434 LINDA SCHWARTZ
objective in Choctaw but subjective in Italian, "sneeze" is subjective in Ita
lian and Dutch but objective in Eastern Pomo and variable in Choctaw,
"bleed" is objective in Turkish but subjective in Italian, etc. (for a fuller
range of comparative data, see Rosen (1984)).
Rosen concludes from this that the Universal Alignment Hypothesis is
invalid (that is, that there is no total cross-linguistic predictability of the
syntactic class of a predicate from the semantics of that predicate). An
implication of this conclusion in Rosen's account is that the syntactic objec
tivity or subjectivity of at least some intransitive verbs (those for which
semantic class does not predict syntactic class) must be specified as syntactic
information. This is done formally within Relational Grammar by specify
ing in the lexicon that a given intransitive predicate takes an initial 1 (=sub
ject) but no initial 2 (=direct object) for subjectives or an initial 2 but no
initial 1 for objectives. Rosen presents this as an argument for an initial syn
tactic stratum which is independent of semantic structure.
Burzio (1981, 1986), working within a Government-Binding
framework, recognizes the subjective-objective distinction in Italian
intransitive verbs and also chooses a syntactic rather than semantic rep
resentation of the distinction. This is done formally by specifying that sub
ject position is or is not a theta-marked position for a given predicate; if its
subject position is theta-marked, then an intransitive predicate will be sub
jective and have only an external argument, and if its subject position is not
theta-marked, then an intransitive predicate will be objective and have only
an internal argument at d-structure. Since it is assumed (Burzio's Generali
zation) that verbs which do not case-mark their objects do not assign a
theta-role to their subjects, the underlying object of an objective predicate
must move to subject position to get case. Again, this in part justifies the
presence of a syntactic stratum (d-structure) which is distinct from both sur
face syntax and Logical Form.
The fact that a given predicate patterns sometimes with the subjective
class and sometimes with the objective class cross-linguistically does not
necessarily mean that a strong hypothesis of semantic-syntactic alignment is
invalid. Rather, it may mean that the alignment parameters may vary lan-
guage-specifically (as argued in Van Valin 1990) or that only one class in
the subjective-objective opposition will be semantically specified, with the
other class functioning as a default (the position taken in Merlan (1985)).
These positions are consistent with a Motivated Alignment Hypothesis,
which would predict that the membership of at least one class of the opposi-