1 On tough-movement* Milan Rezac, University ... - Multimania.co.uk

Agree now plays the role in triggering Predicate Abstraction that for Heim and Kratzer
(1998) is played by Move (13): it yields a syntactic representation where the index of the goal of
Agree is represented at the top of the constituent whose head triggers Agree.
It turns out to be further preferable to formulate Predicate Abstraction so that it is triggered
by the interpretable index feature of the non-thematic DP in [Spec, HP] as in (20), rather than by
the value index feature itself. This lets expletives be ignored if by hypothesis they lack
interpretable index features, correctly allows non-thematic non-agreeing DPs to freely λ-bind as
in (12)b, and obviates the problem that Agree-valued index features will been deleted prior to the
application of Predicate Abstraction (see Rezac 2004a:152ff., 2004b for discussion). 8
(20) Predicate Abstraction Rule (PA) (modified): Let α be a tree dominating two sub-trees, β
and γ, such that β has an index feature [ix=i]; then for any variable assignment a, [[ α ]] a =
λx ∈ D e .[[ γ ]] a[x/i] (β).
However, this approach requires a mechanism to ensure specifier-head agreement between
the derived predicate H', whose head H has the Agree-valued φ-probe identifying the variable to
be λ-bound within H', and the DP freely Merged into [Spec, HP] which is to λ-bind it. I discuss
the various possibilities in Rezac (2004b); here it suffices that a mechanism exists which results
in (21), both for the present case and for independent reasons.
(21) Match Condition: if Merge(α, β), then for any φ-feature F, the value of F on the label of α
and the label of β do not differ.
(22) illustrates how the system works in the case of movement, now a descriptive label only
for an instance of the CTM. The mapping label Syntax is the sequence of the independent
operations Agree, Merge, Trace Conversion. At interpretation, assume for concreteness that the
seventh element in the variable assignment function a, that is a(7), is Kate.
(22)
[T [φ-] [ VP left Kate [φ=3sg,ix=7]]]
→ Syntax [Kate [φ=3sg,ix=7] [T [φ=3sg,ix=7] [ VP left [Kate [φ=3sg,ix=7] identical.to x 7 ]]]]
→ Feature deletion [Kate [φ=3sg,ix=7] [T [ VP left [Kate [φ=3sg,ix=7] identical.to x 7 ]]]]
→ PA [[ Kate [φ=3sg,ix=7]]] [λx.[ [ VP left [Kate [φ=3sg,ix=7] identical.to x 7 ]] ]] a[x/7] ]
→ Lexical entries Kate λx.[[λy.y left] [Kate identical.to x]] a[x/7]
→ FA
Kate λx.[ Kate identical.to x left]] a[x/7]
→ FA
Kate identical.to Kate left
= Kate left
Exactly the same mechanics apply in copy-raising as in movement.
This system ensures that all DPs base-generated in a non-thematic position as sister to a φ-
Agreeing predicate show the core property of movement and copy-raising: the goal to which the
DP is interpretively linked is identified by φ-Agree.
3.2 Syntax-semantics mapping in tough-movement
8