- triggered by ( variant) exte rnalized outputs (Berwic k and Chomsky 2011)
which constitute the input to the learner, based on the Borer-Cho msky conjec-
ture. Under the third-factor underspecification approach, the options of under-
specified rule order are entailed by an independently motivated formulation of
the SMT “computationally efficient satisfaction of the interfaces” without any
ad hoc descriptive stipulations. Based on Obata, Epstein and Baptista (2015),
we further extended the argument in favor of ordering a s a third factor. As a
case study, we discussed Chomsky’s (2013) analysis of Aux-inversion in terms
of Carstens and Shoaff ’s (2014) analysi s of Kisongo Maasai and English matrix
subject wh-moveme nt, and suggested that both T-to-C and D-to-C are equally
possible computationally efficient derivations satisfying the interface conditions,
each of which is derivable by allowable ordering variation entailed by what is
NOT formally stated, within the third-factor underspecification approach. The
theory thus predicts that computationally efficient satisfaction of the interfaces
can be attained by variant rule orderings, giving rise to variant externalized
outputs which serve as inputs to the learner. As is perhaps obvious, ‘construc-
tions’ (or, more primitively, lexical items or their features) in a given I-language
are subject to the exact same constraints which apply ‘across’ I-languages. The
resulting prediction is that any ‘construction’ in a particular language (e.g.
English tough-constructions) are in principle possible constructions (“IN”
another “language”). The particular languages that have existed and do exist
are an accident of many historical factors, but the class of possible (human)
I-languages (our object of inquiry) and hence the class of possible (human)
derivations is a single set of what we might call “human”. Hence it is not
particularly surprising (nor is it necessary) to find that “English tough-construc-
tions are ‘simply’ Kilega”, and “English matrix subject wh-movement lacking
auxiliary inversion is ‘simply’ Kisongo Maasai”. This, of course, further blurs
(if not obliterates) the common sense (non-scientific) notions of languages, such
as English, Kisongo Maasai and Kilega, i.e. if we are on track here, the English
tough-construction ‘is’ Kilega and the English subject wh-movement in matrix
clauses lacking T to C, ‘is’ Kisongo Maasai. Scientifically speaking, the I-language
English is neither a T-to-C “language”; nor is it a D-to-C “language”. Rather,
“it” is both, i.e. my I-language performs both operations, each licensed by third
factor and triggered by input. Similarly, under Obata and Epstein (2011), we
cannot say that English involves subject agreement. Rather in tough-constructi ons,
T agrees with a moved/moving/shifted object first merged in object position.
(i.e. “John IS easy to please” originates as “Be easy to please John”. John then
shifts to embedded C-Spec where matrix T can probe and agree with (what
was a direct object) John. Then John raises ultimately to matrix subject position.
See Obata and Epstein 2011, 2012) Thus “English” is neither a subject agree-
ment nor object agreement “language”. My I-language generate s both subject
agreement derivations and derivations in which a shifted object appearing in
embedded C-Spec is the goal of a higher T probe. Binary parameters of the
form “Subject agreement or Object agreement” seem not only descriptive, but
(descriptively) inadequate. The class of possible human derivations is a single
136 Miki Obata and Samuel Epstein