However, the relationbetween the nouns is not totally free: whilesnowmanmight have meant a man who shovels away
snowor whomakes snow at a ski area, it is notlikelytohavemeanta man whose sister once fellinthesnow. Thus the
situation resembles the possible meaning relations conveyedby raw concatenation. Jackendoff (in preparation), based
on earlier work such as Lees (1960), Levi (1978), Gleitman and Gleitman (1970), and Downing (1977),finds a
repertoire of perhaps twenty relations that can be conveyed in a compound through pragmatics alone—though the
reasonfor thisparticular set ofrelations remainsfor themomentunclear. These relationsare enrichedand constrained
by the meanings of the particular words in the compound, along lines suggested by Pustejovsky (1995). It is this
reliance on pragmatics and the details of word meaning that has made the analysis of compounds resistant to standard
analytic techniques of generative grammar.
Thousands of compounds with partially idiosyncratic meanings are stored in long-term memory. But in addition, one
constantlyencounters novelexamples suchashealth management cost containment servicesandtwo-axle diesel electric engine dump
truck(examples from the BostonGlobe), whose meanings can be computed on the spot. Thus this is a productive
concatenative syste minvolving only words. As observed by, for instance, Sadock (1998), this syste mis an entirely
differentsortofcombinationthanother forms ofmorphology. Kleinand Perduereportthatnouncompoundingisthe
only kind of morphology found in the Basic Variety; and children improvise compounds very early (Clark etal. 1985).
The facts of compounding thus seem symptomatic of protolinguistic“fossils”: the grammatical principle involved is
simplyoneof concatenating twonouns intoa bigger noun, and thesemanticrelationbetweenthem is determined by a
combination of pragmatics and memorization. Determining the meaning of a newly encountered compound is hence
muchlikedeterminingthemeaningofhit tree Freddiscussedabove—oneuses theHead Principle,plustherepertoireof
possible semantic relations, plus a dose of pragmatics, to put together a meaning that makes sense in context.
Whatever the particular details of these sorts of principle that map between semantic roles and pure linear order, they
sharpen communication. They are therefore a plausible step between unregulated concatenation and full syntax. In
fact, unregulated concatenation need not necessarily have preceded the appearance of these principles: the evidence in
modern language is scant, and only possibly the case of Nim shows us raw concatenation without semantically based
ordering principles. Notably, the free utterances of the bonobo Kanzi see mto show so me li mited use of se mantically
based word order (Savage-Rumbaugh et al. 1998; but see Kako 1999 for a less positive assessment).