hypothesis that grammatical functions were perhaps the latest-developing part of the architecture—an extra bit of
power for efficiently regulating the syntactic expression of semantic relations. Deviating from practice in Lexical-
Functional Grammar and Relational Grammar, section 5.10 proposed that this tier applies basically only to NP
arguments, and that oblique arguments (i.e. arguments introduced by prepositions), adjuncts, and“adverbials”such as
those illustrated in (2) and (3) above are“invisible”to it.
In section 5.10, the idea of such a limited system was justified by analogy with phonological tiers. Now a more
powerful analogy suggests itself. Phrasal syntax regulates only a limited aspect of the connection between phonology
and semantics, while leaving other, evolutionarilyprior aspects(including theall-importantword-to-meaningmapping)
to the“earlier”system. We can see grammatical functions in a similar light: the syntax–semantics interface is partially
regulated by the supervening, more“advanced”syste mof gra m matical functions; but the“earlier”syste mis left in
place, revealing itself around the corners of the“new”system.
The architecture now comes out like Fig. 8.5. It is not elegant. But then again, this is what we have come to expect of
the brain. It hardly matches the connectivity of the visual system in its complexity.
8.12 Universal Grammar as a toolkit again
Suppose we take Fig. 8.5 as a sketch of“architectural universals”of language. How much of it is actually universal?
At rock bottom, the open vocabulary, phonology, and word concatenation are surely universal. But then we start
running into exceptions. Some Australian languages, for example Warlpiri, Jiwarli, and Wambaya (Hale 1983;
Nordlinger 1998; Austin 2001) show no consistent evidence of phrase structure. Their word order is almost entirely
free, and adjectival modifiers are routinely separated
Fig. 8.5 Modern architecture