of case valuation opens a novel and natural way to capture case valuation uni-
formly in non-agreeing languages and agreeing languages. This possibility is not
readily available through the standard case system under Agree without stipulat-
ing an otherwise unmotivated (abstract) φ-feature agreement in non-agreeing
languages.
In this proposal, φ-feature specifications of lexical items in a language deter-
mine how a covariance relation is established, which in turn selects a mode of
case valuation in the language. Japanese is among the languages that allow the
covariance relation to be built up by Merge; hence, case valuation occurs through
Merge. Based on the assumption that no φ-features are involved in Japanese
syntax, the covariance relation is established by Merge when it applies to two
elements, such as a verb and an object, since neither is specified with φ-features.^4
Therefore, case valuation takes place at the merger of the two elements, and
the case feature of the object is valued as accusative according to (1a). Similarly,
the merger of a subject and v establishes a covariance relation, since neither has
φ–features. Thus, the case feature of the subject is valued as nominative by (1b).
In this manner, case valuation in this type of language occurs in terms of Merge
that establishes a covariance relation, rendering case valuation in terms of Agree
unnecessary and impossible.
In the Agree-based case system put forth by Chomsky, Agree holds between
a probe with unvalued φ-features and a goal with valued φ-features when certain
locality conditions are met. The assumption underlying this view is that both
the probe and the goal are specified with φ-features, regardless of whether they
are valued. As stated above, this concept is included in (3i); that is, a covariance
relation is established between the probe and the goal. As a consequence, the
case feature of the goal is valued under Agree in agreeing languages.
Why is case valuation in terms of Merge impossible in this type of language?
In other words, why is a covariance relation not established by Merge in this
type of language? Keys to the answers to these questions appear in the feature
specifications of lexical items. Following the general assumption that it is v
rather than a verb that comes with φ-feature specifications, the merger of a
verb and an object does not establish a covariance relation in this type of
language because only the object is specified with F. Therefore, a case feature
of the object cannot be valued at the point of merger with a verb. When a
merger of v occurs that combines it with VP (a verb + an object), a covari-
ance relation is established via Agree, and thus the object receives an accusative
case value.
Note that the unvalued φ-features of v are valued by this Agree relation, and
then deleted in the domain of the vP-phase. Thus, when a subject is merged
with v, its case feature cannot be valued at the merger of the two, since the
subject and v do not covary with respect to F. It follows that the unvalued case
feature of the subject should be valued by Agree in the next phase domain of
C-T. This proposal, given in (2) and (3), together with the properties of
φ-features in the agreeing languages, suggests that Agree is the only option for
case valuation in such agreeing languages.
Case and predicate-argument relations 49