waitresses in the Boston Barmaid contest. ?An efficient waitress is Diane Chambers.

These felicity judgements are accounted for as follows. While the adverb especially enables a uniqueness interpretation, such that Diane Chambers could well be the most efficient waitress within the P-set, it is more difficult to obtain a non-exclusive reading without this adverb. Efficient is a gradable adjective, and so we might expect there to be more than one efficient waitresses within the shared context (even if some are more efficient than others).

It seems then ‘that indefinite noun phrases will be better suited to the specifying function’, and therefore more amenable to the specificational inversion construction, ‘if they contain lots of modifying information’, and if the content of modification does not rule out an interpretation of uniqueness or inclusiveness (Patten Reference Patten2012: 55). Although this claim warrants empirical investigation, the explanation sketched here gains further support from the findings of wider studies of English inversion constructions. In a discussion of adjective phrase inversion, Dorgeloh (Reference Dorgeloh1997: 82) observes that inversion is deemed more acceptable if the AP is marked as ‘a comparative, or if accompanied by also, especially’ and the like, as in (53).

1. (53) Especially worrisome to public health experts is the growing number of cases caused by tuberculosis germ strains that have become resistant to drugs. . .

   (New York Times, 15 July 1990 [Birner Reference Birner1996: 41])

Although other authors have attributed this finding to discourse familiarity (see Penhallurick Reference Penhallurick1984: 43; Birner Reference Birner1996: 40–2), in Dorgeloh's view, ‘this is not the whole story’. She suggests that these APs denote ‘a property which is particularly attributable to the discourse entity that follows [the copula]’; the construction therefore has a contrastive function, singling out the referent that exhibits this property to a greater extent than the other individuals or items that also have this property (Dorgeloh Reference Dorgeloh1997: 82).

As with indefinite noun phrases, shown in (51) and (52) above, the suitability of the adjective phrase in performing this function has consequences for the acceptability of the sentence, shown in (54) and (55); see also the examples in Penhallurick (Reference Penhallurick1984: 43).

1. (54) Particularly impressive was Diane's demeanour.

2. (55) ?Impressive was Diane's demeanour.

Again, we might say that without the adverb particularly creating a sense of uniqueness, or rather distinctiveness, the property of being impressive could characterise a number of things in addition to Diane's demeanour. Clearly, the AP inversion data is relevant to the characterisation of specificational meaning developed here. This lends additional support to the inverse approach to specificational sentences, and builds up a picture of a family of predicative inversion constructions sharing a similar specifying function.

Finally, if we return once more to the behaviour of definite noun phrase predicates, we find yet further evidence to suggest that specificational meaning is at odds with an indefiniteness interpretation. In example (56), the perfect gentleman behaves differently from most definite NP predicates in that it does not function to uniquely characterise Sam in relation to some shared set of individuals. Instead, the intended interpretation is that ‘Sam behaved in a gentlemanly fashion’. As we can see in (57), substituting the indefinite article does little to change the meaning of the sentence. As Declerck (Reference Declerck1986: 35) notes ‘predicational NPs that are formally definite sometimes yield an indefinite interpretation’.

1. (56) Sam was the perfect gentleman.

2. (57) Sam was a perfect gentleman.

It is telling that the sentence in (56) does not enable a specificational reading, and does not correspond to an acceptable specificational inversion sentence, as shown in (58).

1. (58) #The perfect gentleman was Sam.

Such data point once more to the lack of a clear boundary between the concepts of definiteness and indefiniteness, which I have suggested is crucial to understanding the use of definite and indefinite noun phrases in specificational NP be NP sentences. However, while the unacceptability of (58) may be explained by a lack of inclusiveness, it is possible too that what we are witnessing here is the absence of a class inclusion relation in both (56) and (57).

These examples are relevant to the individualising predication relation recognised by Croft (Reference Croft1991: 70). In Bolinger (Reference Bolinger1980), this label is applied to indefinite NP predicates that lack a classifying function (in contrast to bare, articleless nouns, which are always classifying). In order to tease apart these distinct functions of the indefinite article in English, Bolinger (Reference Bolinger1980: 3) considers the behaviour of gradable nouns with degree adverbs that indicate the extent to which the subject has an individual quality. For example, in (59), such an and a complete indicate (or intensify) the extent of Carla's angelic behaviour. Carla is not necessarily a real angel; nor is she being compared to or distinguished from other angels through degree modification. Instead, (59) states that Carla ‘acts/is acting like’ an angel; that is, she ‘has the characteristics or behaviour’ associated with this category (Croft Reference Croft1991: 70).

1. (59) Carla is such an/a complete angel today.

Sentence (59) can therefore be said to have an individualising, rather than classifying, meaning; as Croft (Reference Croft1991: 70) notes, the predicative noun phrase ‘characterizes the subject without subsuming it’. Such examples clearly have much in common with the data in (56) and (57), where, as we saw, perfect gentleman describes Sam's behaviour as gentlemanly, but does not distinguish Sam from some set of gentlemen. The use of the definite article in (56) serves only to emphasise the extent of Sam's gentlemanly behaviour. Consequently, this example seems to function as a definite noun phrase counterpart to the individualising relation recognised for indefinite NP predicates.

It would therefore support the analysis of specificational sentences outlined in section 2, if it were shown that (in addition to the property of non-exclusiveness) the presence of a classifying predication relation was just as important for the felicity of indefinite specificational sentences as definite ones. And indeed this does seem to be the case. For example, in (51), Diane is distinguished from the other members of the set of waitresses in the shared context by a potentially unique (or at least distinctive) description. In contrast, the non-classifying predication relation in (59) does not allow a specificational reading for the corresponding inversion sentence in (60), irrespective of the presence of adjectival modification in the indefinite NP.

1. (60) A complete angel is Carla.

While the sentence in (60) is grammatical, it is acceptable only on an ascriptive reading, in which a complete angel predicates a property of, and ascribes a property to, Carla. That is, the sentence tells us something about Carla, rather than specifying Carla as matching an anchored description. The sentence is an example of a rare construction distinct from our specificational inversion sentences, with an information structure that places focus on the precopular predicate (see Renský Reference Renský, Esser and Hübler1981: 140).Footnote ¹⁰

The importance of class inclusion to the felicitousness of indefinite specificational sentences is made even clearer when we consider the inversions in (61) and (62) below. While (61) involves an instantiation relation, with blue serving as an instance of the type colour, the referring expression this shirt cannot be interpreted as a member of this category. The inversion in (62) therefore corresponds to the non-classifying predication relation discussed in section 3; that is, to Partee's (Reference Partee, Partee, Groenedijk, de Jong and Stokhof2004: 221) ‘special predicative use of attribute NPs’. As we would expect from the account provided in section 2, the two sentences differ in acceptability. Restrictive modification in (61) makes for a perfectly good specificational sentence; however, it cannot improve sentence (62), which does not involve the requisite classifying relation.

1. (61) A particularly attractive colour is blue.

2. (62) #A particularly attractive colour is this shirt.

What this shows is that indefinite NP inversion is subject to the same constraints as specificational copular sentences introduced by definite NPs. Both constructions are governed by the concept of specification; that is, by what constitutes a ‘well-formed list’.

6 Summary and conclusion

This article has re-examined the case for an analysis of specificational copular sentences as instances of predicative inversion. Its focus was on the restrictions on inversion that have proven difficult to account for on formal inverse analyses; and that have motivated alternative analyses of specification as semantic equation. The article demonstrates that we can account for the restrictions on inversion by focusing on the concept of specificational meaning that is integral to this construction. In viewing specification as a construction-specific requirement or stipulation, the distributional facts are shown to fall out straightforwardly.

Inversion restrictions involving definite NP predicates were shown to stem from the claim that only a particular type of predication relation, namely class inclusion (or instantiation), permits a specificational reading; non-classifying semantic relations, in which the predicative NP does not subsume the logical subject, do not sanction a specificational reading and so cannot occur in the specificational inversion construction. This stipulation was shown to capture individual examples exhibiting these restrictions more successfully than an equative approach to specificational sentences, despite repeated claims that they ‘pose a serious problem for predicative inversion analyses’ (Heycock Reference Heycock2012: 226).

Inversion restrictions involving indefinite NP predicates were claimed to relate both to their discourse status (Mikkelsen Reference Mikkelsen2005) and to the implicature of exclusiveness commonly associated with indefinite noun phrases (Patten Reference Patten2012). Both accounts are consistent with the findings of wider studies of predicative inversion, thus supporting an inverse analysis. Finally, it was argued that all specificational NP be NP sentences are subject to the same stipulation: specification requires a class inclusion relation, and prefers NPs that exhibit properties closer to definiteness than indefiniteness.

In conclusion, straightforward explanations for inversion restrictions emerge from a reconsideration of what constitutes specificational meaning, conceptualised here, as in Higgins (Reference Higgins1979), as ‘a well-formed list’. This follows Higgins’ (Reference Higgins1979: 362) suggestion that ‘further work. . .should probably concentrate on. . .testing the fruitfulness of the notion that Specificational. . .sentences are essentially lists’.