Íslenskt mál og almenn málfræði - 01.01.1981, Side 174

172 Ritdómar
syntactic subjects, even though these are deleted or removed by the Equi and
Subject-Raising rules.
But many recent theorists, such as Bresnan (1978) and Gazdar (1981), to name
two, have wished to drastically reduce the role of transformations, or eliminate
them entirely. A natural move to make is to claim that the syntactic represen-
tations of the (b) examples are essentially the same as their surface structures.
On such analyses, the infinitives would be syntactically nothing more than
subjectless VP. The semantic interpretation, however, would work out so that
some matrix NP is understood as if it was the subject of the VP. The semantic
interpretation would furthermore specify that the matrix object in (5)b functions
as an argument to the matrix verb, bearing a specific semantic role, while that
in (6)b doesn’t. This would account for the fact that (5)a, the finite clause para-
phrase of (5)b, has a matrix NP object coreferential to the complement subject,
but (6)a, the main clause paraphrase of (6)b, does not.
Such analyses can be made to work reasonably well for English, although
certain problems remain (see Sag (1980) for an attempted solution to most of
them). But T is able to show that the problems are much worse in Icelandic:
principles of case-marking, agreement, reflexivization, etc., seem to depend on the
underlying structural relationships to such an extent that it is difficult to see how
an adequate theory of sentence structure could avoid the necessity of directly
representing these relationships.
This result about Icelandic does have consequences for the analysis of other
languages, such as English. For we should be reluctant to include in linguistic
theory both the powerful semantic interpretation principles that would be needed
to recover the understood-subject-of relation if it is not represented in the syntax,2
and the powerful syntactic principles that are needed to represent it in syntactic
structure. It is enough of a problem to produce a constrained linguistic theory
containing one or the other of these devices. One would not want to try to produce
one with both unless it were absolutely necessary. Since a syntactic representation
of the understood-subject-of seems consistent with facts of English and necessary
for those of Icelandic, we would want to chose this treatment for both languages.
Non-transformational theories which can easily accomodate T’s data, such as
Relational Grammar, or Lexical-Functional Grammar (Andrews 1981a, 198lb;
Bresnan 1981) succeed precisely because the syntactic structures they assign to
sentences preserve the relevant features of Aspects-style deep structures, such as
‘understood subject-of relations.
Another important result concerns the surface structure of the SOR construction
(6)b. T argues strongly that the postverbal NP in this construction is the surface
matrix object, and not the surface complement subject (though it is the deep
complement subject). Such an analysis has also been widely adopted for the corre-
sponding constructions in English.
But Chomsky (1973, 1981) (and many intervening publications) has argued that
the surface structure for such constructions in English is as in (11):
2 See Dowty, Wall and Peters (1980) for an introduction to these techniques.