|
Abstract:
Two ERP studies (2x20 participants) employing different tasks
(comprehension vs. prosody judgment) were carried out to
investigate prosodic effects on the processing of auditorily
presented sentences. The first two conditions consisted of 48
sentence pairs with different syntactic structures:
(1) [IP1 Peter verspricht Anna #zu arbeiten][IP2 und das Buero
zu putzen.] Peter promises Anna to work / and the office to
clean.
(2) [IP1 Peter verspricht][IP2 Anna #zu entlasten][IP3 und das
Buero zu putzen.] Peter promises / Anna to support / and the
office to clean.
In (1) the NP2 'Anna' becomes the indirect object of verb-1
('verspricht') as verb-2 ('arbeiten') is intransitive. In (2), by
contrast, 'Anna' is required as direct object of the transitive
verb-2 ('entlasten').
These structural differences correspond to distinct prosodic
characteristics: As indicated via bracketing, (1) consists only
of two Intonational Phrases (IP) whereas (2) has to be devided
into three IPs. Exhaustive acoustic measurements of all 48
sentence pairs confirm these differences:
a) Only in (2) a significant pause preceding NP2 is detectable
in the acoustic signal.
b) In (1) the Fundamental Frequency (F0) indicates
accentuation on verb-2 by a rising (L+H*) tonal pattern. In (2),
however, the accentuation (L+H*) is aligned with NP2
('Anna').
c) In (1) the durations of NP1 and verb-1 are dramatically
shorter than in (2), providing very early prosodic cues
concerning the subsequent sentence structure.
These prosodic events in the speech signal covary surprisingly
with (task-independent) polarity changes in the ERP signal
(positive shift) suggesting an ERP correlate of IP boundaries and
immediate encoding of prosodic information.
To test the influence of prosody on first pass parsing, a
third condition was created by cross-splicing the first part of
(2) and the second part of (1) at position '#':
*(3) [IP1 Peter verspricht][IP2 Anna #zu arbeiten][IP3 und das
Buero zu putzen.] Peter promises / Anna to work / and the office
to clean.
If the prosodic cues are immediately used to guide parsing,
the resulting false boundary preceding 'Anna' should prevent the
required attachment of 'Anna' to verb-1. Instead, 'Anna' should
initially be attached to the second intransitive verb (Early
Closure) and cause an argument structure violation (Garden-Path
effect). In fact, verb-2 in (3) elicited the predicted N400-P600
pattern in the ERP, which was modulated by task differences.
Combined performance and ERP data of both studies support the
view that subjects reliably detected the prosodic mismatch in (3)
and then reanalysed the sentence according to syntactic
constraints.
|
