Why successive cyclic movement must be feature-driven: An indirect argument from the lack of successive cyclic A-movement

Michelle Sheehan, Anglia Ruskin University

One of the most surprising and robust findings of generative grammar is the existence of successive cyclic movement. In a wide range of unrelated languages, we find evidence for intermediate steps of movement along the paths of long-distance A-bar extraction (see Abels 2012, Georgi 2014, van Urk 2015, 2019, van Urk and Richards 2015). There several different account of how successive cyclic movement through the phase edge works. Often this movement is taken to be feature-driven, by features on phase heads (Chomsky 2000, 2001, Abels 2012), A-bar moved phrases (Bošković 2007), or both (Nunes 2019) or via feature percolation (Neeleman and Van de Koot 2010). More recently, though, the view has become widespread that intermediate steps of movement are not driven by features at all, but rather by the availability of free merge (Chomsky 2005, 2007, 2008, Boekcx 2010) or because of the labelling problem posed by phrasal specifiers (Chomsky 2013, 2015). Even within a relatively small theoretical space, then, there are several competing accounts of the same effect and ideally we would like to identify one of them as correct and discard the rest.

There is some evidence from the signature effects of successive cyclic A-bar movement that it must be feature driven (see especially van Urk 2019). In this talk, however, I will present a distinct indirect argument for this position from restrictions on A-movement. The argument will go as follows. There are contexts where interclausal A-movement is banned, notably under causative and perception verbs:

  • 1. She was made/let/seen leave.

This restriction is found in many languages with ECM causatives of this kind which can be shown to embed phases (e.g., Portuguese, Spanish, Italian, German, Swedish, Danish). The restriction follows from phase theory if we adopt the weaker Phase Impenetrability Condition in Chomsky (2001) whereby the complement of a phase head is spelled out only when the next highest phase head is merged. As shown in (2), this means that the vP <leave she> is spelled out before the matrix T probes:

  • 2. T was made/let/seen [voiceP voice <leave she>].

This analysis only works if A-movement does not have access to phase-edge escape hatches. If this is the case then each instance of A-movement will only ever be able to cross one phase head and the problem with (1) is that two voice phase heads intervene between the argument position and the matrix T. This seems to capture the facts, as we shall see.

Examples like (1) contrast with examples like (3) because, in (3), the presence of T between the two voice phase heads triggers an additional A-movement and this makes further A-movement possible:

  • 3. Shei was made/seen/expected [TP ti to <leave ti> ]

Unlike A-bar movement, then, A-movement does not have access to phase head escape hatches. Successive cyclic A-movement to the extent that it is possible, must be facilitated by the presence of heads with dedicated EPP features.

It follows that successive cyclic A-bar movement must be driven by A-bar-related features and cannot result from a general labelling problem or free merge.