Quotation does not need marks of quotation

2.1 A note on hybrid quotation

Because the label is not widely used, hybrid quotation (henceforth HQ) deserves a quick introduction. The label is derived from Recanati’s terminology and is intended to encompass ‘mixed quotation’ (e.g., Cappelen and Lepore 1997) and ‘scare quoting’ (e.g., McCullagh 2017) but also a range of instances that do not readily fall under either of these categories (see De Brabanter 2017a: 238).

The best way to make plain the difference between hybrid quotation and embedded quotation is to focus on grammar. Let us start with embedded cases. The typical instances of DD in (6) and PQ in (7) have been recruited as NPs in clause structure. They are replaceable by a non-quotational NP like, e.g., that.

The internal syntactic structure of the quotations is segregated from their surroundings: never mind that the quoted string in (6) is a sequence of an AdjP and an NP. The complement of said is not generated via a phrase-structure rule like VP ⇒ V AdjP NP. Remember that what has been recruited is the communicative act, not a string of words. Semantically, the quotation refers to Kim’s speech act, not to some object that is quite red. Likewise with (7): syntactically, it is not generated by a rule that states that PPs can be subjects; semantically, it is about a certain grammatical structure, not a time interval. Note further that strings instantiating just any syntactic structure (or no structure at all, e.g., She the with me six) can occur as recruited quotations, without altering the grammaticality of the host sentence, witness (8) and (9):

Just the opposite can be observed with HQs, as they involve strings of words whose internal structure is part and parcel of the host structure. Consider (5) again, in which the nominal^[11] billion Muslims who reject their ideology is quoted at the same time as it functions normally as the head of the NP the billion Muslims who reject their ideology: you could not substitute an NP or AdvP or VP or finite clause for that nominal without making the host structure ungrammatical. Similarly with (10): nothing else than a ditransitive verb could fit into the boldfaced slot.

This different grammatical behavior of HQs is a direct consequence of their not being recruited syntactically. Another consequence is that removing the quotational dimension does not result in unacceptability, and that the resulting sentence expresses a proposition that is meant by the speaker. That is because in HQ a quotation is superimposed on a structure that is already viable without.^[12] Compare this to what normally happens with recruited quotation: remove the quotational dimension from (6)–(7), and even more strikingly (8)–(9), and you end up with unacceptable strings of words.

3.1 Pure and direct quotation

I start with cases where it is uncontroversial that we are dealing with a quoted string of words, namely PQ and DD. In those, quotations are purely metalinguistic or metadiscursive. i.e., they refer to linguistic material, be it abstract lexemes or concrete utterances.

First, let me point out that the absence of marking is not carelessness. It is encountered not just in informal genres, such as text messages:

It is also widely observed in carefully edited writing. (12)–(13) and (14)–(15) are examples of DD and PQ, respectively, that occurred in fiction.^[14]

In (12), being a boy’s name denotes a property that can only be ascribed to a linguistic entity, not a person. Similarly with the other boldfaced words. In (13), the non-embedded Lower level is introduced as a ‘new term’; it must therefore stand for a linguistic entity, not a floor in a building. In (14), the complement of said is an unmarked embedded DD report. Use of the imperatives see and look and of the present tense demonstrates that it cannot be an indirect discourse (ID) report. The boldfaced strings in (15) are instances of non-embedded DD. They ‘voice’ a character different from the narrator. Note that the author cannot be suspected of quote-mark avoidance since he marks both the embedded DD that follows and the PQ contained in it.

Quotations in all relevant respects similar to the unmarked quotations in (12)–(15) can also occur with quotation marks, without a meaning difference. I take this to be uncontroversial with respect to DD (cf. the embedded DD that occurs in (14) and in (15)). I also take it that quote-marking lower level is unproblematic and would not affect its interpretation in (13). What of the quotations that occur as predicative complements of call in (12)? Corpus evidence shows that those complements tend not to be quote marked, notably when the quoted material is a proper noun. Still, there exist straightforward quote-marked occurrences of names as predicative complements of call, strongly hinting that the absence-presence of marks makes no semantic difference.^[15] Consider (16) and, strikingly, (17):

Example (17) is particularly illuminating. Both occurrences of calling are semantically indistinguishable, yet one is marked while the other is not. (Beta is both quote-marked and italicized in the original example.) It may be that the journalist made the pragmatic decision to (double-)mark it because a foreign word was quoted.

3.2 Hybrid quotation

So far I have shown that PQ and DD can occur with or without marks of quotation with no difference in meaning. But can HQ also occur unmarked? Again, most accounts of HQ take for granted that the marks are necessary (Benbaji 2005; García-Carpintero 2005; Geurts and Maier 2005; Maier 2014a). The data initially seem compelling:

Neither in (18) nor in previous examples like (5) or (10) is it likely that a reader would detect the quotations in the absence of any marking. It is therefore legitimate to ask if certain types of written quotation are necessarily marked.

I argued in Section 2 that there are good grounds for regarding quotation as one and the same phenomenon in PQ, DD or HQ. That view is threatened if unmarked HQs do not exist. Below, I adduce two types of evidence that they do. The first consists of allusions to well-known sayings in English (see De Brabanter 2013: 140–141), the second of unmarked ‘intrusions’ of a reported speaker’s perspective, typically but not exclusively into ID reports.

In (19) and (20), the boldfaced strings echo a poem and a saying, respectively. Those are ‘others’ words’, to use Benbaji’s (2005) apt phrase, and are expected to be recognized as such by the reader.

The boldfaced NP in (19) refers to a group of Haitians who tried to enter U.S. territory clandestinely. It is also intended to be recognized as an echo of the Emma Lazarus poem inscribed on the base of the Statue of Liberty.^[16] In (20), even without any signaling, the proverb beauty is in the eye of the beholder is sufficiently well-known to be widely identified as an echoic HQ.

Some usage manuals actually advocate not marking echoic allusions. Thus, in a discussion of indirect discourse, Strunk and White (2000: 37) write:

Proverbial expressions and familiar phrases of literary origin require no quotation marks.

These are the times that try men’s souls.

He lives far from the madding crowd.

Target readers of these examples would be expected to recognize the first example as a Thomas Paine quotation. The second quotes the title of Thomas Hardy’s novel Far from the madding crowd. Both quotations are hybrid in the sense that their meaning-without-quotation is meant by their putative utterer.^[17]

The second kind of evidence relevant here is found typically in those frequent instances of ID that stray from the canonical template. Canonically, ID presents an utterance or thought entirely from the reporter’s perspective (Evans 2013: 68). The examples below contain words or phrases that are intended to be understood as being the source speaker’s, not the narrator’s (here, narrator = reporter). We should therefore analyze them as quotational superimpositions on ID, one of the forms that HQ can take.

In (21), the taboo phrases are not the narrator’s, as is unmistakably noticed by the reader, who is by now familiar with the narrator. They are demonstrations of Toner’s words, designed to provide indirect information on that character. A similar analysis applies to (22).

Comparable examples can occur even outside ID. In the following – taken from online readers’ comments on a Guardian article (https://www.theguardian.com/commentisfree/2016/mar/29/grammar-pedant-personality-type) which used the spelling extravert instead of the more frequent extrovert – reader Edgeley pokes fun at the grammar pedants who complain about the unusual spelling, indirectly reminding them that the term was originally coined with an a not an o:

The word extrovagantly is both used to mean what extravagantly normally means and ‘displayed for depictive purposes’: the quoter conspicuously uses the spelling with o as an echo to the grammar pedants’ preferred spelling of extrovert. The point of the demonstration is to gently mock the pedants.

3.3 Intermediate conclusion

Section 3 has shown that all uses of quotation (PQ, DD, HQ) can occur with or without marks of quotation. It must be conceded that there is a preference for not omitting marks in HQ. A tentative explanation is that a non-recruited quotation is less conspicuous than a recruited one. Hence, it would be unwarranted, when the words quoted are not obviously not the writer’s, to expect a reader to recognize a depictive intention behind those words without the help of any marking.

4.1 Systematic prosodic counterpart to written marks?

There have been two main approaches to this question: conversation-analytic (CA) and, for lack of a better word, acoustic. Since the vast majority of the relevant studies focused exclusively on DD, this section does not address PQ or HQ.

The CA studies focus mainly on the distinctions between DD and ID, and between DD and narrative. The findings are that variable prosodic means are used to signal quotation (Bolden 2004; Günthner 1999; Sams 2010). The following is a list of signals identified in various studies:

1. breathy voice;

2. high pitch;

3. slower delivery;

4. faster delivery;

5. combination of breathy voice and low pitch;

6. combination of low pitch and slower delivery;

7. combination of breathy voice and slower delivery;

8. combination of higher pitch and louder volume;

9. combination of creaky voice and lower pitch.

Next to this array of signals, several core functions of prosodic marking are identified too, not all of them the kinds of meanings that supporters of the Necessity Claim ascribe to marks of quotation:

1. demarcating DD from surrounding narrative (Bolden 2004; Günthner 1999);

2. animating different characters and ‘tracking’ them (Günthner 1999; Klewitz and Couper-Kuhlen 1999);

3. depicting the mental states of a character (Günthner 1999; Sams 2010).

Based on the rich descriptive material collected, the CA studies conclude that prosodic marking is not systematic (Günthner 1999; Klewitz and Couper-Kuhlen 1999; for the offset of reported speech see Bolden 2004; also Biber et al. 1999: 1118–1119).

In sum, although stretches of conversational reported speech are frequently marked prosodically, it would be an overstatement to claim that prosodic marking is used systematically as a sign of reported speech in talk the way quotation marks are in texts. For one, not only direct but also ‘indirect’ reported speech is often prosodically marked. Moreover, oral quotations may be cued as such by virtue of being prosodically unmarked if they are embedded in a prosodically marked environment. And more importantly, quotations in speech may not receive any kind of special prosodic formatting at all. (Klewitz and Couper-Kuhlen 1999: 473)

As for the acoustic analyses, most are conducted within an NLP framework. A general observation about meanings that are marked prosodically is that speakers are not constrained to use prosodic marking. When they do, several means appear to be available and used (Collier et al. 1993; Hirschberg 1999). Different studies identify different such means. In some, pitch range and/or pitch reset are singled out as significantly marking DD boundaries (Hanote 2015; Jansen et al. 2001; Oliveira and Cunha 2004). Some studies also highlight intonational phrase breaks as marking boundaries (Hanote 2015; Jansen et al. 2001). Elsewhere, Bertrand and Espesser (2002) also propose that Pitch (F0) significantly marks DD containing self-quotations (but Jansen et al. 2001 disagree). Finally, Hanote (2015) points out that there are major differences in terms of prosodic marking according to the position of the reporting frame – initial, medial, final.

All in all, even though all studies find that there is some marking at least some of the time, the evidence remains inconclusive. First, although some analyses report significant effects of certain prosodic factors on the distinction between DD and ID/narration, these significant effects never amount to systematic marking; hence they provide insufficient support for the Necessity Claim. Second, the considerable variation in the means used to signal quotation does not support the hypothesis that a single morpheme/operator, realized prosodically, is responsible for generating quotation in speech.

4.2 Oral realization of written marks?

The inconclusive results reported in the previous section nonetheless leave open the possibility that readers are sensitive to the presence/absence of marks of quotation. In a recent article, Schlechtweg and Härtl (2020) investigated the potential acoustic correlates of marks in two different contexts, ‘name-informing’ as in (24), and ‘non-name-informing’ as in (25). They recorded their subjects reading short discourses like:

One of their key findings is that “nouns highlighted with quotation marks are pronounced differently from the same nouns without quotation marks” (2020: 15). In other words, marks of quotation have acoustic correlates. This means that readers perceive quotation marks as “linguistically relevant markers” (2020: 22). Still, Schlechtweg and Härtl also found that “[c]rucially, in quotational contexts in which the target item is not embraced by quotation marks an acoustic correlate was not found” (2020: 22–23). In other words, in name-informing contexts, in which a lexical item already invites a quotational reading (nennt in (24)), there are no significant prosodic differences between the quote-marked and unmarked conditions, suggesting that the presence of a predicate like nennen is enough for the identification of quotation. This leads the authors to conclude that their results are “not compatible with approaches that implement quotation marks and their acoustic equivalents as an essential part of a quotational construction and its semantic compositional representation” (2020: 22).

4.3 Taking stock

The data reviewed in Sections 3 and 4 are inconsistent with the view that marks of quotation are mandatory. This confirms the skepticism that arose from the brief outline of the history of marks of quotation in the Introduction.

All the same, I don’t think the case is closed. It makes sense to speculate that type-1 theorists have something else in mind than a literal claim. Incidentally, that does make one wonder why the question of the actual import of the Necessity Claim is not more often tackled head-on. Its proponents might argue that what I have called unmarked quotations are only apparent counterexamples. One reason might be that those unmarked quotations are not, in fact, quotations. In that case, they can be simply dismissed as irrelevant. Another reason might be that there is more to unmarked quotation than meets the eye: though not visible or audible, marks of quotation ‘are there’ in the underlying structure. Both of these responses are examined in Section 5.

But before we proceed to this discussion, I need to consider another possibility. Some writers might submit that the Necessity Claim is valid for only some kinds of quotation, or for some definitions of quotation. Thus, Cappelen and Lepore seem ready to accept that separate theories are needed for written and spoken quotations: “In reply to [the claim that in spoken language, no obvious correlates of quotation marks exist], we will reject the idea that a theory of quotation should accommodate utterances in which there genuinely are no quotation marks […]. We’re interested in theories for sentences that contain the syntactic devices [left and right inverted commas]; other sentences aren’t our concern in this study of quotation” (2007: 13). There is a prescriptive strand in some type-1 theories of quotation that deserves careful consideration.^[19] Still, type-1 theorists have usually presented their proposals as ‘theories of quotation’, thus making it legitimate for me to regard them as descriptive accounts of an empirical phenomenon. That is how I will continue treating them in what follows.

5.1 Pragmatic repair of false statements

Though promoting distinct theories of quotation, several scholars have claimed that utterances like (26) and (27) are false (Cappelen and Lepore 2007: 42–43; García-Carpintero 2004: 677; Gómez-Torrente 2001: 130; Quine 1940: 23) because they predicate a linguistic property of a city and a person, respectively.

Both Gómez-Torrente and García-Carpintero (see also Cappelen and Lepore 2007; Harth 2011: 203) none the less acknowledge that these utterances can succeed in making statements about the words Boston and Donald, via a mechanism of pragmatic repair. Gómez-Torrente appeals to speaker’s reference (vs. the semantic reference that makes (26) and (27) false), while García-Carpintero writes that these examples involve flouts of Grice’s first maxim of Quality which trigger conversational implicatures with the intended metalinguistic meaning.

Though initially attractive, these suggestions are marred by a major flaw. The pragmatic repair described assumes that the writers who produced (26) and (27) intended for it to be manifest to the reader that they were literally false. Let us focus for a moment on García-Carpintero’s proposed solution. Within a Gricean framework, there are no accidental implicatures: a conversational implicature is a proposition the utterer means. Accordingly, one should assume that metalinguistic statements that are not marked for quotation are produced in circumstances where reliance on conversational implicatures are apt, while quote-marked ones are produced in circumstances in which literality is required, and likewise for reliance on speaker’s versus semantic reference. In other words, if Gómez-Torrente and García-Carpintero are right, writers who produce (26) and (27) are happy to omit marks because they consciously rely on readers’ ability to make the right inferences.

There’s the rub. In their own writings, those very writers alternately produce marked and unmarked quotations in relevantly similar contexts. Below, I include additional examples by Cappelen and Lepore, who endorse the pragmatic-repair story for some unmarked cases. I hasten to add that Gómez-Torrente and García-Carpintero are amongst the most careful and precise writers that I know. This hints that the vast majority of writers will display the same kind of variation in their usage.

Each pair of examples in (28)–(30) presents syntactically and semantically matching environments. We can see each writer sometimes using marks of quotation and sometimes not. There is no indication – given the matching contexts – that this variability signals different intentions on the writer’s part. In particular, there is no reason to believe that in the unmarked cases the writers chose to rely on conversational implicature or on the speaker’s reference for the metalinguistic dimension of their statements to be processed by the readers.

I take these examples to demonstrate that the ‘pragmatic-repair’ accounts are empirically flawed.

5.2 Unrealized quotative operator

Linguists of all stripes admit that there may be aspects of structure that are not explicitly realized. So the question whether quotation marks might ‘be there’ even when they are not seen or heard is one that must be taken seriously. Cappelen and Lepore, for instance, claim that “there are elliptical quotation marks […] for example, if the expression without quotation marks is of the wrong grammatical category” (2007: 41), as in Since is a five-letter word.^[20]

Cappelen and Lepore hint at an account according to which marks of quotation are the realization of an underlying functional element that is always there when quotation takes place, in other words, an account that endorses the Necessity Claim. In what follows, I will refer to that putative element of syntax as the ‘quotative operator’, because that is the label under which something like it has most often surfaced in the relevant literature. The understanding is that this operator occurs either overtly (e.g., in the form of inverted commas in writing or a special cluster of prosodic features in speech)^[21] or covertly.

Several semanticists have proposed detailed formal accounts of the formation of quotations which take marks of quotation to be that operator. Predelli (2008) has done this for a Demonstrative account of pure quotation. Pagin and Westerståhl (2010: 405) have proposed a mechanism for the generation of pure quotation via a “syntactic operation k that puts quote marks around any expression e, and interprets the result as referring to e”. And Maier has done the same for PQ and HQ (2014a, 2014b). But these syntactic rules simply postulate that it is marks of quotation that do the relevant syntactic work. They make no predictions as to the conditions under which the marks can or cannot be omitted from surface structure, which is what you would expect if those marks are overt realizations of an underlying element.

An analogy is useful in this regard. Consider the functional projection CP (for Complementizer Phrase) and its head C⁰ in generative syntax. C⁰ is sometimes realized phonetically, and sometimes not (it then has a ‘null phonetic spell out’). Let us for a moment focus on the (non-)realization of C⁰ in English declarative content clauses.

When C⁰ is realized in that environment, it takes the form of the subordinator that (31). When it is not, a ‘null’ complementizer heads the CP (32).

Syntacticians agree that there are conditions under which C⁰ must be overt, other conditions under which it must be null, and finally conditions under which realization is optional and depends on non-grammatical factors. I give illustrations of the first two cases below. Optionality has just been exemplified by (31) and (32).

(For more on those conditions, see Huddleston and Pullum 2002: 952–954; Radford 2009: 112–114). Note also that, within the Minimalist Program, syntacticians widely assume English declarative main clauses too to be headed by a C⁰, albeit one that must be null/covert.

The main point is that there are well-defined conditions for the (non-)realization of C⁰ in finite clauses. The same is true of the (non-)realization of C⁰ in non-finite clauses, just as it is true of other functional heads (e.g., definite and indefinite articles) that may be overt or null in English. It would therefore have been reasonable to expect the semanticists who champion the Necessity Claim to provide an account of the rules that govern the surface realization or not of the quotative operator. But they don’t. Do syntacticians provide such an account?

There are a number of analyses of (particular types of) quotation within generative syntax, the family of approaches to grammar most likely to postulate null elements. The following overview is based on Collins and Branigan (1997), Collins (1997), Suñer (2000), de Vries (2006, 2008, Haddican and Zweig (2012), and Kluck and de Vries (2015). I take this set of articles to be representative of generative approaches. All of them postulate some quotative operator. It is impossible to go into the details of the different proposals, nor is that called for. It will do to highlight the features most relevant to present purposes.

First, these proposals are not intended to account for quotation in general. Collins and Branigan (1997), Collins (1997), and Suñer (2000) postulate a quotative operator in order to explain the grammar of DD with a reporting frame and especially cases of inversion between subject and verb in the reporting frame of examples like (35)–(36); de Vries (2006, 2008 and Kluck and de Vries (2015) posit a quotative operator only in sentences in which the reporting frame is parenthetical (36)–(37); while Haddican and Zweig’s (2012) proposal concerns quotative be like and its Dutch counterpart hebben zoiets van (literally ‘have such-something of’) (38):

None of these proposals invokes a quotative operator to explain pure quotation or non-embedded DD, for example. None of them is conceived as a syntactic account of quotation at large.

On all the proposals, the quotative operator is base-generated either as a direct object or an adjunct within the VP of the reporting frame or as a nominal complement of like in Haddican and Zweig (2012). It is explicitly not part of the quotation. That is its raison d’être: it must be able to move without the quotation itself moving. As for the quotation, which supplies the quotative operator with an antecedent, Collins and Branigan (1997: 11–12) and Suñer (2000: 540) agree that it is ‘discourse generated’, i.e., not generated by some lexical element like the reporting verb (or like). These facts throw light on why the syntactic analyses reviewed here could not be extended to all quotations.

It is striking that no mention is made in any of the proposals of any role of marks of quotation in generating the quotation.^[22] The quotative operator is never associated to marks of quotation or involved in making predictions about the (non-)occurrence of marks of quotation. For Collins and Branigan, Collins and Suñer, the quotative operator is necessary for feature-checking purposes: some element must move to the specifier position of a higher functional projection, and that element clearly cannot be the DD itself, since it can appear not just in initial position (36), but also in final position (as in most examples of DD so far) or discontinuously (37). For de Vries and Kluck and de Vries, the operator makes it possible to understand how reporting frames in Dutch and German, which seem to have V1 word order, actually instantiate V2, as befits main clauses. For Haddican and Zweig, evidence shows that the DD itself cannot be the complement of like, hence something else must hold its place: the quotative operator.

On all the proposals but one, the quotative operator is always null (in Standard English), quite unlike an operator that could surface as inverted commas. The one exception is found in de Vries (2006) and Kluck and de Vries (2015), who take it that their operator can be realized overtly, as zo in Dutch and so in German:

For Dutch at least, they explain the possible (and actually very frequent) non-realization of the operator as follows: zo is anaphoric to the immediate syntactic context, hence it is a topic. Dutch permits ‘topic drop’, and that is what happens when zo is unrealized (2015: 115–117). This way Kluck and de Vries fulfill the requirement voiced previously that when a functional element can be overt or covert, any proper theory should state the conditions for its non-realization at the surface – a requirement that, to my knowledge, is never met by type-1 theorists. The main point here, however, is that this laudable explicitness of Kluck and de Vries’s account is of no help to type-1 theorists: the (non-)realization of the zo/null operator and the (non-)realization of quotation marks are separate and independent issues.

A last point, again in connection with Kluck and de Vries (2015), is that their zo/null operator is not assumed to be involved only in quotational contexts: it is a feature of ‘comment and reporting parentheticals’ at large, including examples like (40):

In other words, it is not exclusively a quotative operator.

The lesson from this scrutiny of the generative analyses of quotation is clear. There is currently no syntactic account that takes marks of quotation to be the overt realization of an abstract operator. In other words, generative analyses provide no support for the Necessity Claim.

5.3 Intermediate conclusion: are unmarked examples just apparent counterexamples?

Type-1 theorists proposed basically two (non mutually exclusive) ways of dealing with unmarked quotations: pragmatic repair and a null quotative operator. I have made a case for the empirical untenability of the repair account. Then, I have shown that the quotative operator postulated by several generative proposals offers no support for the Necessity Claim, for two reasons. First, that operator is not articulated with the presence/absence of quote-marking. The generative analyses start from the observation that there is a quotation (DD) and then posit that a null element is necessary to explain certain grammatical peculiarities of sentences containing DD. The second reason lies in the way the quotation itself is generated: on the generative accounts, it is discourse-generated; on the type-1 theorists’, it is generated syntactically. This is an irreconcilable difference.

To those advocates of the Necessity Claim that might want to object that semanticists and syntacticians are talking about different quotations, the former about PQ (and HQ), the latter about DD, I would respond that most type-1 theorists “make a not unreasonable assumption that [the syntax of sentences containing DD] is pretty much the same as that of pure quotation sentences” (Cappelen and Lepore 2007: 131). This means that the semanticists’ and the syntacticians’ analyses can be compared. And shown to be incompatible in their current state.

There is more. Even if this irreconcilable difference was put right, it is unlikely that a single quotative operator will ever be proposed which would match the range of uses of marks of quotation. We saw in the previous section that only framed DD was linked to a quotative operator. In the studies (co-)authored by de Vries, however, embedded DD was not. And no study suggested that unframed DD would require a quotative operator. PQ is not tackled in the set of studies reviewed above. If a quotative operator was postulated for PQ, it would have to be different from the one for DD. Remember that the latter was base generated as a complement or adjunct to a reporting verb. Such reporting verbs are not normally involved in PQ (though distinct metalinguistic predicates often are).

And then there is the case of HQ. On the theory I defend, sentences with hybrid quotation are just ordinary sentences on a segment of which a quoting act is superimposed (Section 2.1). My contention (contra Maier 2014a, for example) is that these sentences are nothing special from a grammatical point of view, unlike sentences containing recruited quotation. Hence, I see no reason to postulate a quotative operator in an account of their grammar.^[23] However, if it turned out that a quotative operator was proposed for HQ too, it would again have to be different from the one for DD and any putative one for PQ. That is because the operator could not be base generated by a reporting verb (which may or may not be there, and which licenses ID when it is) or by a metalinguistic predicate. Another crucial reason is that the putative HQ operator would not be employed to explain any of the grammatical effects associated with framed DD and PQ. As pointed out in Section 2.1, the internal syntactic structure of (embedded) DD and of PQ is segregated from the syntax of the host sentence, leading to all sorts of interesting grammatical consequences. That is not the case with HQ.

Section 6 will briefly sketch how a Depiction theory can account for the grammatical consequences just mentioned. But before getting there, it is useful to see how type-2 theories fare in the light of the generative proposals reviewed in Section 5.2. Surprisingly – maybe – the generative proposals are compatible with type-2 theories, as they (i) make no connection between the quotative operator and marks of quotation, and (ii) assume that quotations are discourse generated, not generated by any syntactic process that encloses strings of words within marks of quotation. They are therefore consistent with the view that marks of quotation are disambiguators. In consequence, type-2 theorists can accept a null quotative operator as the instrument of syntactic recruitment. After all, there are manifest grammatical consequences to recruitment. At the same time, the type-2 theorist is not committed to such an operator. Whether recruitment needs to be captured by such an operator depends on whether the type-2 theorist judges that these grammatical consequences must be traced to an element of linguistic structure or can be explained pragmatically.