The main difference between (81a) and the primeless examples in (82) is related to the truth of the proposition expressed by the embedded clause; cf. Kiparsky & Kiparsky (1970). Non-factive verbs are used to assert the truth of the argument clause with varying degrees of decisiveness: by using (83a) the speaker expresses that Jan can be held responsible for the truth of the proposition “Els is going to emigrate”, whereas this holds only to a lesser extent when he uses (83b).

Factive verbs, on the other hand, are used when the speaker presupposes the truth of the proposition expressed by the embedded clause; the two examples in (84) express that the speaker considers the proposition “Els is going to emigrate” to be true. Furthermore, the speaker asserts something about this proposition: (84a) asserts that Jan revealed it, and (84b) that Peter regrets it.

That the speaker is not committed to the truth of the proposition expressed by the object clauses of the non-factive verbs bewerento claim and vermoedento suspect in (83) is clear from the fact that he can easily deny that the proposition is true if he thinks/knows that the source of the information is wrong; the denial of the proposition “Els is going to emigrate” in the examples in (85) yields a semantically coherent result.

Things are different in sentences such as (84) with the factive verbs onthullento reveal or betreurento regret; by using these verbs the speaker presents the proposition “Els is going to emigrate” as true, so that the denial of this proposition in the examples in (86) leads to a semantically incoherent or at least surprising result.

Whether an object clause allows a factive reading or not depends mainly on the meaning of the verb/predicate in the matrix clause. In (87) we list some verbs that are typically used in factive or non-factive contexts, as well as some verbs that can be used in both contexts; cf. Kiparsky & Kiparsky (1970) for a similar list for English.

Kiparsky & Kiparsky (1970) proposes several tests that can be used to determine whether or not we are dealing with a factive verb/predicate. Some of these appeal to specific properties of English, so we will only discuss those tests that make the desired distinction for Dutch as well. We will add a number of tests suggested in Barbiers (2000).

The discussion in Subsection II may have suggested that the verb/predicate of the matrix clause fully determines whether the embedded proposition can be construed as factive or not. However, there are a number of additional factors that can affect a verb’s ability to take a factive complement; Barbiers (2000:193) even claims that a factive reading can be imposed on the complement clause of most of the verbs in (87a).

Factive clauses that occupy a position in the middle field of the matrix clause are typically separated from the clause-final verbs by one or more adverbs (if present). This is illustrated in (107) by the modal adverb waarschijnlijkprobably.

The pattern in (107) is different from the pattern we find with the noun phrase het feit dat ...the fact that ... in (108). Since (107b) and (108b) differ in acceptability, this this could be seen as a possible problem for the hypothesis in Kiparsky & Kiparsky (1970) that factive clauses are reduced noun phrases.

One way to solve this problem for Kiparsky & Kiparsky’s hypothesis would be to claim that the difference in word order between (107a) and (107c) is sufficient to make the information-structural distinction between focus (“discourse-new information”) and presupposition (“discourse-old information”), whereas in (108) this distinction depends rather on the position on the nominal part het feit; cf. Sections 13.2 and N21.1.4 for discussion. It remains to be seen, however, whether this line of thinking would lead to a fully satisfactory account of the contrast between (107) and (108).

If we accept the proposal in Sections 13.2 and N21.1.4 that the word order in (108c) is derived by leftward movement of the nominal object, it seems rather attractive to assume that the order in (107c) is derived by leftward movement of the factive clause. An empirical argument for this is that it allows us to appeal to the freezing effect to explain the contrast between the two examples in (109): the factive clause is a transparent to wh-extraction when it occupies its base position after the clause-final verbs, but freezes, i.e. becomes a strong island for wh-extraction, when it is moved to the middle field of the matrix clause.

However, recall from Subsection IIE that factive clauses are weak islands in the sense that wh-extraction is restricted to nominal objects; wh-extraction of adverbial phrases is excluded regardless of the position of the factive clause; this is illustrated again in (110).

The observation that factive clauses are weak islands is actually another problem for Kiparsky & Kiparsky’s hypothesis that factive clauses are reduced noun phrases; complex noun phrases are generally strong islands in the sense that they also block extraction of nominal objects from their complement clause. The examples in (111) show that this is true regardless of whether the clause precedes or follows the clause-final verbs.

So far, we have more or less adopted Kiparsky & Kiparsky’s hypothesis that factive clauses are reduced noun phrases, but Subsections IV and V have discussed a number of possible problems with this hypothesis. Thus, it may be prudent to look for another analysis to account for the differences in behavior between non-factive and factive clauses. One such analysis is provided in Barbiers (2000), where it is argued that non-factive clauses are complements of the verb, whereas factive clauses are adjuncts. This proposal is interesting because it would immediately explain why factive clauses can occur in the middle field of the clause, since this is generally possible with adjunct clauses, as shown by the examples in (112).

If factive clauses are indeed adjuncts, we expect them to have a looser relation to the matrix verb than non-factive verbs. Barbiers (2000) claims that this expectation is indeed borne out, which is demonstrated by pointing to the fact, illustrated by the primed examples in (113), that non-factive clauses must be pied-piped under VP-topicalization, whereas factive clauses can be stranded.

Another observation made in Barbiers (2000) that may point in the same direction is that stranding of the clause can disambiguate examples such as (114a): while (114a) can be factive (the speaker knows that Jan has been ill) or non-factive (the speaker expects Jan to tell a lie, e.g. to excuse his absence), example (114b) can only have the former reading.

However, there are at least three possible problems with this proposal. First, as already pointed out in Barbiers (2000), the hypothesis does not account for the fact that factive clauses are weak (and not strong) islands, since adjunct clauses normally block wh-extraction of nominal objects as well. Second, assigning adjunct status to factive clauses would lead to the expectation that factive clauses can be omitted (which adjuncts generally allow), which is not borne out: *Jan betreurde. Third, we might expect that VP-topicalization must pied-pipe the non-factive clause and must strand the factive clause, while pied-piping seems to give a relatively good result in both cases. We therefore leave the question of whether the hypothesis is tenable to future research.

The term factivity is mostly restricted to verbs that select declarative clauses, because it is defined in terms of the truth value of the propositions expressed by complement clauses. A typical example of such a definition can be found in Crystal (2008): the term factivity is “used in the classification of verbs, referring to a verb which takes a complement clause, and where the speaker presupposes the truth of the proposition expressed in that clause”. The application of this definition is again illustrated in the examples in (115), where S1 ⊨ S2 stands for “by uttering sentence S1, the speaker presupposes that the proposition P expressed by S2 is true”.

Definitions of this kind exclude the existence of factive verbs that select an interrogative complement clause: interrogatives differ from declaratives in that they do not express complete propositions, as they are characterized by indeterminacy in the value of some variable represented by the yes/no operator or the wh-phrase; cf. Grimshaw (1979). Whether or not this exclusion is justified can be tested by examining factive verbs like vergetento forget and wetento know, both of which can also take an interrogative complement. Consider first the examples in (116).

This sentence pair suggests that verbs taking an interrogative object clause are non-factive: by uttering sentence (116b), the speaker does not commit himself to the truth of the proposition expressed by the sentence to the right of the arrow. Of course, this is not surprising: the speaker’s reference to Jan as a source of further information about the truth of the proposition only makes sense if the speaker does not know the answer to the embedded question himself.

However, things seem to be different with embedded wh-questions. Consider the contrast between the examples in (117). By uttering the sentence in (117a) the speaker does not imply that the proposition “Els is leaving” is true, while by uttering the sentence in (117b) he does.

The verbs vragento ask and wetento know thus differ in that the former is clearly non-factive, but that the latter is factive in the somewhat restricted sense that the truth of the proposition expressed by the non-wh part of the complement clause is presupposed by the speaker. The examples in (118) show that this difference between vragen and weten holds not only in cases where the wh-phrase is an adjunct of the embedded clause, but also in cases where it is an argument.

As shown in Subsection II, factive verbs have the property that negating or questioning the clause they head does not affect the entailment; the examples in (119) have the same entailment as example (116a).

The examples in (120) show that the (b)-examples in (117) and (118) also pass this litmus test for factivity; by uttering the sentences to the left of the arrow the speaker entails that the propositions expressed by the sentences to the right are true.

The syntactic tests for factivity yield somewhat ambiguous results. Like the factive declarative clause in (121a), the factive interrogative clauses in (121b&c) can be introduced by the anticipatory pronoun hetit.

However, it seems that placement of a factive complement in the middle field of the matrix clause produces a less felicitous result when the complement clause is interrogative than when it is declarative; whereas (122a) is merely stylistically marked, the examples in (122b&c) seem degraded (although they may improve somewhat with a contrastive accent on the wh-word).

Note that the distinction between two types of wh-questions is not new and goes back at least to Groenendijk & Stokhof (1984:91ff), where the distinction is formulated in terms of pragmatic implicatures instead of factivity, i.e. the speaker’s presuppositions. As far as we know, there is no detailed study of the syntactic behavior of factive interrogative clauses, so we have to leave further discussion to future research.

The previous subsections have shown that there are a large number of systematic differences between non-factive and factive clauses, suggesting that they must also receive a different syntactic analysis. Kiparsky & Kiparsky (1970) argued that the difference should be expressed by assuming a difference in categorial status: non-factive clauses are clausal complements and factive clauses are reduced nominal complements. Barbiers (2000) argued that the distinction is related to syntactic function: non-factive clauses are complements of the verb, whereas factive clauses are adjuncts. Both proposals are embedded in a larger set of theoretical assumptions, and each has its own problems. In any case, the discussion has shown that the fact that factive clauses can occur in the middle field of the matrix clause is not just an isolated fact, but part of a larger set of facts that has yet to be explained.