2.1 Relative clause introducers vs relativised variables

First of all, what is a ‘relative clause’? Essentially, an ‘integrated’ relative is a subordinate clause which is embedded within an NP, and which functions as a modifier (and not as a complement) of the head noun of that NP (see e.g. McCawley Reference McCawley1981; Fabb Reference Fabb1990; Huddleston & Pullum Reference Huddleston and Pullum2002). A restrictive relative acts as a ‘syntactic adjective’ (Benveniste Reference Benveniste1966: 222), a ‘restrictor’ (Dik Reference Dik1997: 25) or an ‘attribute’ (Schachter Reference Schachter1973: 42), in the sense that it further specifies the defining property that any entity capable of being referred to by the NP as a whole must necessarily possess. As a predicate, it serves to restrict the broad intensional categorisation already made of the referent at issue by the head noun (itself a predicate, as a common noun) which is modified: see also Chierchia & McConnell-Ginet (Reference Chierchia and McConnell-Ginet1990: 330). An a-restrictive relative (see below), however, serves simply to elaborate the denotation of the head noun of the containing NP.

As we have just seen, RRCs are not the only type of NP-integrated relatives: see sections 3.2, 3.3 and below for a discussion of what we might call ‘a-restrictive’, but nevertheless ‘integrated’, relative clauses.Footnote ² An ARRC, often embedded within an indefinite NP, conveys discourse-new non-identifying information which is nonetheless relevant to the entity being designated by the containing NP as a whole in terms of its role in the broader discourse context being developed. An attested example is given in (1) (see sections 3.2 and 3.3 below for more in-depth discussion):

1. (1) COMEDY Joe Orton's play is about money hidden in a coffin that gets buried in a churchyard. . . . (Film synopsis of Loot, Radio Times 30.07–05.08.16, p. 37)

Since it is clearly necessary to find a term which covers the subset of NP-integrated relatives which in no sense ‘restrict’ the host head noun's sense, we are proposing the term ‘a-restrictive’ (i.e. ‘neither “restrictive” nor (strictly speaking) “non-restrictive”’), retaining ‘restrictive’ for integrated relatives which genuinely serve to restrict the sense of their head noun host.

In communicative terms, while RRCs help recipients pick out or identify a particular referent, NRRCs, which are in apposition to their host phrase or clause, do not: their use assumes that the referent of the relative pronoun has already been identified, and they serve simply to add further, non-identifying (though discourse-relevant) information about it. Indeed, as fully fledged discourse units corresponding semantically to complete predications, carrying a potentially independent illocutionary force in relation to the main clause, NRRCs perform certain specific functions within the wider discourse: see Loock (Reference Loock2007, Reference Loock2010) and section 4.2. Integrated relatives (ARRCs and RRCs), on the other hand, correspond semantically only to (complex) predicates. As such, they do not correspond to potential discourse units, unlike NRRCs (see Cornish Reference Cornish2009: 584).

Now, since the relation involved in the case of RRCs as well as ARRCs is contracted with the head noun of its containing NP (as will be argued below: see section 3.1), the relation cannot be one of coreference, stricto sensu, where each member of the relation individually targets the same external referent. Hence, quantified, non-referential hosts are just as acceptable as definite, referential ones, though often unacceptable in NRRCs (see section 4.1, and in particular fn. 23, for evidence and illustration). RRCs and ARRCs involve a relation of anaphoraFootnote ³ between an NP's head noun and the relative pronoun, while NRRCs normally contract a relation of coreference between the relative pronoun and an independently referring phrase or clause in its immediate vicinity (see section 4.2). Yet many authors (e.g. Malan Reference Malan1999: 43 and passim) write indiscriminately of a relation of ‘coreference’ between relativisers and their antecedents in the case of NP-integrated relatives, and between relative pronouns and phrasal heads of various types in that of clause-external or non-restrictive relatives (though Malan's book is in effect restricted solely to integrated relative clauses).

Arguably, the host noun in the case of integrated relatives is not an ‘antecedent’, in the traditional sense, but rather an ‘antecedent trigger’ in Cornish & Salazar Orvig's (Reference Cornish and Orvig2016: 634) conception:Footnote ⁴ it simply provides the sortal category of entity which is to be built into a fully fledged referent via the anaphoric (here relative) clause itself. It is only when the interpretation of the whole NP is set up that such a referent may be established, the relativised variable (see below) constituting a proxy for this. And in the case of NRRCs, it seems clear that the RC has a similar function, perhaps clearest of all in the case of propositional NRRCs, where the RC has the effect of actually creating its referent from within the immediately preceding co-text, as in (2):

1. (2) He's [Garrison Keillor] just stopped hosting A Prairie Home Companion, which is a tragedy. . . (Interview with Susan Rae, Radio Times 30.07–05.08.16, p. 17)

Here, the NRRC consists of a predicate nominal clause interpretable as ‘X being a tragedy’. As such, for reasons of discourse coherence, it is more likely to target a propositional host, namely ‘the fact that GK had just stopped hosting the programme A Prairie Home Companion at the time of utterance’, than the programme in question itself (which would constitute a nominal NRRC host): but the programme itself is not a ‘play’, and its title could not sustain an eventive interpretation.Footnote ⁵

Syntactically, a relative (of whatever type) is a constituent of the type C´´ (i.e. it has the structure of a clause) in X´ syntax.Footnote ⁶ In English, it may either be introduced by a relative pronoun in wh- or (if an integrated relative) by a relative marker – henceforth ‘RM’: that or zero – followed by an incomplete clause: what is missing here is the constituent which has been relativised. This missing constituent (more exactly, the unexpressed constituent which is bound by the pronoun and which Dik Reference Dik1997: 3 terms ‘relativized variable’ – henceforth ‘RV’) is perceptible once the introducer is omitted, since the sequence of elements which remains will not then correspond to an independent, stand-alone sentence. Rather, it will contain a gap in a particular functional position in the clause, a gap whose functional-syntactic role is reflected by those clause-introducing wh-pronouns that show case information (i.e. who, whom and whose: see also Chierchia & McConnell-Ginet Reference Chierchia and McConnell-Ginet1990: 331). In the case of the place and time wh-markers where and when (which are relative adverbs rather than ‘pronouns’ stricto sensu), the gap is less easy to detect in this context, these forms not being strictly argumental in character. For example, in The city [where John met Mary] is in France, the outcome John met Mary appears at first sight to correspond to a possible sentence; however, it is a ‘system sentence’, not a ‘text sentence’, in Lyons's (Reference Lyons1977: 622–35) terms, since it would be wanting as a potential utterance. The place (as well as time) at which the meeting indicated occurred would be highly assumed. As a potential text sentence, then, such a construction would be felt as incomplete in context (unless of course the place variable was bound by default in a given context).

Henceforth, a relative marker realised in the English examples by zero will be symbolised as ø, and the relativised variable as ®. The relative pronoun both signals the subordination of the clause it introduces in relation to the main clause, as well as to the head noun of the containing NP, and binds the relativised variable that corresponds to the referent that is common to the two clauses. The relative marker is fused morphologically with the relative operator in the shape of a relative pronoun (see also Dik Reference Dik1997: 48).

This distinction between ‘relative marker’ and relativised variable (expressed in the languages exemplified below by an ordinary pronoun: nó in (3a) and masa in (3b)) can be illustrated by examples drawn from two unrelated African languages, Akan (3a) and Hausa (3b):

1. (3)

   

^{Footnote 8,Footnote 9}

These data reveal very clearly the distinction between the relativised variable (marked in situ by an ordinary pronoun) and a simple clause-introducing element in relative clauses: áà in (3a) and da in (3b). According to Jacobson (Reference Jacobson1989: 201), a number of other languages (e.g. Albanian and Tagalog) introduce relative clauses by means of a marker identical to a subordinator; and McCloskey (Reference McCloskey1979) argues that this is also the case in modern Irish (cf. also Chinese, according to Pu Reference Pu2007, where it is null).

However, the existence of the RV is left out of account by descriptions of relatives in standard English such as Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985) and Malan (Reference Malan1999), who systematically conflate RC introducers (whether wh-pronouns, that or zero) and RVs by characterising all these introducers, respectively, as directly assuming syntactic-functional roles in the clause, or as being ‘in subject or complement position’. Of course, they do and are not, the structural positions within the clause being distinct in each case – even when the RV occupies subject position in it.

2.2 Relative markers vs wh-marked relative pronouns: distribution and functions

In English too, RMs and relative (wh-)pronouns may be distinguished: that in English is arguably not a relative pronoun,Footnote ¹⁰ but simply marks a departure from and (indirectly) a connection with the main clause, and signals that the clause it introduces is subordinate.

Yet Malan (Reference Malan1999: 12) and Larreya & Rivière (Reference Larreya and Rivière2010: 354) specifically claim that whereas the complement clause introducer that is a pure ‘conjunction’, the relative clause introducer that (and even zero) is a relative ‘pronoun’, which, as such, fulfils syntactic functions within the clause. (See also Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 366, table 6.33, as well as van der Auwera Reference Van der Auwera1985 and Loock Reference Loock2010: 14–15 for such a position.) We may surmise that this conception is a direct consequence of these authors’ conflating the RV and the relativiser, a position which necessarily entails that that and zero as RC introducers must also be ‘relative pronouns’. However, this section gives evidence that the distinction between complement-clause that and RM that is invalid, and that the that (as well as zero) of relative clauses is also a simple subordinator. In this context, it is a pure relative marker (see also Dik Reference Dik1997: 48, fn. 4, Miller Reference Miller2008 and Miller & Fernandez-Vest Reference Miller, Jocelyne Fernandez-Vest, Bernini and Schwarz2006):

1. (4) <x c32> . . . they were cracking jokes about it during one of the shows we went to see at night about [the little ladies that [RM] ® [RV] wear gym shoes and ® [RV] keep five machines going]. . . (Spoken extract from ‘The Great American Sandwich’ corpus)

Note here that the RC introducer that, a purely grammatical morpheme in its guise as a subordinator, is thereby neutral as to the ‘human’ vs ‘non-human’ distinction encoded by the veritable relative pronouns who vs which, respectively. See also Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985: 1248).

Let us first briefly review some of the main members of the set of English relative pronouns. In standard English, there is the series of relative pronouns in wh-: what (indefinite, non-human), which (definite, specific non-human), w ho/whom (human), whose (who + the possessive morpheme),Footnote ¹¹where (place), when (time), etc. However, in current spoken as well as written English, the ‘place’ relative pronoun/adverb where tends to be used more abstractly, in a more general ‘locating’ sense (see examples (5a and b) below). Apart from the forms who (subject), whom (non-subject) and whose (genitive), wh-pronouns are not specified for grammatical function: they may bind subject, direct, indirect or oblique object RVs. These features (case-marking where it applies, restriction or otherwise to human referents, to places or times, or to specific or non-specific entities – e.g. which vs what) are characteristic of pronouns cross-linguistically, being potentially referring expression types, though not of subordinators, which are purely grammatical elements. Note that what, unlike the other wh-pronouns, contains its own ‘antecedent’, namely that (+ which): see Quirk et al. (Reference Quirk, Greenbaum, Leech and Svartvik1985: 1060). It is this factor, arguably, that accounts for the fact that, alone among the wh-pronouns, there is no ‘gap’ (RV) in relatives whose introducer is what.

We find wh-pronouns in a-restrictive relatives, as in (5a), restrictive relatives (5b), non-restrictive (appositive) nominal relatives (6) and propositional relatives (7): see also example (2) above:

1. (5)

   1. (a) Jim Davidson hosts the snooker game show where world-class players compete against the clock to win prizes for the contestants. . . (TV programme note, Radio Times 11–17.08.01, p. 66)

   2. (b) All through school I knew that I wanted a job where I used my brain but also helped people. . . . (Interview with Jo Hardy, Radio Times 16–22.08.14, p. 21)

1. (6) A recent study by Oxford University, which compared blood levels of B12 in 689 men, found that 52 per cent of vegans, 7 per cent of vegetarians, but only one omnivore, were vitamin B12 deficient. . . (Michael Mosley, Radio Times 16–22.08.14, p. 9)

1. (7) <x c28> . . .so he said if you don't play the machines somebody else is losing an awful lot of money you know which is true (Spoken extract from ‘The Great American Sandwich’ corpus)

Note first of all that neither that nor zero could replace the wh-pronoun/adverb where in the nonetheless integrated relative clauses in (5a) and (5b). These RMs would also be ruled out almost by definition as clause introducers of the non-restrictive relatives in (6) and (7) (see Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 1257; Dik Reference Dik1997: 39; see Loock Reference Loock2007, Reference Loock2010 and section 4.2 below for this subtype). Corpus data confirm the virtual absence of that as an NRRC introducer too.Footnote ¹² The reason for this exclusion may be adduced as follows.

This function evidently normally requires the use of a pronominal, referential expression, something which RMs are not. Since the relative clause is non-restrictive in each case, it is syntactically and semantico-pragmatically quasi-independent in relation to the main clause. This may be expressed formally by the possible presence of a comma before and after the relative clause (and by pauses or other disjunctures in the spoken realisation).

Now, in order to co-refer with a phrasal or clausal antecedent across the ‘barrier’ represented by the comma or the pause/disjuncture, a referentially stronger expression-type than a simple subordinator like that or ø is needed. Wh-marked relative pronouns may occur here, since they are precisely pronouns, a type of expression capable of having a reference. Subordinators, on the other hand, serve only to ensure the connection between main and subordinate clauses, simultaneously signalling the presence of the latter. But there is no tight connection between main clause and NRRC – in fact there is a syntactic break between them: hence an essentially clause-connecting marker such as subordinator that cannot naturally occur, given that the two clauses are by definition not being integrated into a subsuming syntactic unit.Footnote ¹³ However, this characterisation is no doubt too strict a description, in view of the (albeit undoubtedly marginal) attested instances of that introducing NRRCs noted in fn. 12. To avoid any possible confusion, it is important to strictly distinguish between the distal demonstrative determiner/pronoun that (as in Look at that bird, What is that?, respectively) and the subordinator that which we also find as RM – which has a somewhat different phonology: /ðət/ as opposed to /ðæt/ (see also Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985: 1251, n. (a); Malan Reference Malan1999: 12).

Note also the close parallelism between that and zero in their RM function, on the one hand, and the very same morphemes as introducers of other types of subordinate clauses, on the other. Just as that and zero may introduce relative clauses modifying a head noun within an expanded NP (the idea that/ø Wilbur just expressed ® is ridiculous), so these same subordinators may herald clauses acting as complement of an (abstract) head noun (and also of certain head verbs, adjectives or prepositions): the idea that/ø Wilbur might not accept the invitation is ridiculous.Footnote ¹⁴

Moreover, in contexts where a wh-pronoun introduces an NRRC as an independent utterance, unaccented that is excluded (see (8b) in contrast to (8a)):

1. (8)

   1. (a) . . .I may now drive a Ferrari, but at heart, ‘my other car’ will always be that Vauxhall Cavalier. Which is as it should be. . . (J. May, ‘My other car's a Ferrari’, Radio Times, 9–15.08.14, p. 19)

   2. (b) *. . .That (/ðət/) is as it should be.

Let us look now at the situation where the relative pronoun is governed by a preposition:

1. (9) . . .Twistily plotted while also subtly knowing this [film] has nimble wisecracks with which couples everywhere can readily identify. . . (Film synopsis of Date Night, Radio Times 30.07–05.08.16, p. 33)

1. (10) I'm looking for a piece of paper on which to write my list.

When the preposition (where it is a short and common one: see Quirk Reference Quirk, Greenbaum, Leech and Svartvik1985: 664) is left stranded at the right edge of the clause, the RC introducer may be expressed as that or zero, occurring in situ. This realisation is characteristic, precisely, of the spoken language, but it also occurs even in careful writing:

1. (9)

   1. (a) . . . nimble wisecracks ø couples everywhere can readily identify with ®. . .

1. (10)

   1. (a) I'm looking for a piece of paper ø to write my list on ®.

As is well known, the RM that cannot ‘replace’ the relative pronoun when it is preceded by a preposition (see (9b) below), but it may occur when the preposition appears at the end of the clause: see (9c), also below. This is yet another indication that this form is a relative marker and not a fully fledged ‘relative pronoun’, as is often stated – since in these positions following a preposition, it is clear that an argument filler is needed.

Now, relative pronouns, which are indeed argument fillers, may occur here, but relative markers (purely grammatical elements, in fact subordinators) are excluded. In addition, in such a position, that assumes its usual function of subordinator, signalling the presence of a subordinate clause (see (9c) below):

1. (9)

   1. (b) *. . .nimble wisecracks with that couples everywhere can readily identify . . .

1. (9)

   1. (c) . . .nimble wisecracks that couples everywhere can identify with ®. . .

Again, this correlates with the possibility of ellipsis of that in (9a): where that is possible, the zero form is too (on condition that the RV is non-subject). But the converse is not true, as can be seen in (10a) as compared to (9a): whereas in (9a), that and ø are acceptable in the same position as introducers of the relative, in (10a), only the zero form is (cf. *I'm looking for a piece of paper that to write my list on). This is certainly due to the fact that that as a relative marker/subordinator is restricted to occurring in finite clauses (which the subordinate clause in (10a) is not). If it were a ‘relative pronoun’, however, it should be able to occur in this position. Purely in terms of respective distributions, note that this restriction does not apply to genuine (wh-marked) relative pronouns, at least in the case of the so-called ‘free’ relatives: He doesn't know [what to do]/[where to go]/[who to turn to], etc.

3.1 Restrictive relatives

First of all, as we will be seeing later, it is important to distinguish between presupposed and non-presupposed RRCs, a distinction orthogonal to that between definite and indefinite RC-containing NPs. Examples (11) and (12) illustrate the first subtype within the former distinction, and example (5b) the second – all three of these RC-containing NPs being indefinite. So although a restrictive relative may well serve to set up a narrower category type than the one denoted solely by the head noun which it modifies, nevertheless, its content is not necessarily presupposed by the speaker or writer. The ‘presupposed/non-presupposed’ opposition is limited to RRCs, ARRCs being by definition non-presupposed, and NRRCs by definition corresponding only to assertions (or other mood and illocutionary categories) at the utterance level. To begin, here is an example of a restrictive relative, where the illocutionary force of the containing NP in context was that of a question (pronounced with a rising tone):

1. (11) <x k4> any other fascinating drinks [that you tried ®] (Spoken extract from ‘The Great American Sandwich’ corpus)

As a modifier, the relative clause may, in principle, be elided without any consequence for the remainder of the sentence – on condition, from the discourse point of view, that the referent of the resulting NP (any other fascinating drinks in (11), an NP which arguably does not exist as such prior to the elision of the RRC – see below), is already grounded in the preceding context.

By contrast, Fox & Thompson (Reference Fox and Thompson1990: 298, fn. 4) claim a contrario that what is modified by a relative clause is in fact an ‘NP’ (see also Bache & Jakobsen Reference Bache and Jakobsen1980; Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985), and not the head noun of the embedding NP containing the relative. On the basis of the data yielded by their corpus of American English conversational utterances, these authors do not, however, distinguish between RRCs and NRRCs (this is also the case with Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan1999: 603).

In the approach adopted here, however, the sequence Determiner + Noun preceding the RRC is not in fact a constituent as such: an ‘NP’ should be substitutable by a pronoun, yet this is impossible in the context of a restrictive relative: The book/*It that I read yesterday. Furthermore, the article or other determiner which introduces the NP as a whole functions to establish in context the reference of the entire content of this NP (i.e. head N + relative clause), and is in a paradigmatic relation with every other member of the determiner set. The analysis we are claiming to be valid here, then, posits the structure of English integrated relatives as being [_NPthe [_N´ [_N´book] [_C´´that I read ® yesterday]]], where the RC forms a head-modifier constituent with the head noun of the expanded, containing NP, and not [_NP[_NPthe book] [_C´´that I read ® yesterday]] (see also Loock Reference Loock2010: 10 and section 4.1 below).Footnote ¹⁵ Under this analysis, the definite article the determines the entire head noun + RRC complex, and not simply the N book. The same is certainly true of other NPs containing modifiers (e.g. APs or PPs: respectively, the [book [bound in gold braid]]/. . .[about philosophy]]). See also as evidence the ungrammaticality created by attempting to substitute an indefinite pronoun for the ‘NP’ preceding the relative clause in (12):Footnote ¹⁶

1. (12) . . .two films about art [that ® is produced by those with no formal training]. (Film synopsis of Outsider Art, Radio Times 11–17.08.01, p. 85)

Replacing the ‘introducing NP’ (but see fn. 16 for an alternative ‘substitutum’) two films about art here by ones would not result in a grammatical matrix NP: *Ones that is produced by those with no formal training. For in (12), the singular value of the finite passive auxiliary verb is in the relative clause relates to the ‘host’ of this integrated relative, namely the head noun art (sg.), and not to the ‘NP’ (at least, in the context of this restrictive relative clause) two films about art (pl.). This formal-syntactic relation reflects the fact that the relative clause in (12) functions to set up a restriction upon the denotation of the head noun art, thereby creating a subcategory of ‘art’. Namely, art produced by artists without any formal training in the field.

3.2 ‘A-restrictive’ relatives

Let us look now specifically at ‘a-restrictive’ relative clauses (ARRCs). Below are several attested illustrations:

1. (13) Philip LeRiche is a retired software developer and information security adviser and Ben Skidmore is an electronic engineering student. Both volunteer with The Restart Project, a London-based social enterprise that encourages and empowers people to keep their electronic goods for longer. . . (The Guardian Online, 16.06.14).

Here, the NP-integrated relative clause that encourages and empowers people to keep their electronic goods for longer conveys discourse- and hearer-new information for the reader at the point of occurrence. It does not restrict the reference of the NP that contains it, thereby helping the reader to identify the referent at issue. No contrast is implied with other ‘London-based social enterprises’, which would be the case if the relative clause were restrictive (see sections 3.1 above and 4.1 below). Indeed, the indefinite NP that contains it is in apposition to the proper noun which precedes (The Restart Project), and as such serves to inform the reader about this venture. Notice that no radical change in terms of the referent targeted by the residue of the omission (in effect, forming a simpler indefinite NP) would be caused if the ARRC were elided. This state of affairs is in direct contrast with the situation obtaining with restrictive relatives. Note, however, as we have already seen in section 3.1, that RRCs may well also occur within indefinite NPs. As such, they often serve to characterise a type of entity.

(14) and (15) are further attested indefinite-NP-containing examples of ARRCs:

1. (14) It's inspired by the most successful rally car of the 90s, so expect a sports saloon that exhilarates. . . (Advertisement for Subaru cars, Radio Times 7–13.08.99, p. 11)

1. (15) . . .[(actor) Steve] Martin displays a comic dexterity which since then he seems to have lost or perhaps put aside. . . . (Barry Norman, ‘Last of the clowns. . .’, Radio Times 7–13.08.99, p. 42)

In (14) and (15), the governing verbs expect and displays imply that the states of affairs evoked by the complex NPs which they govern are either one which is predicted to materialise in the future (from the intended buyer's point of view in (14)), or one that is claimed to hold, as in (15). These indefinite NPs therefore constitute discourse-new, focal information. (16a), however, is an ARRC contained within a syntactically definite NP:

1. (16)

   1. (a) . . .DOCUMENTARY Chris Rock's amusing yet piquant look at the billion-dollar industry that thrives on the care and treatment of black hair weaves a worthy tale. . . (Film synopsis of Good Hair, Radio Times 30.07-05.08.16, p. 38)

   2. (b) Bobby Charlton tells the story of the Munich air disaster. He charts a journey that ended in the tragic events of 1958. (Film synopsis, Yesterday TV channel, Radio Times 30.07–05.08.16, p. 49)

In (16a), the nature of the industry examined in the documentary at issue is in fact discourse-new (and also reader-new) information: as such, the NP which expresses it would be indefinite in the absence of the relative (in fact, the definite article introducing the original NP could be replaced by the indefinite one). So this is an example where, by hypothesis, it is the addition of the relative clause itself which has caused the NP as a whole to be syntactically definite. After all, eliding the RC here implies that the referent of the residue is discourse-new (indeed, there would be no justification for a definite (i.e. functionally anaphoric) RC-less NP at this point in the brief synopsis, namely, the very first sentence). The referent in each case is thus essentially identical. The ‘lie (contradiction) test’ (see section 3.3) confirms this status, since it yields a positive result when the content of the RC is contradicted: That's not true at all, it doesn't!. We may say that the RC in (16a) is at the same time speaker-presupposed (since the journalist has manifestly already seen the documentary in question) as well as discourse- and reader-new.

As for (16b), this represents the opposite configuration to (16a): for here, the RC-contained NP is indefinite, rather than definite (as would befit the highly presupposed character of the RRC involved here: indeed, the definite article in place of the indefinite one would be highly preferred). The ‘heralding’ NP the Munich air disaster and the evocation of the famous former Manchester United and England footballer ‘Bobby Charlton’ in the initial sentence, as well as the also definite NP the tragic events of 1958 within the RC itself, serve to trigger the memory of this well-known air catastrophe. The RC thus represents highly presupposed specific knowledge on the reader's part – and yet the containing NP is indefinite.

Malan (Reference Malan1999: 5–6) also notices the distinction between what we are calling restrictive and a-restrictive integrated relatives. However, her solution (1999: 6) is to abandon the term ‘restrictive’, since clearly it is only a ‘subset’ of this category which really serves to ‘restrict’ the reference of the NP which contains the relative clause. She proposes instead the cover term ‘relative déterminative’. Yet this does not capture the specificity of what we are calling ‘ARRCs’ (she calls these ‘fausses déterminatives’, which is not a very helpful term, ARRCs not being ‘déterminatives’ – i.e. ‘restrictive’ – at all).

Depraetere (Reference Depraetere1996: 718–20), for her part, tacitly recognises the distinction between ARRCs and RRCs, though she continues to use the cover term ‘RRCs’ to categorise both subtypes. According to the author, ARRCs (in our terms) occurring within indefinite NPs serve to signal discursively foregrounded information, whereas relatives embedded within definite ones only convey backgrounded content.Footnote ¹⁷ However, our example (16a) is clearly a counterexample to this latter claim. Rydén (Reference Rydén1974: 544–5) similarly distinguishes a ‘descriptive or characterizational’ (p. 544) RC type, though he is equivocal about its distinctiveness with regard to NRRCs, and does not give it a particularising label.

3.3 Distinguishing between a-restrictive and restrictive relatives

Three tests, amongst a number of others noted in the relevant literature, make it possible to recognise (a-)restrictive relative clauses. The first of these corresponds to an interpretation, or initial analysis, of the clause in terms of the type of semantic contribution being made by the RC; and the next two seek either to confirm this initial understanding (test 2) or to provide evidence that it is presupposed (confirming its RRC status) or not (indicating that it may either be a non-presupposed RRC or an ARRC: test 3).

Test 1: The nature of the semantic contribution made by the subordinate clause as a whole. If it is a restrictive relative, then it serves to determine more precisely which ‘x’ is involved in the reference of the NP as a whole. A-restrictive relatives lack this property (see (17b)):

1. (17)

   1. (a) The book [that was ordered yesterday] is out of print. (= identificational, reference-restricting)

   2. (b) You bought a book [that is being considered for the Pulitzer prize] yesterday. (= nonidentificational, non reference-restricting; merely elaborating in function: implied evaluation by the speaker of the book in question)

Test 2: If we elide the relative clause as a whole, on condition that the unmodified head noun of the residual NP is not a general or highly abstract one, then if it is a restrictive relative (as in (18a), modified from (17a)), the resulting NP's referent will be a different one. However, if it is an a-restrictive relative, then the referent will be essentially the same, whether the relative is present or not, as in (18b):

1. (18)

   1. (a) #Footnote ¹⁸ The book is out of print. (referent of NP the book is contextually unanchored; probably distinct from that of the subject NP in (17a))

   2. (b) You bought a book yesterday. (referent of NP a book is newly introduced; it could well correspond to the book evoked in an utterance of (17b))

This test often yields more perspicuous results when the containing NP is indefinite: viz., I wanted a job vs I wanted a job where I used my brain but also helped people, as in (5b) above. Here, the referents of the object NPs are clearly distinct (that of the latter being a more specific subtype of the former). Test 2 indicates whether a RC is restrictive or otherwise, not whether its content is presupposed by the speaker or not. This is the purpose of test 3. In the case of (5b) and (19) (below), for example, the subclauses are not presupposed, even though they are clearly RRCs:

1. (19) . . .Here's my pick of the players who'll make this one of the most exciting seasons yet. (Alan Hansen, ‘The players to watch. . .’. Radio Times 7–13.08.99, p. 26)

Test 3, the so-called ‘lie-test’ (see Erteschik-Shir Reference Erteschik-Shir2007: 39, 164): This is really a test of what is in focus in a given utterance. If the content of a given segment can be felicitously contradicted, then it is indeed in focus, discourse-pragmatically; but if it cannot, then it is part of the presupposition of the utterance concerned.Footnote ¹⁹ The test should be construed in relative, not absolute terms: the relative degree of ease with which the contradiction may be realised reflects the degree of ‘foregroundedness’ of the information unit at issue. Let us apply this test to both the matrix and the relative clauses presented earlier in this subsection (see (20) for RRCs, and (21) for ARRCs). The relevant contrast is that between (20a.B) and (21a.B):

1. (20)

   1. (a) A, to B: The book [that was ordered yesterday] is out of print.

      B: #That's not true at all, it wasn't! (Contradiction of the content of the RC)

   2. (b) A, to B: The book [that was ordered yesterday] is out of print.

      B: That's not true at all, it isn't! (Contradiction of the matrix proposition)

1. (21)

   1. (a) A, to B: You bought a book [that is being considered for the Pulitzer prize] yesterday.

      B: That's not true at all, it isn't! (Contradiction of the content of the RC)

   2. (b) A, to B: You bought a book [that is being considered for the Pulitzer prize] yesterday.

      B: That's not true at all, I didn't! (Contradiction of the matrix proposition)

Now let us apply these three tests to the attested examples that follow:

1. (22) . . .[Will it be] perhaps a genuine novelty such as Sally, Irene and Mary (Saturday), a 1938 Alice Faye musical [that apparently has never before been shown on British TV]?. . . (‘Quentin Cooper's guide through the multichannel maze’, Radio Times 7–13.08.99, p. 39)

1. (23) When you sell a home in America you have to pretend you have never lived there. The inspector's report on the house [ø we sold in Washington] mentioned all manner of minor imperfections. . . (Justin Webb, ‘Viewpoint’, Radio Times 02–08.08.14, p. 9)

Applying test 1 to (22), it is clear that the content of the italicised subordinate clause serves to convey the writer's noting of a feature associated with the musical in question (notice the presence of the style disjunct apparently), namely the fact that this showing would constitute a première on British TV. There is no concern to distinguish it from other comparable musicals in order to identify it. As for test 2, eliding the RC would not essentially change the referent at issue. Test 3 also yields a positive result: That's not true at all, it has!, clearly indicating that the content of the RC is part of the focus, but not the presupposition of the containing NP. So it seems we are dealing with an ARRC here.

As for example (23), test 1 shows that the subordinate clause places a limit or restriction upon the denotation of the noun house: it is a question of the particular ‘house’ which was situated in Washington and which the writer sold prior to the time of utterance. This is clearly characteristic of a RRC. Test 2 confirms this status: for if the relative clause is elided (The inspector's report on the house mentioned all manner of minor imperfections), then it is not clear at all which ‘house’ is being evoked: indeed, there is no way that it would specifically be the writer's house that is at issue. So the clause would seem to be restrictive, and not a-restrictive in this respect. Test 3 indicates that the content of the clause at issue is part of the presupposition of the matrix subject NP, and does not constitute focused information (#That's not true at all, you didn't! versus That's not true at all, it didn't!).

In discourse terms, the content of the ARRC in (22) represents discourse- and reader-new information (see also Depraetere Reference Depraetere1996: 718); but in (23), the existence of a particular ‘house’ formerly owned by Justin Webb in Washington is presented as constituting backgrounded discourse informationFootnote ²⁰ in the case of the RRC.

Table 1 summarises the criteria distinguishing RRCs and ARRCs in English.

Table 1. Tests for ‘a-restrictive’ vs restrictive relative clauses

4.1 Integrated vs non-integrated relative clauses

Integrated RCs (specifically RRCs) may serve to create a type of entity, often as part of an indefinite containing NP. Closely connected with this function is their role in the formulation of definitions (whether in dictionaries or encyclopaedias). They may also help to set up a bound-variable construction, and can function as an extended predicate within an NP (either demonstrative or definite) that acts to direct the recipient towards reactivating an assumed shared item of knowledge buried in the speech participants’ long-term memory. NRRCs, however, are not capable of or suited for achieving any of these discourse goals. Let us look at each of these functions in turn.

Since NP-integrated relatives bear upon a particular (common) noun (a form type which corresponds semantically to an intension and only denotes, but does not refer), and not upon an NP, a type of expression which may well be used to refer (to an entity, a set of entities or to a genus – i.e. a generic reference), the nature of the reference of the containing NP is irrelevant:

1. (24)

   1. (a) All through school I knew that I wanted a job where I used my brain but also helped people. . . (Interview with Jo Hardy, Radio Times 16–22.08.14, p. 21: example (5b) above)

In (24a), the reference of the global indefinite NP whose head is job is non-specific: there is no particular ‘job’ identified via the use of the expression a job where I used my brain but also helped people. The entity evoked constitutes a type, not a token (note the semantic effect of the governing modalising, ‘world-creating’ verb want here). The restrictive relative clause serves precisely to specify the type at issue here. As we have seen, this syntactic dependency upon a head noun in the case of restrictive relatives has a semantic correlate, in that the content of the relative clause serves to extend the sense and, in doing so, to narrow down the denotation of the head noun upon which it is constructed. This discourse function could not be realised via an equivalent NRRC:

1. (24)

   1. (b) !Footnote ²² All through school I knew that I wanted a job, where I used my brain but also helped people. . . .

In (24b), since the now ‘orphaned’ indefinite NP is non-referential, the purported NRRC has no ‘anchor’ to support it. The quantifier corresponding to the indefinite article needs to have scope over a constituent corresponding to a coherent semantic unit – i.e. more than simply the denotation of the head noun by itself, as would be the case for the ‘NP’ consisting of indefinite article + noun in the semantically infelicitous appositional clause in (24b). On the other hand, in the original example (24a) such a constituent is available in the shape of the head noun together with its modifying relative providing the relevant property for its denotatum. In more general terms, quantified or otherwise non-specific, non-referring expressions are not acceptable as hosts for potential NRRCs (see Del Gobbo 2003; Loock Reference Loock2010: 20; Emonds Reference Emonds1979: 235–6), unless they can be interpreted as generic in function.Footnote ²³

(25) is an illustration of the function of (non-presupposed) RRCs in formulating dictionary definitions of nouns:

1. (25) Drayman. 1581. A man who drives a (brewer's) dray. (The Shorter Oxford Dictionary on Historical Principles (3rd edition). Oxford: Clarendon Press, 1973, p. 606)

Here, the RRC is an essential part of the definiens of the defined noun drayman. NRRCs, by contrast, tend to be used in dictionary definitions for adding non-essential details about the meaning, or about the type of world entity which the noun in question denotes.

Now, inasmuch as they are modifiers, relative clauses are in principle omissible. However, as we have already seen in section 3.3, an NP containing a RRC and the same NP without one are not equivalent, whether in semantic or in pragmatic terms. For a RRC post-modifying a noun serves above all to identify the referent targeted, by providing a property enabling the addressee or the reader to differentiate it with respect to other referents with which it might be confused:

1. (26)

   1. (a) <x c23> . . . and in the middle of the night the longest train ø i have ever heard in my life went past the bedroom . . .’ (Spoken extract from ‘The Great American Sandwich’ corpus)

   2. (b) and in the middle of the night the longest train went past the bedroom.

Again, as in (24a), in (26a) the intensional element corresponding to the superlative component of the adjective longest requires a coherent semantic constituent over which it can have scope, forming an expanded property (viz. ‘longest X the speaker had ever heard in her life’). This property is predicated of a restricted subset of long trains, namely ‘the particular one which the speaker heard at the time being narrated’; whereas, by default, we understand (26b) as asserting that the longest train in the world, objectively speaking, went past speaker <x c23>’s bedroom at the time being reported. This is potentially a very different entity. As in the case of (24b) as compared with (24a), an equivalent NRRC could not perform the semantic function fulfilled by the RRC in (26a).

As in (26a), when the NP in question contains a logical element, then an RRC is virtually obligatory for the sentence to make sense:

1. (27)

   1. (a) <x c19> . . . it was the ONE place where i had to stay in a hotel (Spoken extract from ‘The Great American Sandwich’ corpus)

   2. (b) !It was the ONE place, where i had to stay in a hotel.

Here the restrictive phrase one (+ pitch accent) X (followed by relative clause), where the numeral one, which when preceded by the definite article, is equivalent to the restrictive adverb only – i.e. ‘the one and only. . .’ –, imposes a logical scope over the whole of the noun group place where i had to stay in a hotel. And it is this that makes the presence of the RRC obligatory in (27a). Note that the restriction here stems both from the presence of the definite article and the fact that one is accented (compare with . . . one place where i had to stay in a hotel).

By contrast, in the absence of this integrated relative (see (27b)), the accented numeral one can only have scope over the sole noun head place, a situation which would result in an absurd proposition since this noun has an extremely broad, nonspecific denotatum (see also Bache & Jakobsen Reference Bache and Jakobsen1980: 260).

Administrative prescriptions defining a category of individuals concerned by some specification or condition also require integrated RCs (in (28a) below, an RRC) in order to achieve this function of delimiting a class of entities denoted by the nouns upon which they operate (see section 2.1 above).

1. (28)

   1. (a) Those students who have obtained 26 points at Part I may register for their chosen degree. (Translation from French of a notice displayed in the Language Sciences department, Université de Toulouse II)

Here, the restrictive relative (in italics) delimits a potential subset of students: namely, those who have obtained 26 points in Part I. It is not a question of all students, whatever their situation, being said to be authorised to register for their first degree, but only those who have obtained 26 points in Part I of their undergraduate programme. Compare this with (28b), where the RC has NRRC status:

1. (28)

   1. (b) Those students, who have obtained 26 points at Part I, may register for their chosen degree.

Though this is also a possible utterance, it does not mean the same as example (28a) at all. For in (28b), all the students in question (i.e. ‘those particular students’, a group presupposed to have been identified either independently or deictically, via the situation of utterance) are asserted as being permitted to register for their chosen degree, and not only those who have obtained 26 points in Part I (though assertion is not the only type of illocutionary force a NRRC may assume). There may well be other students at the university in question who have obtained 26 points in Part I, but who are not included in the group of students designated by the subject of the main clause. Note also that in (28b), the demonstrative determiner those assumes its usual deictic-referential function. Whereas in (28a), this is clearly not the case. Here the demonstrative may be replaced by a quantifier such as all or any, a substitution which would be impossible salva veritate in the case of (28b). This difference in referential value is a direct function of the differing RC subtypes at issue in this pair of examples.

Finally, (29) is an example of the use of RRCs to enable the addressee/reader to locate and extract an item of shared, stored information from their long-term memory:

1. (29) Pyle had a flat in a new villa near the rue Duranton, off one of those main streets which the French continually subdivided in honour of their generals. . . (G. Greene, The Quiet American, London: Vintage Books, 2004, p. 18)

As we have seen in sections 3.1 and 3.3, the proposition expressed via an RRC may or may not be presupposed by the speaker as true of the referent of the whole NP containing it;Footnote ²⁴ however, it usually conveys only (relatively) background information, unlike ARRCs or one subtype of NRRCs (‘Continuative’ NRRCs: see subsection 4.2): see figure 1 in section 4.2 below for an overall representation of the positions adopted by the various subtypes of RC described here on the background-foreground axis, at the level of discourse.

Figure 1. Placement of each of the six subtypes of relative clauses examined along the ‘Background–Foreground’ axis at the level of discourse

RRCs are crucial for the containing NP's reference, unlike non-restrictive (or ‘supplementary’, in Huddleston & Pullum's Reference Huddleston and Pullum2002 terminology) relatives, to which we now turn.

4.2 Non-restrictive (appositive) relatives

By contrast, non-restrictive relative clauses, which are precisely, by that token, appositive in function, to some extent function like parentheticals. That is, they break up the flow of the discourse on some referent in order to elaborate or comment on it in some way. See in particular Loock's (Reference Loock2007, Reference Loock2010) subtype of appositive relatives which he calls ‘Subjective’, where the NRRC evokes an aside, a comment or an evaluation of the host referent or the state of affairs evoked via the main clause. An example:

1. (30) A routine Randolph Scott western, uninspiringly directed by Gun Crazy’s Joseph H. Lewis who fails to secure a satisfactory conclusion to the tale, which results in a particularly soppy ending. . . (Film synopsis of A Lawless Street, Radio Times 7–13.08.99, p. 49)

Here, the (propositional) NRRC which results in a particularly soppy ending constitutes a subjective evaluation on the journalist's part of the lack of a satisfactory conclusion to the story in the film at issue. Note in particular the subjective adjective soppy used to express the critic's point of view on the film at issue here. A test for Subjective status could be the possibility of inserting, without altering the intended interpretation, one of the discourse markers in my view or to my mind after the relative pronoun in (30).

The other two subtypes of NRRC recognised by Loock are ‘Continuative’ and ‘Relevance’. Loock's subtype ‘Continuative’ involves the NRRC in post-matrix clause position evoking a state, process or event which comes into being after the one evoked by the preceding main clause. As in the case of the Subjective status of the NRRC in (30), a test for Continuative status could be the possibility of felicitously inserting the time adverb later between the passive auxiliary was and forced in (31), for example:

1. (31) I used to help out with a local community group, which I was forced to give up due to all the admin. (The Guardian Online, 21.01.14)

This is a ‘Continuative’ NRRC whose host is a second-order referent (the eventuality ‘the writer's helping out with a local community group at the time designated’). Denison & Hundt (Reference Denison and Hundt2013: 144–5ff.) use the same term, though they do not (apparently) restrict the subtype to matrix-final clauses which advance the narrative being developed – and hence constitute not only new but also discursively foregrounded information.

Another such example is (32), where it seems that a ‘Relevance’ relation (see below) is involved simultaneously:

1. (32) Donna Mills plays the battered wife of nasty Corbin Bernson, who summons up the courage to leave him only to discover that he has no intention of allowing her to break up the family. . . (Film synopsis of Dangerous Intentions, Radio Times 7–13.08.99, p. 48)

Note here that it is not the referent ‘Donna Mills’, the actress, that is anaphorised by the relative pronoun who in this NRRC, but the character she plays in the film (the battered wife of nasty Corbin Bernson). The NRRC evokes a continuation of the plot description initiated by the first clause, and also enables the reader to infer that this event is a consequence of the character in question's having been ‘battered’ by her husband: hence there is also a relation of ‘Relevance’ (specifically, ‘Cause-Consequence’ here) holding between the two propositions: see below on this subtype. To test for these statuses, the markers as a result (for ‘Relevance’) and subsequently (for ‘Continuative’) could be naturally inserted (in that order) after the relative pronoun who in (32). Notice that, where this occurs, one of the discourse-pragmatic statuses is dominant, the most dominant one of all characterising the status of the NRRC as a whole (see figure 2 below). In (32), this is clearly the status ‘Continuative’, ‘Relevance’ being subservient to it. The subtypes of NRRC recognised by Loock are thus not mutually exclusive.

Figure 2. Scale of relative dominance of simultaneous discourse statuses that may be manifested in context by a given NRRC

The NRRC subtype named ‘Relevance’, as the term implies, enables the recipient to grasp the relevance of the situation evoked via the main clause, as illustrated by (33):

1. (33) The first recorded eclipse prediction may have been made by Greek philosopher Thales, who is alleged to have forecast totality for an eclipse in 585 BC. . . (‘Eclipses throughout history’, Radio Times 07–13.08.99, p. 25)

Here, the content of the NRRC provides the (tentative) evidence for the (again, tentative) claim made by the journalist in the initial, independent clause. To test this, the marker indeed could be felicitously introduced after the passive auxiliary is within the RC. Note also that the states of affairs evoked via the matrix and relative clauses are identical, so the NRRC here cannot (also) be of the ‘Continuative’ subtype.

See Loock (Reference Loock2010: 95–139) for a detailed discussion of the various discourse functions of the subtypes of NRRCs he recognises. Of the examples of NRRCs presented earlier, the ones in (6), (in part) (32) and (33) would be of the ‘Relevance’ subtype, those in (2), (7), (8a) and (30) would be ‘Subjective’ ones, and (31) and (32) ‘Continuative’. As for the remaining example (28b), none of the three subtypes recognised in Loock's typology seems to be relevant, the NRRC simply evoking an additional, elaborative piece of extra information about the referent of the host expression.

In no sense do NRRCs serve to restrict the denotation of their host expression, as we have seen (section 4.1 in particular). Unlike NP-integrated relatives, with NRRCs, the clause does not form a constituent with its host, which, contrary to integrated relatives, is a phrasal (NP, PP, AP) or clausal, not a lexical category (N). In such instances, the reference of this phrasal or clausal category is assumed already to be established independently of the content of the relative itself, as we have seen.

As regards the often-mentioned ‘suppressibility’ property of NRRCs,Footnote ²⁵ Loock's (Reference Loock2007, Reference Loock2010) corpus-based surveys suggest a significant qualification. For of his three subtypes, Loock claims these surveys show that only the first (‘Subjectives’) may be omitted in context without consequences for the remainder of the surrounding discourse, the third (‘Relevance’ NRRCs) being omissible only under certain conditions. However, in general, it appears that the presence or absence of the NRRC does not differentially affect the identification of the host referent to which it relates.

In sum, then, we may say that Continuative NRRCs serve to update the discourse context already created (in part, at least) by the preceding independent clause, while Subjective NRRCs simply act ‘metadiscursively’ to manage and ‘punctuate’ the discourse created via the preceding or surrounding co-text. In this they are akin to parentheticals (see Dehé Reference Dehé2014 in relation to spoken English discourse). Relevance NRRCs, for their part, serve to facilitate, by contextualising it, the interpretation by the addressee or reader of the matrix clause within which, or following which, they occur.

Moreover, according to the appendix given in Loock (Reference Loock2007: 361), virtually all of the three subtypes of NRRCs recognised in the corpus conveyed discourse-new information (respectively, 100%, 98.9% and 100%). However, with the exception of Continuatives, this information, though discourse-new, is (relatively) backgrounded in relation to the content of the main clause; see figure 1 below. Figure 1 ranges the six subtypes of RC examined in this article along an axis stretching from ‘Background’ to ‘Foreground’, through ‘Midground’, at the level of discourse.

Table 2 summarises the differential properties of the three basic subtypes of relative clause. The table shows clearly the bi-directional relationship between the central category of ARRCs and the other two categories flanking it to its left (RRCs) and right (NRRCs). There are four properties which they hold in common with the former of the flanking subtypes, and only two with the latter. This indicates that ARRCs tend to pattern as an NP-integrated rather than clause-external type of unit – which contradicts what both Malan (Reference Malan1999) and Denison & Hundt (Reference Denison and Hundt2013) claim.

Table 2. Criteria for restrictive, a-restrictive and non-restrictive relative clauses

Now, as pointed out earlier (see the analysis of example (32) above), a given NRRC may manifest in context more than one discourse status at the same time. When this occurs, there is always a relative dominance relation holding between or amongst the statuses concerned. This may be captured via the following scale. Starting from the right-hand pole in figure 1, this scale is reflected in the reverse relative order of the positions occupied by these subtypes of NRRC in the former scale; see figure 2.