Relational vocabulary plays an important role in thought. It has been argued (Dennett 1991; Jackendoff 1996b, among
many others) that language enhances thoughts by making them available as perceptual objects (namely sentences), so
thattheycanbeattended to, focused on, modified, and remembered. Upontheinventionofthis“utilityvocabulary,”it
would all of a sudden be possible explicitly to wonder ifpand suppose thatp, and to give reasons and purposes for
actions, with a tremendous effect on the power of individual and communal reason and planning. (“WhatshouldIsay
toso-and-so?Ifhesaysthis,then maybeI'lldothat;but if....”Trytoperfor mthisreasoningwithouttheitalicized words.)
Suppose we add phrase structure and all this utility vocabulary to a protolanguage. We still don't yet have modern
language. In particular, there is no notion of subject and object of a sentence—only semantically defined notions like
Agentand Patient.There is no grammatical differentiationof parts of speech, only Objectwordsversus Actionwords.
There is no inflection for case and agreement, and no use of pronouns and other proforms. There is no regimented
way of constructinglong-distancedependenciessuch as therelationof awh-phiase at thefront of a clause tothe“gap”
or“trace”elsewhere within the clause, as in[Which bananas]ido you think Fred advised us to buy tifor our soup?
However, we are movingtowards something thatbegins tobe recognizable as modern language. In particular, wefind
“fossils”of this stage in the very productive syste mof“sentential adverbials”of various syntactic categories that
appear freely at the beginning of the sentence, after the subject, or at the end. Consider the examples in (2).