In this article I intend to show that the concept of ‘grafts’ that I have argued for in a number of articles (cf. Van Riemsdijk, 1998b, 2000, 2001a) can be given a natural interpretation in a framework that makes use of internal and external merge. More specifically, I will argue that when all the logical combinations of internal and external merge are taken into account, grafts must exist and are not, as hitherto believed, an odd problem for phrase structure. It follows also that, in addition to my prime examples of grafts (free relatives (FRs) and transparent free relatives (TFRs)), we should expect there to be more constructions with the telltale properties of grafts. One such construction, “Horn Amalgams” will be introduced here in combination with an argument that Grosu’s (2003) attempt at reducing TFRs to regular FRs fails. 1. Introductory remarks1 In a number of articles I have argued that there are constructions in which classical syntactic structures in terms of so-called well-formed trees are insufficient.2 A typical simple example is (1). (1) a far from simple matter Clearly this is a noun phrase. The question is, what is its left-hand modifier? Is it the adjective simple, or is it the adjective far? Semantically, simple seems to be the head, but then what modifies it? Far from is not really a constituent. If far were the head, then it might be thought to have a PP-dependent in the form of from simple. But the latter option is unlikely anyway since adjectival heads in prenominal position must be adjacent to the noun.3 This means we have a paradoxical situation. Under the assumption of grafts, trees can ‘grow together’ in unconventional ways, specifically in sharing terminal elements. Under such an analysis, (1) would consist of two simple trees, one corresponding to a simple matter, the other one representing far from simple. the adjective simple is the shared element, the ‘callus’ in my botanical metaphor. The paradox is avoided by the idea that the root node of the PP is not connected to any position in the host tree (the DP). Even in a simple case like this, interesting evidence exists in favor of an approach like this. Consider the fact that in Dutch attributive adjectives inflect, while predicative adjectives do not. In the equivalent of (1), the adjective is inflected by means of schwa: 1 The ideas presented here have benefited from discussions with the students in my classes at the University of Vienna in June 2004, and from thoughtful comments by Edwin Williams. Thanks are also due to audiences at Tilburg University and at the Incontro di Grammatica Generativa in Rome, February 2005, as well as subsequent comments by Alex Grosu, Viola Schmitt and an anonymous reviewer. Parts of the present article appear in Van Riemsdijk (2004, to appear-b). 2 See Van Riemsdijk (1998b, 2000, 2001a) with antecedents in Nakau (1971), Kajita (1977), Lakoff (1974), and McCawley (1982, 1988). Lakoff coined the term ‘syntactic amalgams for what I call grafts. My own work grew out of a critique of Wilder (1998, 1999). 3 Cf. Williams’ (1982) Head Final Filter, going back to the Surface Recursion Restriction of Emonds (1976:19, Emonds, 1985:131). See also Van Riemsdijk (1998a) for more recent discussion. (2) a. een alles behalve eenvoudig-*(e) oplossing a anything but simple solution b. deze oplossing is alles behalve eenvoudig-(*e) this solution is anything but simple On the graft approach we can say that the uninflected adjective is the shared element, while the inflection marker remains outside the callus. The most compelling evidence for grafts comes from the construction called ‘transparent free relatives’ which I will return to in section 4. Before doing so, however, consider the question of whether the introduction of grafts into the arsenal of descriptive tools in linguistic theory4 does not constitute an excessive extension of power. The point of the present article is to argue that, on the contrary, the existence of grafts follows directly from the logic of external and internal merge, as developed in Chomsky (2001, 2005). 2. The logical necessity of displacement In two recent lectures,5 Chomsky has sharpened the reasoning concerning what used to be called the displacement problem. Until recently, the existence of movement (displacement) was seen as a problem: why does it exist?6 Chomsky now argues that this is a misguided question. Recursive merge is now thought to be the defining property of the human language faculty (cf. Hauser et al., 2002). What Merge does is, it takes two elements, call them A and B, and puts them together. Now, in the simplest case, A and B are both simplex. That would be a case of ‘very first merge’. Apart from this case, however, A and B will generally be syntactic trees that have been formed by previous instances of merge. Consider the following situation. (3) A B
Stored only when both papers are in the corpus — 72 citations exist upstream.