Architecture and Experience in Sentence Processing:
 the Source of Lexical Effects in Parsing  Reduced Relative Clauses

			        Paola Merlo                   
	            U. of Geneva/U. of Pennsylvania        


Models of the human sentence processing mechanism have traditionally
appealed to innate architectural restrictions to explain observed
patterns of behavior.  Recently, a number of proposals have instead
emphasized the role of linguistic experience in guiding sentence
interpretation, suggesting that various frequency measures play a
crucial role in ambiguity resolution.  What has been lacking thus far
is a detailed analysis of the linguistic and computational properties
that could explain why those particular aspects of experience are
effective in shaping behavior. 

We explore these issues in the investigation of potential predictors for
resolving the main verb/reduced relative (MV/RR) ambiguity.  The classic
example of processing difficulty has long been the famous sentence `The horse
raced past the barn fell' (Bever, 1970), which contains a main verb/reduced
relative (MV/RR) ambiguity at the verb `raced'.  Recently it has been observed
that the difficulty of the MV/RR ambiguity depends on the particular verb in
the construction.  A number of proposals assert that this differential degree
of difficulty crucially depends on verb-specific frequency data.  We instead
propose an account in which the lexical-semantics of the verb rather than
frequency is the critical factor.

In support of this view, we present acceptability judgments and corpus
data demonstrating that the frequencies previously proposed as factors
in resolving the MV/RR ambiguity do not match the pattern of human
preferences.  The data further reveals consistent behavior across
lexical semantic verb classes.  Based on a detailed linguistic
analysis of these classes, we propose a different frequency factor as
the crucial environmental variable.  There are two corollary results to
this finding.  First, the environmental factor that we propose models
the preference behavior with a single frequency value, eliminating the
need for a complex model of frequency combination.  Second, the
feature whose count best matches the data is one that, surprisingly,
is not itself an element of the ambiguous linguistic construction, and
whose potential relevance is only revealed by a deeper analysis of the
lexical and syntactic properties of verbs.


This is joint work with Dr. Suzanne Stevenson, Rutgers University.