| |
Abstract:
Prosodic boundaries can bias or disambiguate the
interpretation of constituent structure ambiguities. Four studies
explored the processing of ambiguous sentences like Susie learned
that Bill telephoned after John visited to determine why particular
prosodic boundaries have the effects they do. They manipulated the
prosodic boundary between the matrix clause and its complement and
the boundary before the adjunct clause after John visited. They
disconfirm an account where only the local boundary before the
adjunct (the 'late boundary') is important. Instead they support
the use of global prosodic context, crucially the relative size of
the local boundary before the adjunct and the boundary preceding
the complement clause (that Bill telephoned), the 'early boundary.'
Consider an account in which only the acoustic magnitude of the
boundary preceding the adjunct clause matters. On this hypothesis,
a physically larger late boundary should increase high attachment
interpretations. Experiment 1 manipulated the acoustically-defined
size of the break before the adjunct. Results disconfirmed this
hypothesis. Experiments 2-3 tested an account in which only the
phonological category of the late boundary matters. On this
account, an Intonational Phrase (IP) late boundary should result in
more high attachments than an ip (lacking a boundary tone) or no
boundary, regardless of the type of the early boundary. Experiment
2 tested three conditions that had early IP boundaries and IP, ip,
or no late boundary, and one condition that had ip boundaries at
both positions. The IP-ip and IP-0 conditions, in which the late
boundary was phonologically smaller than the early boundary, showed
significantly fewer high attachments than the IP-IP and ip-ip
conditions, in which the boundaries were equal at the two
positions. The identity of the late boundary did not matter except
in relation to the early boundary. Experiment 3 compared conditions
with a larger late boundary than early boundary to a condition with
equal ip boundaries at both positions. High attachments were less
frequent in the condition with equal boundaries than in the three
conditions with larger late boundaries (ip-IP, 0-IP, 0-ip). These
results disconfirm the claim that IP or ip boundaries before the
adjunct clause will have invariant effects on interpretation, and
thus that this local boundary is the only important factor.
Experiment 4 reinforced these conclusions.
By itself, the presence or size of a particular prosodic
boundary before the adjunct clause did not affect interpretation.
Instead, the interpretation of this local prosodic boundary
depended on its phonological size relative to other prosodic
boundaries in the sentence. If the largest prosodic boundary
preceded the adjunct, then high attachments increased. Why should
the relative size of prosodic boundaries in a sentence matter? We
suggest that listeners assume that speakers are rational: they will
not prosodically mark one break as larger than another if they
intend an interpretation where the smaller break corresponds to the
larger syntacticboundary.
|
