This exposition of the information structure tier has been extremely sketchy. The idea has been only to show its
general properties and their role in the present system. It is again a matter for future research to integrate the sizable
literature in this area into the framework, and to see what advantages or disadvantages may ensue.
12.6 Phrasal semantics and Universal Grammar
We have not talked about the mentalistic basis of semantics much in this chapter. It is time to come back to it, if only
briefly. What parts of phrasal semantics have to be learned, and what parts come from Universal Grammar? Of the
parts that come from UG, which are aspects of more general-purpose capacities, and which are specific to language?
First, it is clear that all these aspects of phrasal meaning are available in all the languages of the world. As has been
mentioned here and there, languages differ in their syntactic strategies for expressing phrasal semantics; but the
organization of what is to be expressed seems universal. The elements of the descriptive, referential, and information
structure tiers see mthe sa me across languages, and the principles of co mbination, especially in the descriptive tier,
see muniversallyavailable.At least on preli minary reflection, the possibility of learning any of thiswouldsee mseverely
limited by the poverty of the stimulus. Thus Ifind it plausible that the basic architecture of conceptual structure is
innate.
Ontheotherhand, conceptualstructureissupposedtobethemedium ofthought—thementalstructurethatlanguage
evolved to express—rather than part of language per se. So it seems reasonable to guess that this architecture is not a
part of Universal Grammar, narrowly construed. Whatispart of Universal Grammar, of course, is the architecture of
the interface components that allow conceptual structures to be expressed in syntactic and phonological structures.
Is there independent evidence for such a conclusion? Consider the descriptive tier. Monkeys, especially in the real mof
social relationships, appear to be working with abstract relations such as kinship and dominance, whose variables are
filled with known individuals (Cheney and Seyfarth 1990). On the other hand, in the absence of language, it is hard to
establish whether they use function-argument relations recursively in the manner pervasive in linguistically expressed
concepts. Turningtothereferentialtier,anyanimal that distinguishes individuals from eachother on a lastingbasis has
concepts with indexical features, hence a referential tier of at least a primordial sort. Again, though, in the absence of
language it is not clear how to test whether for instance they have indexical features embedded under irrealis frames.