Section 34.1.3, sub IB, has discussed examples such as (105a) and argued that achterop is a compound. The main reason is that after R-pronominalization the R-word must precede achterop, as shown in example (105b). The fact that it cannot intervene between achter and op would be unexpected if achter acts as a premodifier of the PP op de auto or if the PP op de auto is a complement of the preposition achter; cf. the discussion in Section 32.2.1, sub 2.

What we have not discussed so far, however, is that there are seemingly similar constructions that behave slightly differently. This will become clear when we look at the examples in (106). Although at first sight the clause structures of (105a) and (106a) seem to be completely parallel, R-pronominalization is excluded in the latter case. Note that the number sign is added to (106b) to express that this order is acceptable in the unrelated reading “there is a garbage can on the back of it (e.g. the cover of a book)”, but in this reading (106b) is clearly not related to (106a).

For a proper understanding of the difference between (105a) and (106a), one should know that adpositions like achter, voor, boven and beneden can also be used as referring expressions: in a domestic situation, achter can refer to the rooms at the back of the house, the backyard, etc.; voor can refer to the rooms at the front of the house, the driveway, etc. and boven and beneden can be translated as upstairs and downstairs, respectively. The semantics of the construction in (106a) seems to be that achter refers to a certain part of the premises and that the PP op de plaats further specifies the intended place, i.e. that the PP functions as a kind of modifier of the referring expression achter.

The difference between the two constructions can be clarified by the ambiguous example in (107a). On the compound reading of bovenop, the further sentence expresses that a new shirt lies on top of the closet. On the postmodification reading, the sentence expresses that the shirt lies on the closet upstairs. As would be expected from the data in (105) and (106), R-pronominalization is only compatible with the compound reading: (107b) can only mean that the shirt lies on top of the closet. Another difference between the two readings is that it seems that the phrase boven op de kast can be split in its modification reading, but not (for obvious reasons) in the compound reading; example (107c) is therefore only compatible with the modification reading. Note in passing that the question mark in the comment in brackets is used to indicate that modification may actually be the wrong technical term (as we will show below), but we will use it a bit longer for convenience.

The ambiguity that arises with boven is due to the fact that it can be used both as the first member of the compound bovenop and as a referring expression meaning “upstairs”. A similar ambiguity does not arise with onder, because it cannot express the meaning “downstairs”; as a result, example (108c) is uninterpretable with onder.

The meaning “downstairs” is expressed by the form beneden; Section 32.2.2 has shown that beneden differs from boven in that its use as a preposition is limited to a small number of collocations. The contrast in acceptability between (107b) and (109b) further shows that it also differs from boven in that it cannot occur as the first member of a complex PP.

As mentioned above, the technical notion of modification may be the wrong term to characterize the relation between bovenupstairs/benedendownstairs and the locational PP following it: the latter is not a syntactic modifier of the former. The main reason for this claim is that the two need not be adjacent in the examples in (110). This suggests that we are dealing with separate clausal constituents: boven/beneden simply functions as a locational adverbial phrase, while the PP in de kast functions as a locational complementive. This analysis can be supported by the fact that the adverbial phrase can be replaced by a locational PP such as in je kamerin your room.

If the proposed analysis of (110) is correct, this may raise the question as to whether (107a) and (108a) are indeed ambiguous in the sense suggested above: if beneden does function as an independent clausal constituent, an analysis of (108a) in which the PP functions as a modifier of beneden seems ruled out, and the same should therefore apply mutatis mutandis to (107a). This would lead to an analysis, in which the PP is actually an epenthetic phrase; such an analysis can be supported by the fact that the PP can easily be preceded and followed by an intonation break, which are indicated by the commas in (111).

Finally, for the sake of completeness, it should be noted that it is also possible for boven/beneden to follow the PP, as in (112a). In these examples, boven/beneden clearly modifies the nominal complement of the preposition: we are dealing with the closet, which is located on a higher/lower floor. That this kind of modification is possible can be supported by (112b). These examples are therefore not relevant for our present discussion.

We conclude by briefly considering another possible case of ambiguity, which can be found in (113a). On one reading, this example seems to express that the plane is flying high, with the PP boven de wolken acting as an (epenthetic) modifier of the adjective hooghigh, specifying more precisely the position of the plane. The alternative reading expresses that the plane is flying above the clouds, with the adjective hoog acting as a premodifier, specifying the distance between the plane and the clouds. The first reading seems similar to the modification reading that we discussed above for bovenupstairs and benedendownstairs, and the second reading is similar to that of constructions such as 10 km boven de wolken10 km above the clouds, in which a nominal measure phrase is used as a premodifier of the PP. The fact illustrated in (113b&c) that both hoog and boven de wolken can be used as the predicative part of the construction is consistent with this account of the observed ambiguity.

The facts in (114b&c) suggest that (114a) is ambiguous in a way similar to (113a), although in this case it does not seem readily possible to identify a corresponding semantic difference.