The form of indirect quotes is largely determined by the main verb: verbs like zeggento say and denkento think select declarative clauses, while verbs such as vragento ask select interrogative clauses. See Section 5.1.2.1 for a more detailed discussion of the selection restrictions on declarative and interrogative object clauses.

The examples in (131) show that indirect quotes typically follow the clause-final verbs (if present) and that placing such quotes in the middle field is usually marked and triggers a factive reading; cf. Section 5.1.2.3. Topicalization and left-dislocation of indirect quotes is possible; cf. Section 5.1.2.2 for a more detailed discussion of the placement of direct object clauses.

The use of an anticipatory pronoun seems possible, but marked; the examples in (133) are more likely to be interpreted with a deictic (i.e. discourse-related) interpretation of the pronoun, which leads to a factive reading of the indirect quote; cf. Section 5.1.2.3, sub IIIB.

Embedded declarative clauses are fully transparent to wh-extraction in the sense that both arguments and adjuncts can be extracted. Recall from Section 5.1.1, sub III, that wh-extraction becomes impossible when an anticipatory or deictic pronoun is added.

Wh-extraction is not possible from embedded interrogative clauses. The standard analysis in generative grammar is that this is due to the fact that wh-extraction cannot occur in one fell swoop, but must proceed via the clause-initial position of the object clause; this position is available in declarative examples such as (134), but is occupied by a wh-phrase in embedded wh-questions such as (135), or a phonetically empty question operator in embedded yes/no questions.

Note in passing that, contrary to what has been reported for English, wh-extraction of the subject in (134a) is acceptable despite the presence of a complementizer, and that most Dutch speakers find the three examples in (135) equally unacceptable. We will not dwell on these issues here, but refer the reader to Section 11.3.1 for a detailed discussion of the restrictions on wh-extraction in Dutch. Note also that example (135c) is perfectly acceptable if wanneerwhen is understood as a modifier of the matrix clause, but this is of course not the reading intended here (as indicated by the placement of the trace tj within the embedded clause).

Referential personal pronouns contained in an indirect quote can be bound by an antecedent in the say-clause; cf. Section N22.2 for a detailed discussion of binding of such pronouns. Because referential pronouns can also co-refer with referential expressions as a result of accidental coreference, we illustrate this with examples in which the antecedent is a quantified expression like iedereeneveryone or niemandnobody. Example (136a) first shows that the pronoun hij cannot be referentially dependent on a universally/negatively quantified expression if the latter is part of another sentence; in such cases the pronoun must refer to some known entity in the domain of discourse. The fact that the pronoun can have a bound-variable reading, i.e. can be interpreted as referentially dependent on the quantifiers in (136b), shows that we are not dealing with accidental coreferentiality but with binding; italics indicate the intended binding relation.

The acceptability of the bound-variable reading in (136b) unambiguously shows that we are dealing with an object clause; if the indirect quote were not the object of the verb zeggento say, there would be no c-command relation between the subject of the say-clause and the pronoun, and consequently binding would be wrongly predicted to be impossible, just as in (136a).

That indirect quotes are object clauses is also shown by the fact that negative polarity items (NPIs) such as ook maar ietsanything can be licensed as part of an indirect quote by a negative element in the say-clause. The reason is that NPI licensing, like binding, requires c-command between the NPI and its licenser. NPI licensing is excluded in (137a) because the NPI and its potential licenser niemandnobody are not in the same sentence, and consequently there is no c-command relation between them; NPI licensing is possible in (137b) because the subject of the matrix clause does c-command the NPI in the embedded object clause. Italics indicate the relation between the NPI and its intended licenser.

The previous subsections have shown that quotes in indirect reported speech constructions are direct object clauses. They exhibit the properties of regular object clauses, which were discussed in more detail in Sections 5.1.2.1 through 5.1.2.3. In addition, the discussion of binding and NPI licensing has established that subjects of say-clauses c-command the constituents in indirect quotes, which lends additional credence to the claim that such quotes are regular direct object clauses.

Direct reported speech constructions are often ambiguous. We will argue that such constructions allow not only an analysis as regular transitive constructions in which the quote functions as a direct object, but also an analysis in which the quote can function as a main clause with an embedded parenthetical say-clause; cf. De Vries (2006b).

Semi-direct reported speech constructions exhibit more or less the same syntactic behavior as their direct counterparts. For example, the direct reported speech constructions in (139) can easily be transformed into the semi-direct reported speech constructions in (154). This shows that semi-direct quotes, like direct quotes, do not have to function as direct objects of the verb zeggento say. It is therefore possible that in the examples in (154) the semi-direct quotes are not actually part of the first sentences, but consist of a series of independent sentences.

The hypothesis that semi-direct quotes are independent sentences may also explain why semi-direct reported speech constructions cannot be embedded whereas direct reported speech constructions can. The contrast in acceptability between the two primed examples in (155) illustrates this. However, it may not be a syntactic issue after all: embedding semi-direct speech constructions may simply be incompatible with the fact that a semi-direct quote is a stylistic means of expressing the inner thoughts of the protagonist(s) of a story; cf. the discussion in the introduction to Section 5.1.2.4.

Such a pragmatic account of the contrast between the two primed examples in (155) may be preferable in light of the fact that there are reasons to assume that semi-direct quotes can also function as direct objects. This is clear from the fact, shown in (156), that the direct quotes in (142) can easily be replaced by semi-direct quotes.

The acceptability of (156a&c) again show that semi-direct quotes can be independent sentences with a parenthetical say-clause; this is confirmed by the examples in (157), which show that the say-clauses can split the quotes.

The acceptability of (156b) shows that semi-direct quotes can also function as direct objects; this is supported by the fact, illustrated in (158a), that the say-clause with resumptive dat cannot split the quote. Note that we put aside the fact that this sentence is perfectly acceptable if it is interpreted as an assertion of the speaker himself, who in addition points to Jan as his source of information by means of a parenthetical clause, as in (158b); cf. also the discussion of the examples in (148).

The difference between the semi-direct reported speech constructions in (157) and the non-quote construction in (158b) can be brought out more clearly by considering coreferentiality: whereas the pronouns in the quotes in (157) can be interpreted as coreferential with the subject Jan of the say-clauses, the pronoun in (158b) does not allow this so easily; this is even clearer in the somewhat more elaborate primeless examples of (159). Example (159b') shows that the proper name in (159b) must be replaced by a pronoun to allow the intended coreferentiality reading, and even then this reading is often emphasized in speech by the addition of the emphatic marker zelfhimself. Note that some of our informants do allow the intended coreference relation indicated in (159b); this is indicated by the percentage sign.

Although the discussion above suggests that semi-direct quotes can function not only as independent sentences but also as direct objects of the verb of saying, it should be noted that they never occur in the middle field of the say-clause: the direct reported speech construction in (140b), repeated here as (160a), has no semi-direct counterpart: the number sign indicates that example (160b) is unacceptable in the intended reading and can at best be interpreted as a direct quote, i.e. with Jan literally saying “Hij was ziek”.

Furthermore, the examples in (161) show that semi-direct quotes differ from direct quotes in that they cannot be used as the complement of a noun.

If say-clauses without resumptive dat are indeed parenthetical clauses, we expect that, as in the case of direct quotes, we should find a wider range of examples that do not involve verbs of saying/thinking. This expectation is again fulfilled, as the cases in (162) show. As in (152), the quote cannot be analyzed as an argument of the verbs beginnento begin, vervolgento continue, and besluitento conclude, because these already have a direct object, namely zijn verhaalhis story. Therefore, analyzing the say-clauses as parentheticals seems to be the only option.

This subsection discusses additional arguments, mainly taken from Corver (1994b) and Corver & Thiersch (2003), in favor of the conclusion reached in the previous subsections that (semi-)direct reported speech constructions can be structurally ambiguous. The starting point is the observation in Reinhart (1983b) that (semi-)direct reported speech constructions can be semantically ambiguous in the sense that the say-clause can be either subject-oriented (i.e. pertaining to a certain subject matter) or speaker-oriented. We will see that this semantic ambiguity correlates with the structural ambiguity discussed in the previous subsections.

The discussion in the previous subsections has shown that analyzing say-clauses of the kind in (170) as parentheticals is a feasible option. However, it is still far from clear what the internal structure of these parenthetical clauses is.

An analysis has been proposed for English in Branigan & Collins (1993), Collins & Branigan (1997) and Collins (1997): it involves the movement of a phonetically empty quotative operator, which functions as the object of the verb of saying, into the clause-initial position of the parenthetical say-clause. The desired interpretation is then derived by assuming that the empty operator is coindexed with the quote. Applied to the Dutch cases, this would correctly account for the fact that the verb of saying can be the first visible element in the parenthetical clause in (170), since the first position of the clause is filled by the phonetically empty element OP. Note that the finite verb is in the second position of the parenthetical clause, as can be seen from the fact that it precedes the subject Jan (and other clausal constituents, if present); cf. also Kluck & De Vries (2015: §2.2). Note that the finite verb is in the second position of the parenthetical clause, as can be seen from the fact that it precedes the subject Jan (and other clausal constituents, if present); cf. Kluck & De Vries (2015: §2.2) form more detailed discussion.

However, the proposal in (171) does not take into account the fact, established in the previous subsections, that the overt counterpart of the postulated empty operator is zoso, not datthat (which in fact also applies to English). A proposal that would solve this problem can be found in Corver (1994b) and Corver & Thiersch (2003), where it is assumed that zo is phrasal and actually contains a phonetically empty pronominal element pro functioning as a direct object; cf. De Vries (2006b:217) and Schermer (2024:342) for similar intuitions.

The proposal in (172) raises all sorts of technical issues (such as the fact that Dutch does not normally allow pro objects, and that pro is not directly related by movement to the object position of the parenthetical clause): however, it has a number of favorable consequences. For instance, it voids the need of postulating a specialized phonetically empty quotative operator, given that the operator zo is a resumptive proform, which can usually be phonologically deleted by topic drop, as in Dat boek, (dat) heb ik niet gezienThat book, I have not seen it; cf. Section 11.2.2. The proposal would also account for the fact that het cannot be present in parenthetical say-clauses, despite the fact that they are typically headed by obligatorily transitive verbs; in (173) the direct object is expressed twice, once by pro and once by the pronoun het.

The downside, however, is that zo can sometimes be combined with a direct object; cf. also Kluck & De Vries (2015: §3). This was already shown in (152) and (162) above, where the (optional) noun phrase zijn verhaalhis story clearly functions as a direct object. In examples such as (151b), repeated below as (174b), a direct object is even obligatory: while the pronoun het can be omitted in the transitive direct reported speech construction in (174a), it must be present in the construction with zo in (174b). This means that next to resumptive phrasal zo in (173), which is obligatorily moved into clause-initial position, there must be a simple proform zo acting as a pure adverbial phrase of manner;

However, there are also transitive direct reported speech constructions with zo, in which the direct object can be omitted; this is illustrated for the examples in (175), in which the presence of the discourse particles nog and maar seems to make the omission of the object pronoun het possible. This leads to the rather ad hoc stipulation that in certain constructions zo obligatorily contains the empty pronoun pro, while in other constructions it cannot or only optionally does so.

However, if we reject Corver & Thiersch’ proposal, we must conclude that we can leave the direct object of the verb zeggen unexpressed in examples like (149) and (150), despite the fact that example (138a) has shown that the verb zeggen cannot normally occur without a direct object; cf. Schermer (2024: §3.3) for a proposal in this direction. At the present moment, such a position seems equally ad hoc, in the absence of independently motivated evidence. We therefore conclude that we do not yet have a fully satisfactory analysis for parenthetical say-clauses. As we have nothing more illuminating to say on the subject, we leave it to future research and simply repeat our previous conclusion that direct and semi-direct reported speech constructions can be ambiguous.

For completeness’ sake, note that we have not discussed the tentative parasitic gap analysis in Kluck & De Vries (2015: §3), because the proposal seems to fail in that the configuration of examples without an overt object violates the anti-c-command restriction on the parasitic gap (the object of the parenthetical clause) and the trace of its presumed antecedent (the wh-moved adverbial phrase zo); cf. Section 11.3.8, sub IIIB, for a discussion of the restriction.

The previous subsections have discussed reported speech constructions with a parenthetical clause introduced by zoso, illustrated again in example (176).

Kluck & De Vries (2015: §4) have noted that Dutch has a different construction with a parenthetical clause introduced with zoalsas, which is illustrated in (177a&b). The two constructions are similar in that they are both common with the verbs of saying, writing, thinking, observation and explanation in (153a&b). However, verbs of manner of speech and sound emission, such as schreeuwento shout, vloekento curse, and zuchtento sigh in (153c), do not seem to be easily possible in the construction with zoals; cf. example (177c).

The dollar sign in (177c) is used to indicate that the unacceptability of this example may be related to a difference in meaning between the two constructions. Example (176) with zo assigns the responsibility for the quote to the subject of the parenthetical clause and, in addition, clarifies the subject’s relation to the quote related to: (i) the manner of transfer in the case of the verbs of saying and writing; (ii) the epistemic status of the utterance in the case of the verbs of thinking, observation and explanation; (iii) the subjective attitude toward the utterance in the case of the verbs of manner of speech and sound emission. In examples such as (177a), on the other hand, the responsibility for the quote is fully assigned to the speaker, who only mentions the subject of the parenthetical clause for recognition or as an authority; this makes the subjective attitude of the latter irrelevant, although it should be noted that the verb vrezento fear would still be perfectly acceptable in example (177c).

We leave this for further exploration and focus in the remainder of this subsection on some important syntactic properties between the two constructions. The most important property is that parenthetical clauses introduced by zo are main clauses while parenthetical clauses introduced by zoals are non-main clauses.This is clear from the fact that the finite verb zei in (176) is in verb-second position, while the finite verbs in (177) are in clause-final position. The latter placement led Kluck & De Vries (2015: §4) to conclude that the complementizer position of the parenthetical clauses in (177) is filled by the element als, which blocks verb-second of the finite verb. The surface order of the parenthetical clauses in (177a) is therefore as indicated in (178).

This analysis would imply that zoalsas is not a single word but consists of two words: the element zo and the complementizer als. This proposal implies that the two constructions are essentially the same as far as the element zo is concerned. This is supported by the fact that they seem to behave similarly with respect to long wh-movement: the adverbial zo is extracted from the complement clause of the bridge verb horento hear and placed in the initial position of the parenthetical clause. The main difference between the two constructions is that in (179a) zo can be optionally omitted, while in (179b) it must be overtly realized.

The fact that the element zo is identical in the two cases will make it easier to account for the fact that the otherwise obligatory transitive verb zeggento say, does not tolerate an (overt) direct object in the prototypical cases of both constructions; cf. Subsection IIB and Kluck & De Vries (2015) for some exceptional cases.

The discussion of the syntactic properties of parenthetical reported speech constructions will have made it clear that there are still several unanswered questions that we will need to leave for future research.

This subsection examines the quotative van-construction and is organized as follows. Subsection 1 discusses the internal structure of the quotative van-phrase, Subsection 2 discusses the meaning of the quotative van-construction as a whole, and Subsection 3 discusses some of its syntactic properties. Along the way, we will point out a number of differences between quotative van-constructions and reported speech constructions without van.