We will start with two arguments in favor of the complex-complementizer analysis. The first argument in favor of this approach to voordat etc. is that the paradigm is not complete. If temporal adpositions can take a clause as their complement, there is no obvious reason why we use morphologically simple words like toen and terwijl instead of the clumsy sounding sequences tijdens datduring that and gedurende datduring that. If formations like voordat are listed in the lexicon as complex complementizers, the awkwardness of tijdens dat and gedurende dat could just be considered a matter of accidental morphological gaps.

The distribution of the element dat seems largely unpredictable: in (60a&b) and (62b) the presence of dat seems optional, in (60c) and (62a&c) dat is obligatory, and in the examples in (61) dat can never be realized. Since in embedded declarative complement clauses the complementizer dat is usually present, the fact that dat can or must sometimes be omitted is a problem for the analysis in (58c), according to which the clause is a complement of the preposition, and thus supports the analysis in (58b), which would be compatible with the claim that we are dealing with complex complementizers listed in the lexicon, some of which have shorter allomorphs.

The arguments in Subsection 1 favor the complex-complementizer analysis, but there are also arguments in favor of the competing analysis in (58c), where the preposition takes a clause as its complement. An obvious argument for the analysis according to which we are dealing with a preposition with a clausal complement is that it analyzes the adverbial clauses as regular PPs in which the nominal complement is replaced by a clause. Also the semantics would be completely regular: an example such as (58a), repeated here as (63a), expresses that the “kissing” event denoted by the main clause precedes the “leaving” event denoted by the adverbial clause; in this respect there is no crucial difference with example (63b). The same applies to the other sequences P+dat ... in (60).

The counterargument from subsection 1, viz. that not all temporal prepositions take a clause as their complement, can be countered by claiming that sequences like tijdens/gedurende datduring that are awkward but in fact grammatical, which can be supported by the fact that such cases can easily be found on the internet; cf. Barbiers (2005). Furthermore, proponents of the regular PP analysis might point out that while voordatbefore, nadatafter, and totdatuntil are explicitly treated as subordinators in traditional grammars and dictionaries, this is not the case for tegen datby the time that and sinds/vanaf datsince then. This may support the claim that temporal prepositions in general can take a finite clause as their complement. Finally, the clumsiness of tijdens/gedurende dat mentioned above could be attributed to lexical blocking, i.e. to the fact that the lexicon contains the more or less synonymous complementizers toenthen and terwijlwhile.

Even though proponents of the complementation analysis in (58c) might acknowledge the problem of the distribution of dat, they might point out that it should make us suspicious that the supposed complex complementizers always have a prepositional element as their first member. This could perhaps be used as an argument for the idea that in (60) and (62) we are actually dealing with prepositions taking a finite clause as their complement. A much stronger argument for the claim that we are dealing with PPs is that there are correspondences with undisputed PPs in terms of modification. As shown in (64), the temporal prepositions voor and na can be modified by the same elements as the sequences voor dat and na dat.

A conclusive argument in favor of a clausal-complement analysis would be an example in which an element intervenes between the preposition and the finite clause, i.e. an example comparable to cases like tot vlak voor de deur until right in front of the door or voor daar bij for with it, which were adduced in Section 33.2 to show that tot and voor can take prepositional complements. Unfortunately, such examples cannot be constructed because complement clauses do not allow modification or R-extraction. Nevertheless, there are semantic reasons to think that certain phonetically empty elements occur between the preposition and the complementizer, but these seem to support the relative-clause analysis rather than the clausal-complement analysis.

Consider example (65a), which is ambiguous between two readings, cf. Larson (1990) and Haslinger (2007). In the first reading, the temporal expression naafter takes scope over the whole string it precedes (dat hij beweerde dat hij vertrokken was): this leads to the paraphrase in (65b). In the second reading, the scope of the temporal expression is limited to the more deeply embedded clause (dat hij vertrokken was): this leads to the paraphrase in (65c), which can be understood as expressing that the culprit committed perjury.

The crucial fact is that the slightly awkward construction in (66a) has the same scope properties as example (65a).

We can explain the ambiguity of (65a) by assuming that the temporal expression is related to its scope position by a phonetically empty relative element rel with the function of an adverbial phrase of time: rel is moved into the clause-initial position before the complementizer dat and takes an empty noun as its antecedent. This would mean that the reading in (65b) corresponds to the structure in (67a), where rel originates in the matrix clause, and the reading in (65c) corresponds to the structure in (67b), where rel originates in the embedded clause. For comparison, the structures attributed to the two readings of (66a) are given in the primed examples.

For more evidence for a wh-movement analysis of scope ambiguities of this sort, but without the postulation of a concealed relative-clause structure (which was adopted from Barbiers 2005), we refer the reader to Larson (1990).

The discussion in the previous subsections makes it clear that, on the basis of the currently available evidence, it is far from easy to establish explicit criteria that could definitely settle the question of the correct analysis of the sequences P + dat. The three options discussed here all have their merits and we leave it to the reader to decide which proposal is the most promising for further research.