A further generalization of the analysis extends it to verbs expressing propositional relations: at the proper degree of
abstraction,entailandimplyare analogous to verbs of causation,permitandbe consistent withare analogous to verbs of
letting, andreinforceandsupportare analogous to verbs of helping. These patterns of commonality are recognized as
significant by many lexical semanticists. But we cannot pull them out without resorting to considerable sublexical
formal abstraction, not readily expressible as definitions. And once we do so, are we down to primitives? We don't
know. All we know is that we are three stages of generalization away fro mactual words.
It is all well and good to extract significant parts fro mword meanings; but what about the part that is left? For
example, suppose we accept, even approximately, McCawley's (1968) famous analysis ofkillas [CAUSE[BECOME
[NOT[ALIVE]]]] (where I use words written in small capitals to designate the meanings of those words). Thefirst three
elements are widespread among lexical concepts and are candidates for primitives or near-primitives. But what about
ALIVE? Can it be decomposed, or do we have to accept it as a primitive? Laurence and Margolis (1999) call this the
“proble mofcompleters.”
This proble malso arises in analyzing variation a mong closely related words. For instance, how doesbreakdiffer from
shatterandcrumble! It does not appear likely that the differences among them are elements of any larger generality, such
that we wouldfind other groups of verbs that differ in exactly the same dimensions. Such facts, which we confront at
every turn, threaten tounderminethe prospect of completely decomposing wordsintoprimitives thatare descriptively
useful and that have some plausibility for innateness.
For some linguists this is not taken to be a problem. Pinker (1989: 167):“I will not try to come up with a small set of
primitives and relations out of which one can compose definitions[sic]capturing the totality of a verb's meaning.”
Grimshaw (1993):“Linguistically speaking, pairs like[breakandshatter]are synonyms, because they have the same
structure.The differences betweenthe mare notvisible to thelanguage.”Pustejovsky (1995:58):“What I wouldlike to
do is to propose a new way of viewing decomposition, looking more at the generative orcompositionalaspects of lexical
semantics, rather than decomposition into a specifiednumberof primitives.”For Pinker and Grimshaw, all that is at
issue is the syntactic behavior of words; since the syntax-semantics interface cannot“see”all of meaning, a complete
decomposition is unnecessary and even irrelevant (section 9.7.3). Pustejovsky is interested in previously unexplored
aspects of lexical meaning that go beyond mere feature decomposition (see sections 11.9 and 11.10 ), so he too can
bypass the question of primitives.
However, the requirement of learnability forces us to take the problem of primitives seriously. If children can acquire
thefine-scale differences among