world—and thewordsused todescribeit—witha host ofbuilt-inbiases that constraintheir f-hypotheses about whata
word can pertain to.
Most of the experimental work cited above pertains to words for objects, with subsidiary emphases on words for
substances (milk), properties (heavy), and locations (on, under). A whole other area of research, impinging more closely
on thelearning ofgrammar, concerns thelearning ofverbs, whichexpress actionsand states. Differentkinds of action
involvedifferentnumbersand typesofcharacter. Forinstance, sleeping and sneezinginvolveonlyonecharacter;eating
requires an eater and something eaten; giving requires a giver, a recipient, and something given. This has implications
for thecorrespondingverb'ssyntacticbehavior. Forinstance,sleepandsneezeareintransitive,theirsinglecharacterbeing
expressed as the verb's subject.Eatis transitive, the thing eaten being expressed as direct object—or omitted.Giveis
ditransitive:therecipientand thing givenare expressed as indirectobjectand directobjectrespectively—or as objectof
toand directobjectrespectively. (The principlesbehind these correspondences willbediscussed in sections 5.8 and 5.9.
) So the question arises: To what extentdo childrenfigure out a verb's meaning and then its syntactic possibilities, and
to what extent do they have to hear its syntactic possibilities—including various alternatives as in the cases above—in
order to decide what it means? A fascinating and intricate discussion in the literature (e.g. Landau and Gleitman 1985;
Pinker 1989; 1994a; Fisher et al. 1994; Grimshaw 1994;Brent 1994; Steedman 1994; Tomasello and Merriman 1995;
Gillette et al. 1999) detects influences in both directions. However the answer works out, one has to assume that the
child has certain biases as to what to look for.
This is only the semantic side of word learning. As Macnamara (1982) points out, the child must also f-identify the
spoken word as a significant perceptual object, in order to have something to link a meaning to. Here too there is a
significant body of research (e.g. Cutler 1994; Jusczyk 1997). Statistical regularities in syllabic structure (Saffran et al.
1996; Kuhl 2000) undoubtedly play a role in the child's determining which strings of sounds are words at all. More
generally, the connectionist tradition (Rumelhart and McClelland 1986a, b; Elman et al. 1996) has shown that rather
elementary statistical procedures can lead to much more sophisticated behavior and learning than could have been
imagined in 1965. As pointed out in Chapter 3, such procedures do not come to terms with the all-important
combinatorialityof language. However, we will return in Chapter 6 to points where this sort of learning proves useful.
The discussion in this section can hardly do justice to what has by now become a vast tradition of research. I allude to
this tradition only to make clear how rich, varied, and difficult the questions of word learning are. Again we meet