(1) a. When a nominal is merged with a lexical head, its case feature is valued
as accusative.
b. When a nominal is merged with a phase head (v or n), its case feature
is valued as nominative or genitive.
c. Otherwise, the case feature of a nominal is valued as dative.^2
For example, when the nominal hon ‘books’ is merged with the verb yomu
‘read’ to form {hon, yomu}, the case feature of the nominal is valued as accusa-
tive by (1a). Likewise, when the nominal Taroo is merged with a syntactic object
headed by v, its case feature is valued as nominative by (1b).^3 When a case
feature of the nominal is not valued by either (1a) or (1b), (1c) features as a
last resort to save the nominal that has an unvalued case feature.
While case valuation occurs in terms of external Merge as described in (1)
in non-agreeing languages, Zushi argues, following Chomsky (2001) and sub-
sequent work, that a case feature is valued in terms of Agree in languages that
exhibit φ-feature agreement. Therefore, this proposal suggests two distinct modes
of valuing case features – Agree-based and Merge-based – one of which is chosen
for a particular language.
One question that arises here is what determines the mode of case valua-
tion in a given language. Given that UG has the option of case valuation via
Merge or Agree, why is the Merge-based case valuation not an option in
agreeing languages? Similarly, why is the Agree-based case valuation not an
option in non-agreeing languages? To address these questions, Zushi proposes
that case valuation is conducted by a syntactic operation – either Merge or
Agree – if the operation establishes a covariance relation R, which is defi ned
as follows.
(2) R holds between α and β iff α and β covary with respect to a set of
φ-features F.
(3) α and β covary with respect to F iff either (i) both α and β are specifi ed
with respect to F, or (ii) neither α nor β is specifi ed with respect to F.
From this viewpoint, case valuation is a consequence of the covariance rela-
tion between α and β. The covariance relation holds if and only if either of two
conditions is satisfi ed: both α and β bear a set of φ-features (3i) or neither one
bears any φ-features (3ii). The former condition is tantamount to a standard
agreement relation established by Agree between a probe and a goal, where
both are specifi ed with a set of φ-features. The gist of this proposal is that the
covariance relation also holds between two elements, both of which have no
φ-feature specifi cations, as stated in (3ii). Such a covariance relation is established
by Merge when two elements with no φ-features are introduced into a structure
and combined by Merge. When a covariance relation is constructed in this way,
case valuation becomes available based on Merge. This makes the operation
Agree unnecessary, insofar as a covariance relation (hence, a case valuation) is
concerned. Introducing the notion of covariance relation into the general system
48 Mihoko Zushi