Avoiding Overgeneration

The analysis of Dutch verb-clusters presented so far has demonstrated that cross-serial word order can be accounted for by means of a lexical constraint equivalent to a generalized version of division. The account still overgenerates, however, as nothing forces cross-serial word order for VR verbs. In this section we argue that overgeneration can be avoided by requiring that VR verbs take a verbal complex as argument and by introducing a feature that marks certain constituents as not being a verbal complex.

A VR or PE verb may combine with the verbal head of its complement before combining with the (non-verbal) arguments of that head, but nothing in the analysis presented so far enforces that this is obligatory. For PE verbs this is fine, as in these cases both the cross-serial and the extraposed word order is allowed, as well as cases that are a mixture of both (the `partially extraposed' VP appears is italic):

For VR verbs, however, only the cross-serial word order is grammatical:

The constraint we have failed to incorporate is that for VR verbs, cross-serial word order is obligatory. In other words, a VR verb must combine with the verbal head of its complement, before this head has been combined with a complement of, for instance, category NP or PP.

In previous work [11,13,10] it has been suggested that VR verbs take an argument that must be a lexical item. We believe that this suggestion is problematic for a number of reasons.

First of all, whereas it is usually the case that only cross-serial word order is grammatical, there are important exceptions.

The particle aan in (23a) is a so-called seperable prefix of the verb aanspreken, which is illustrated in (23b) by the fact that a VR verb may seperate the prefix/particle from its root. One could argue that eventhough aan in (23b) is clearly a particle, aan te spreken in (23a) is a morphologically complex verb, so that this example is not immediately problematic for an account that requires that the verbal argument of VR verbs must be lexical.

Next, consider (24), where an auxiliary has been added to the verb cluster. In these cases, the particle may occur in three positions.

Example (24b) is problematic if VR verbs take lexical arguments. A verb cluster V₁ V₂ V₃, with V₁ and V₂ VR verbs, must be left-branching (i.e. [[V₁ V₂] V₃]), as otherwise V₁ would not take a lexical argument.¹ For (24b), however, left-branching implies that heeft has to combine with the particle aan before combining with its verbal argument. Such a derivation is not possible given the rules presented in the previous section.

A related problem for analyses requiring that VR verbs take lexical arguments is the fact that there is a - somewhat heterogeneous - class of complements that can occur to the right of a VR verb. Some examples are listed below (see also ANS [8], p. 1012 ff.):

While there are many idiomatic expressions in this class, it seems unlikely that phrases such as tot stand gebracht, schuldig gemaakt, duidelijk gemaakt are in fact lexical items.² Note also that, as with particles, cross-serial word orders are possible as well (and, for some speakers, are preferred):

These can only be derived if the idiomatic expressions involved are not treated as lexical units.

Second, an account which assumes that the arguments of VR verbs are lexical, must either assume that complex verb clusters are left-branching or else that verb clusters themselves are derived lexically. We already pointed out that a left-branching analysis of verb-clusters cannot easily account for the distribution of seperable verb-prefixes or other complements within the verb-cluster.³ An analysis which assumes that the verb cluster itself is derived lexically (as has been proposed in [13]) is not very plausible either. The examples in (25) can only be explained by assuming that quite complex phrases can have lexical status. Furthermore, if verb clusters are complex lexical items, it remains to be explained why parts of it can be fronted (27) and why coordination of parts of the verb cluster is possible (28).⁴ Note that in example (28c) (from [15]), the conjuncts consist of two NP's and a verb. In an analysis where verb clusters are derived lexically, this would require that coordination may compose phrasal and (sub-)lexical elements into one conjunct.

To avoid the problems that arise form the assumption that the arguments of VR verbs must be lexical, we impose the constraint that the verbal argument of a VR verb must be a verbal complex. Assuming a feature VC to distinguish between verbal complexes and other (verbal) constituents, we can implement the relevant restriction as follows (where the subscribt Sign indicates a reentrancy with the second argument of division):

The next, and more difficult, matter that needs to be settled is when exactly a constituent is marked as -VC. Consider a simple example.

Example (30a) is acceptable, whereas (30b) needs to be ruled out. A simple way to achieve this would be to assign the category NP\VP[-VC] to lezen. This blocks the derivation of (30b), since een boek lezen is now of category VP[-VC], which is not unifiable with the argument of wil. (30a), on the other hand is still derivable, as NP\VP[-VC] and (NP\VP)[+VC] are unifiable (note that the category of lezen only specifies that its value is -VC).

This solution does not work for more complex examples, however. Consider for instance the following example:

If we add the specification -VC to the value of lezen only, the complex phrase een boek willen lezen will remain unspecified for the feature VC. One might consider solving this problem by adding the specification -VC to the value of VR verbs. But this would block the derivation of right-branching verb clusters, an option we argued against above. For instance, this solution would block the derivation of verb clusters containing particles:

In (32), the specification -VC is added to the value category of moeten. Since particles can occur as part of the verb cluster, no specification -VC is added to the value of schieten. The derivation is blocked nevertheless, as op moeten schieten is -VC, irrespective of the fact that the value of schieten is unmarked for VC.

The problem with the solutions just considered is that they fail to take into account the fact that it is the argument of a verb which usually determines whether the result of combining that verb with its argument is -VC or not. This information should be preserved if a verb does not combine with its argument directly, but instead is combined with a VR verb that `inherits' the arguments of the verbs it governs. One way to achieve this is to extend the rule of division with a constraint that makes the values of VC on the value categories of functor and argument reentrant:

The derivation of simple as well as more complex verb clusters now proceeds as required. Example (30a) is still derivable, whereas (30b) is not:

Note that the phrase een boek wil lezen is marked -VC. This implies that the ungrammatical example (31) is no longer derivable:

For complex verb clusters containing a particle, however, no problem arises. In particular, the word order in which the particle appears in the middle of the verb cluster (i.e neither at the left-periphery nor adjacent to its governing verb) remains derivable:

The analysis can be extended to verb clusters containing a modal or auxiliary inducing inverted word order. In section 4 we observed that if such a verb selects an extraposition verb, which in its turn selects for a full VP to its right, only cross-serial word order is allowed:

This suggests that inverted modals and auxiliaries are like other VR verbs in that they select a +VC argument⁵ and that extraposition verbs give rise to -VC constituents:

A slightly more complex situation arises if the verb inducing inverted word order is itself governed by a VR verb. This is possible for auxiliaries, but not for modals, as with the latter the possibility of inversion is restricted to finite forms. The examples below show that in such cases, the participle selected by the auxiliary must occur at the left periphery of the verb cluster:

To account for the ungrammaticality of (40b), we must assume that hebben not only selects a +VC argument but also that it marks its value as -VC:

Under this assumption, the derivation of the verb cluster in (40b) is blocked (TV abbreviates NP\VP):

whereas the verb cluster in (40a) can be derived as follows: