University of North Carolina at Chapel Hill
Research on children's verb learning strategies has shown that children use
information from a verb's subcategorization frames in narrowing down the possible
meanings of a verb (Gleitman 1990, i.a.). For example, the verb in (1) might
mean something like 'eat' or 'buy', but is unlikely to mean 'inhale' and could
not mean 'sleep'.
(1) Mary
gorped an apple.
This strategy is a good one: it exploits the strong regularities in the syntax-semantics
mapping that fall out from the Theta criterion. Adults make use of this information,
too. Given sentences in which all content words are changed to nonce forms,
adults can make quite specific and consistent predictions
about what a novel verb could or could not mean, based only on the syntactic
frame (whether it is transitive, intransitive, ditransitive, takes a sentential
complement, etc.) (Kako 1999).
In this talk, I
explore the problem of learning to define a class of verbs that do not select
any NP arguments, namely, the class of so-called "raising verbs" (e.g.
seem, appear, tend). Thus, in a sentence like (2), the NP subject, Mary,
is not an argument of the main verb, seem.
(2) Mary seems to be happy.
What makes the problem more interesting is that there is another class of verbs,
control verbs, which also happen to occur in this sentential environment, as
in (3).
(3) Mary wants to be happy.
Unlike (2), the subject of (3) is an argument of the main verb, want. Three questions arise. First, what strategy does a learner use to figure out that the subject of (3) is selected by the main verb, but the subject of (2) is not? Second, how does the learner determine the respective structures of these sentences, given only string (linear) input? Third, do children actually use these strategies? I will present data from both children and adults that bear on these questions. I'll argue that 1) multiple cues are necessary for distinguishing raising from control verbs (hearing a verb with an expletive is not sufficient), and 2) pilot data suggest that children may not begin by assuming a control structure for a string such as (4).
(4) Mary gorps
to be happy.
This second claim raises an interesting learnability puzzle and appears to contradict
the predictions made by Borer & Wexler (1987) and Frank (1998).
If you are interested in
meeting with the speaker, please contact Kaori
Ozawa.
To go back to the Colloquium schedule, click
here.