We want to conclude the discussion of the classification of verbs on the basis of the number and type of nominal complements they take by pointing out a potential problem for one of the unaccusativity tests used in the preceding discussion: the claim that selection of the auxiliary zijn is a sufficient condition for assuming unaccusative status for a verb. If this test is indeed valid, we predict that there are no transitive verbs selecting zijn. This indeed seems to be true in the general case but there are a small number of potential counterexamples, which we will discuss in this section.
      We will begin with a number of apparent counterexamples: verbs like bijspringen'to help out', ontkomen'to escape', ontlopen'to escape', ontvluchten'to flee', tegemoet gaan/komen'to meet', volgen'to follow' all take an object although they form their perfect tense with zijn. This is not really surprising given that these verbs all take a dative object in German. However, the verb volgen still may be a potential problem given that it can be passivized, which was taken to be a sufficient test for assuming ergativity (which implies transitivity in this specific case).

The seeming contradiction is resolved once we realize that the verb volgen exhibits ambiguous behavior with respect to the auxiliary test; it combines not only with zijn but also with hebben: De politieagent is/heeft de verdachte gevolgd'The police officer has followed the suspect'. This suggests that volgen is undergoing a process of reanalysis; it develops from a verb with a dative object into a verb with an accusative object. A reanalysis of this sort has applied in other cases as well; the German verb hilfen'to help', for example, takes a dative argument and cannot be passivized, whereas its Standard Dutch counterpart helpen exhibits prototypical transitive behavior in that it can undergo regular passivization: Het slachtoffer werd door een voorbijganger geholpen'the victim was helped by a passer-by'.
      Even if we ignore those cases that are susceptible to a dative object analysis, we at least have to deal with the following two (notorious) problems: the transitive verbs vergeten'to forget' and verliezen'to lose', which can take either hebben or zijn in the perfect.

Perhaps we may set the case of verliezen aside as being part of the formal register given that Haeseryn et al. (1997: 79) claim that the use of zijn is not generally accepted and more commonly found in written language than in speech. The case of vergeten is harder to account for. Perhaps we can understand the acceptability of zijn in (152b) better by relating this example to examples such as (153), in which the noun phrase mijn paraplu does not function as a complement of the verb vergeten but as an argument of the embedded infinitival predicate meenemen 'to take along' . One may therefore assume that (152b) has some phonetically empty embedded predicate.

But even if this were viable, it would leave us with cases such as (154), in which vergeten is more specifically interpreted as "to not remember": although Haeseryn et al. (1997: 79) claim that zijn is much preferred in this case, postulation of a phonetically empty embedded predicate seems less tenable. We will therefore not speculate any further on this issue, and simply leave it for future research.

Another incidental case is beginnen'to start' in (155a); example (155b) shows that passivization gives rise to a reasonable acceptable result. Perhaps such cases are relatively new innovations in the language given that beginnen can also be combined with a PP-complement with aan; cf. Zij is aan een nieuwe boek begonnen 'she started on a new book'. Perhaps this can be supported by the fact that is counterpart stoppen can only be combined with a PP-complement; Zij is *(met) haar nieuwe boek gestopt'She has stopped with her new book'. We have to leave further investigation of cases of this sort to future research.