The examples in (10) show that finite verbal argument clauses come in at least two different forms, and that the choice between the two is largely dependent on the matrix verb: the verbs zeggen'to say' and vragen'to ask' differ in that the former takes declarative clauses as its complement, whereas the latter takes interrogative clauses (that is, yes/no- or wh-questions) as its complement.

Although we occasionally find similar differences in the domain of nominal complementation (cf. Jan stelde een vraag/*antwoord'Jan asked a question' versus Jan gaf een antwoord/*vraag'Jan gave an answer'), this distinction is quite basic when it comes to complementation by finite clauses.
      Since Grimshaw (1979) it has often been claimed that verbs are subcategorized for specific semantic types of complement clauses: embedded declarative clauses such as (10a) are of the type "proposition" and embedded questions are of the type "interrogative". Grimshaw adds the type of "wh-exclamative", which is found in the examples in (11); the wh-phrases in these examples are not interrogative but express "high degree" modification, just as in the exclamative main clauses given in the primed examples. Observe that there are a number of differences between the main and embedded clause (e.g. concerning word order and the form of the wh-word), which we will ignore for the moment, but to which we will return in Section 11.3.4.

The fact that Grimshaw (1979) includes exclamatives suggests that the list of semantic types is open-ended in the sense that it would be possible to add more semantic types to it; so it seems desirable to restrict it by imposing principled constraints on the set of possible types. An attempt to do this can be found in Nye (2013), who proposes that complement clauses are selected on the basis of two binary features: ±wh and ±factive. These features characterize the four different constructions in (12) provided we adopt the following definition of factivity: factivity refers to constructions with verbs which take a complement clause, and where the speaker presupposes the truth of some proposition expressed by that clause; see Kiparsky & Kiparsky (1970) in the interpretation of Broekhuis & Nye (2013). In the (a)-examples the relevant proposition is expressed by the full complement clause, whereas in the (b)-examples it is expressed by the non-wh part of the complement clause. For the two types of wh-questions, see also Groenendijk & Stokhof (1984:91ff.) who define the distinction in terms of pragmatic implicatures, that is, the speaker's presupposition instead of factivity.

The binary feature approach thus gives rise to the four construction types in Table 1, which now includes the new class of factive interrogatives illustrated in (12b').

	[-wh]	[+wh]
[-factive]	non-factive declarative (12a)	non-factive interrogative (12b)
[+factive]	factive declarative (12a')	factive interrogative (12b') wh-exclamative (11)

Another advantage of adopting the binary features ±wh and ±factive is that they enable us to account for the fact that betreuren'to regret' and weten'to know' impose different selection restrictions on their complement; the unacceptability of (13a) shows that the verb betreuren is only compatible with declarative clauses, whereas the acceptability of (13b) shows that weten is compatible both with declarative and with interrogative clauses. This can be expressed by assuming that betreuren selects a -wh,+factive complement clause, but that weten does not impose restrictions on the [wh]-feature and thus simply selects a +factive complement clause. Providing a similar account in a non-ad hoc fashion seems harder if we adopt Grimshaw's claim that verbs select semantic types like proposition, interrogative or exclamative.

Note in passing that examples like Ik betreur [wat je hier schrijft]'I regret what you write here' are not relevant in this context: the bracketed part is a free relative, therefore we are dealing with a nominal complement and not a complement clause.
      In a similar way, we might account for the fact that verbs like betwijfelen'to doubt' in (14) can be combined with an embedded yes/no-question, but not with an embedded wh-question by claiming that its interrogative complement clause must be -factive—although it should be noted that this still leaves open why the embedded wh-question in (14) cannot be interpreted as non-factive. Again, providing a similar account is not possible under Grimshaw's proposal where yes/no- and wh-questions are claimed to be of the same semantic type.

For completeness' sake, it should be noted that a less fortunate aspect of a binary feature approach is that it does not account for the fact that factive verbs like weten can also take yes/no-questions: Jan weet (niet) of Marie morgen komt'Jan knows/does not know whether Marie is coming tomorrow', which can never be used to express a non-null proposition. This, as well as the problem noted for example (14), shows that the binary feature approach is still in need of some fine-tuning, but we leave this issue for future research.
      The new class of +factive,+wh verbs does not seem to be restricted to factive interrogative constructions. If we assume that the feature +wh does not refer to a semantic feature but to the formal (syntactic/morphological) feature that wh-elements have in common and that enables them to undergo wh-movement, it may also include verbs taking exclamative complements; cf. the primeless examples in (11) above. Another construction that may be included, which is discussed in Nye (2013), is the one illustrated in (15a); the complement clause in this construction, which is especially found in narrative contexts, is introduced by the wh-word hoe'how' but seems to be more or less semantically equivalent with the factive declarative dat-clause in (15b).

This section has shown that the semantic selection restrictions on finite complement clauses exceed the dichotomies between (i) declarative and interrogative clauses and (ii) yes/no- and wh-questions normally found in descriptive grammars. In addition, we have shown that Nye's (2013) binary-feature approach to the selection of complement clauses has certain advantages compared to Grimshaw's (1979) approach based on semantic types.