The examples in (33), repeated here as (42), have made it clear that it is not the preposition itself that is modified by an adverbial phrase but the full PP: this follows from the fact that in the case of R-pronominalization, the R-word er can intervene between the modifier and the preposition.

However, if we look at cases with the elements links/rechts, achter/voor/midden and boven/onder, it looks as if the R-pronoun cannot intervene between most of these elements and the PP. This is shown in the primed examples in (43).

It is not immediately clear how decisive the primed examples in (43a&b) are for the hypothesis that the preposition and the element preceding it form a complex PP, because what we see in (44a&b) brings out that the order in which the R-pronoun precedes the supposed complex preposition is also marked. However, the hypothesis is supported by the fact that this order is acceptable in (44b'&c).

The acceptability of (44b'&c) thus shows that the cause of the unacceptability of the primed examples in (43) is not that R-pronominalization is impossible; it rather suggests that the sequence of the modifier and the preposition is impenetrable, which would be consistent with a compound analysis; cf. 32.2.1, sub 2. This raises the question as to why (44a&b) are marked/unacceptable. A possible answer might be that these constructions compete with the shorter and thus more economical expressions in (45a&b) without the locational PP. This appeal to blocking would be supported by the acceptability of (44b'&c), given that (45b'&c) are not acceptable.

We conclude from the discussion above that the adverbial modification analysis of the primeless example in (43) is highly problematic, and that the proposed compound analysis fares much better. Note that we have ignored a possible (and quite plausible) alternative analysis of the (a) and (b)-examples in (43)/(44), namely one in which the PP acts as a modifier of the element preceding it. This would explain the acceptability judgments in (43)/(44), as well as those in (45), while preserving our conclusions for the analysis of the (c) and (d)-examples. We refer the reader to Section 34.5 for further discussion of this type of analysis.

There are cases in which the complex forms can be used as intransitive adpositions, while the simple forms cannot. If achter acts as a modifier of the preposition op in (46a), then the unacceptability of (46a') with op as an intransitive adposition would of course be highly surprising. In (46b), we give similar examples with bovenin.

If we are dealing with compounds, on the other hand, the difference in acceptability between the primeless and primed examples could simply be explained by appealing to a difference in syntactic valence of the simple and complex prepositions.

Section 34.1.5 has shown that modified intransitive adpositions cannot permeate the clause-final verb cluster; the complex forms under discussion, on the other hand, can, as is shown in (47).

This again favors a compound analysis of the complex forms; the complex preposition voorop functions as a verbal particle, which can thus permeate the verbal cluster; cf. Section 32.2.4, sub III, for discussion.

The discussion in the previous subsections suggests that elements like links/rechts, midden, voor/achter and boven/onder in (39) and (40) are not adverbial modifiers but the first members of a compound. Table 5 shows the possible combinations of these elements with prepositions that denote the null vector. The fact that there are so many question marks in this table indicates that more research is needed.

It is important to note that the “—” mark does not necessarily mean that the pertinent sequence cannot be found, but rather that the intended modification relation is not present. The examples in (48) do not involve modification of the PP by the elements achter/voor, as will be clear from the paraphrases.

That we are not dealing with modification of the PPs is also supported by the following two facts. First, the PPs need not be present: without them, the examples simply express that the ladder was at the back and that Jan walked (e.g. up and down) in front of the house. Second, the examples in (49) show that voor and achter cannot pied-pipe the PP under topicalization but must be fronted in isolation; this shows that they are independent constituents, comparable to the more common element boven with the meaning “upstairs”; cf. Section 34.5.

The fact that the combinations in Table 6 may occur with a relation other than the modification relation discussed in this subsection may perhaps also account for at least some of the question marks in this table. The fact mentioned earlier that there are grounds for positing that links and rechts do not function as the first member of a compound but rather as independent constituents in their own right, may also be responsible for some of the unclear cases.

The meaning expressed by the putative modifiers in (53) is of a completely different nature than that of PPs denoting a set of vectors. This is most obvious in the case of the putative modifiers of distance. In the case of PPs denoting a set of vectors, these modifiers indicate the distance between the reference object and the located object. In the case of in/uit, on the other hand, the modifier indicates how deep/far the located object penetrates/protrudes from the wall; this is shown in Figure 6A&B. Similarly, in the primed examples, schuin indicates the way (i.e. direction) in which the nail penetrates or protrudes from the wall; this is shown in Figure 6A'&B'.

If the PPs headed by in and uit involve vectors, these should differ from the vectors discussed earlier in that they are not necessarily directed outwards with respect to the reference object, but can also be directed inwards. Note, however, that the nominal measure phrases in (53a&b) above are in a paradigm with gedeeltelijkpartly, helemaalentirely, and voor de helfthalf (lit: for the half), which seem to modify (or be predicated of) the located object rather than modifying the PP, so that it is not a priori clear whether we are really dealing with PP modification.

The elements discussed above can be used not only in the locational constructions in (53), but also in constructions involving a change of location, as in (55). Observe that uit differs from in in that it (marginally) takes the adjectival verfar as its (putative) modifier of distance, not diepdeep.

The unacceptability of the examples in (56) shows that these modifiers can only occur with PPs denoting the null vector if some physical contact between the reference and the located object is implied. This is consistent with the suggestion above that the modifiers in question do not actually modify an (inwardly directed) vector, but the located object itself: if there is no physical contact, the located object cannot penetrate or protrude from the reference object and so modification is excluded. If this is true, the notion of an “inwardly directed vector” can be discarded, and we should conclude that modification of adpositional phrases denoting the null vector is not possible. Example (56b) is given a number sign because it (marginally) allows a reading in which schuin/recht refer to Jan’s posture, which is not relevant here.

Nevertheless, it should be noted that examples like (57), where the distance of the located object from the outer boundary of the reference object seems to be measured, are acceptable. Perhaps this is because these examples do not actually involve multidimensional space. For example, if the budget is one million euros, example (57a) simply expresses that the estimate is less. If the time limit is two hours, (57b) expresses that it took the athlete less time to finish. And if the transmitter has a range of 100 km, (57c) expresses that Jan lives less than 100 km from it; so, only the distance between Jan’s house and the transmitter is relevant in this example.

In other words, we are not dealing with an outer boundary in multidimensional space, but with an upper limit on a (one-dimensional) scale; this means that the PPs in (57) cannot be described in terms of vector theory.

Another reason to doubt that the examples in (53) involve modification of the PPs is that topicalization of the allegedly modified PPs gives rise to a marked result. This is clearest for the examples with the preposition uit in the (b)-examples in (53): the examples in (58) seem unacceptable. This is suspicious, because in cases where we are unambiguously dealing with modification, topicalization of the entire modified PP is easily possible: Een meter boven de deur hangt een schilderijA painting hangs one meter above the door.

Leftward movement of the noun/adjective phrase in isolation gives rise to significantly better results: topicalization, as in the primeless examples in (59), may still be a bit marked, but wh-movement, as in the primed examples, is perfectly acceptable. Unfortunately, however, this does not tell us much, since we have seen that it is also possible to extract nominal and adjectival modifiers from the PP.

Example (60a) appears to be slightly better than (58a). However, the interpretation of this example differs from that of example (53a); it can no longer express the situation in Figure 6A, but suggests that the nail has completely entered the wall and is now located at a distance of three centimeters from the surface of the wall. The other examples in (60) do not receive such a deviant interpretation, and the judgments are more or less similar to those in (58) and (59).

The topicalization data discussed in this subsection at least suggest that example (53) does not involve modification of the PP. We will defer postpone discussion of this issue to Section 34.1.4, sub III, where we will examine the modification of directional adpositional phrases.