A Puzzle about P-Stranding and a Possible Solution

In English, Ā-movement operations that move elements to the left are able to strand prepositions—(1)—but those that move elements to the right—(3b)—cannot, as observed by Ross (1967).

1. 1.

2. 2.

   1. a. John saw [the man who lived next door] in the living room yesterday.

   2. b. John saw t₁ in the living room yesterday [the man who lived next door]₁.

3. 3.

   1. a. John looked at [the man who lived next door] in the living room yesterday.

   2. b. *John looked at t₁ in the living room yesterday [the man who lived next door]₁.

            cf. John looked in the living room yesterday at [the man who lived next door].

Why the difference? Bresnan (1976) presented an ingenious account based on the A-over-A Condition (Chomsky 1964). Indicating that "Heavy NP Shift" can apply to PPs as well as to NPs, she formulated the process in terms of [−V], the feature assumed to be shared by N and P. She then observed that the operation of a transformation extracting a heavy NP out of a heavy PP would violate the A-over-A Condition. There is at least one difficulty with this account: rightward movements of NP and PP seem to have different properties, hence are unlikely to be captured by the same transformation (short of Move α, of course, but then all bets are off). In particular, rightward movement of NP does indeed usually require that the NP be (phonetically) heavy.

1. 4.

   1. a. ?*I saw yesterday Jim.

   2. b. ??I saw yesterday Harry.

   3. c.     I saw yesterday Alexander.

But there is no such requirement for PP-movement.

1. 5. I spoke yesterday to him. [or even to'm]

To our knowledge, there exists no current explanation for the directional asymmetry in preposition stranding. We would like to suggest that it can be explained given three currently common assumptions:

1. a. Spell-Out is cyclic (Chomsky 2000, Uriagereka 1999), and the output of the linearization operation that gives an ordering at PF has the form proposed by Fox and Pesetsky (2005). More specifically, linearization applies at every phase, delivering a set of linearization statements of the form "X precedes Y."¹ The union of the sets of these statements must be consistent for the derivation to converge at PF.

2. b. PP is a phase. Languages differ in whether or not intermediate movement to Spec,PP is possible—that is, in whether or not PP has an "escape hatch." P-stranding is possible only when such an escape hatch exists.² This approach to P-stranding goes back to Baltin 1977 and Van Riemsdijk 1978 and has recently been revived in a minimalist setting in Abels 2003.³

3. c. Linearization applies so that P⁰s linearly follow their escape hatches. This seems an empirically correct assumption, which may follow on independent grounds if escape hatches are specifiers. Chomsky (2004:110) notes that (whatever the merits of the strict antisymmetric theory; Kayne 1994) specifiers are typically or always on the left. For current purposes, any linearization procedure that situates PP escape hatches on the left edge of PP will serve.

Given (a)–(c), the asymmetry noted above follows trivially. In order to "escape" a PP, a DP must move via Spec,PP. However, as PP is a linearization domain, this yields a linearization where the DP precedes the preposition. This blocks any operation that subsequently moves the DP to the right in another linearization domain, since the resulting set of linearization statements would be inconsistent. Movement to the left suffers no such fate for obvious reasons. The relevant structure at the point of Spell-Out of the PP is illustrated in (6).

1. 6. … [_PP DP₁ [_P′ P t₁]] …

The same logic seems to apply to recent antisymmetric approaches to rightward movement (Kayne 2000:250), where apparent rightward movement of XP is taken to be leftward movement of XP followed by remnant movement of a constituent immediately (or very nearly immediately) below the landing site of XP. Once Spell-Out has applied at the point in the derivation...