With the order of (2), (8b) can be derived. When C searches, the subject which
man is still within vP, so that T undergoes Internal Merge to C. After that, the
subject which man undergoes Internal Merge to Spec-T (counter-cyclically, see
Epstein, Kitahara and Seely (2012)) and to Spec-C. With the order of (3), on
the other hand, (8a) can be derived. The subject which man moves to Spec-T
before C’s minimal search. In the case of Kisongo Maasai, D ore can be morpho-
phonologically separated from its complement NP cow. However, English D
can never be isolated because of its morphological properties (a form of morpho-
phonological variation) which are analyzed in the process of externalization, so
that D pied-pipes the whole DP for SM-convergence. As a result, DP which
man moves to Spec-C. Thus the exceptional vexing (8a) is simply the Kisongo-
Maasai type of derivation, just like the mysterious tough-construction (i.e.
mysterious only “in English”!) is seen as nothing more than the Kilega type of
derivation (Obata and Epstein 2011, 2012).
The important points worth emphasizing are that the orde rs in (2) and (3)
are ea ch computationally efficient, in that C finds a goal (as required by hypoth-
esis for convergence) and in each derivation the exact same operations are applied
(subject-raising and C minimally searching) and something moves (overtly) to
C by EF. Crucially, then, the theory predicts that there is greater than one
computationally efficient type of derivation: convergence requires that C search
and attract some category, but as far as optimal computation goes, EF on C can
be satisfied by any category moving to C-Spec. EF on T can be similarly satis-
fied. But optimal computation leaves underspecified the ordering of these two
operations, each of which is necessary for convergence. Both orders are com-
putationally efficient. Equally optimal distinct derivations generate predicted
variation. Prediction (the theory of computationally efficient satisfaction of the
interfaces predicts the possibility of two distinct orderings) via underspecification
in third-factor machinery is arguably the furthest we can get from ad hoc stipu-
lation, including the stipulating syntactic parameters as part of UG. That is, if
this approach is viable, intra and inter “language” non-morphological variation
are unified and in part captured by what is NOT stated (in general, third-factor
laws NOT specific to the human language faculty.)
The empirical domain and viability of the application of such an approach remain
to be seen. But the idea that language-specific ‘syntactic phenomena’ (like tough-
constructions “in English” or optional Aux-inversion “in English” matrix subject
wh-movement) might be partially accounted for, by what is NOT stated in laws
NOT specific to the human language faculty, was until very recently inconceivable.
If scientific explanation is one’s goal, this form of third-factor analysis seems very
worthy (in our view) of further exploration, no matter where it might lead.
5 Discussion: what’s an (I-) “language”?
We reviewed some approaches to crosslinguistic variation and fur ther explored
third-factor underspecification as a locus of I-language variation (Richards 2008,
Boeckx 2011, Obata and Epstein 2011, 2012, Obata, Epstein and Baptista
Eliminating parameters from narrow syntax 135