| |
Abstract:
Bader (1998) and Fodor (1998) have proposed that prosodic
forces apply even in silent reading. For first-pass parsing we
formulate this as the Implicit Prosody Hypothesis (IPH):
IPH: In silent reading, a default prosodic contour is projected,
and it may influence ambiguity resolution. Other things being
equal, the parser prefers the analysis associated with the most
natural prosodic contour for the construction.
There are results bearing on Fodor's Same-Size-Sister principle,
which concerns preferred positions of prosodic phrase boundaries
depending on constituent lengths. RC-length effects on attachment
choices in reading are reported by Pynte (1998), Fernandez &
Bradley (1999), Hirose (1999), Walter et al. (1999). But if the IPH
is true, not only prosodic phrasing but also (other) differences in
intonation contours could affect silent reading. Bader demonstrated
this for garden-path reanalysis. We make a case for it in
first-pass parsing of RC-attachment ambiguities in Standard Arabic
and French.
French shows a clear high-attachment tendency (for long RCs;
Pynte & Colonna, 1999). Arabic judgments show mild low
attachment (Abdelghany & Fodor, 1999). Thus the IPH predicts
that the prosody of French complex NPs is such as to favor
attachment to N1, while the prosody of Arabic is of a kind more
likely to favor attachment to N2. Since prosodic principles are not
universal, it is a matter of speculation what these favoring
contours would be. We have borrowed an idea from a study of English
by Schafer et al. (1996), and the results suggest this measure is
not inappropriate. Schafer et al. (also Maynell, 1999) manipulated
the relative prominence of N1 and N2 in explicit prosody in complex
NPs. They found more high RC-attachment (to N1) when N1 had a H*
pitch accent than when N2 did. Based on this work, we examined the
F0 values of N1 and N2 in sentences disambiguated for high/low
attachment, read by naïve native speakers, first silently for
comprehension, then aloud.
Testing the IPH requires establishing the 'unmarked' prosody for
a construction. This is not an easy matter, particularly for
ambiguous constructions. We were helped by the discovery that in
both languages, the F0 difference between N1 and N2 is in the same
direction whether the sentence is disambiguated to high or low
attachment. Though data analysis is not complete, we have found
highly consistent effects. In French there is a mean drop of 35 Hz
from the accented vowel of N1 to that of N2. In Arabic there is a
mean rise of just 3 Hz. In Arabic the slight rise held across
long/short RCs, and high/low attachment. French showed a 10 Hz
effect of length (40 vs 30) and a 21 Hz effect of high/low
attachment (45 vs 24). It appears that each language has its
preferred contour (falling or flat), relatively resistant to
meaning. If this preferred contour is assigned to ambiguous
sentences in silent reading, then the difference in N1 salience
across languages could explain their different attachment
tendencies. It will be of interest to test more languages in
future, to determine how much of the cross-linguistic variance can
be accounted for on this basis.
|
