Decoupling pitch range and peak delay in the L+H pitch accent

Kenneth Konopka, Northwestern University

 	This study investigates the individual contribution of two cues to the 
perception of meaning conveyed by the rise-fall-rise contour.  By presenting 
subjects with stimuli phrases with manipulated contours that varied in peak 
delay and pitch range, the study confirmed that these cues combine to convey 
meaning, but also, that the two cues make independent semantic contributions. 
The data indicate that height variation of the early peak did not affect 
meaning determinations, while a reduction in the late peak height caused 
subjects to fail in making meaning distinctions.
 	Earlier research has established that intonation patterns convey 
meaning, but the nature of intonational meaning and the specifics of its 
realization in the intonational contour are areas of active research.  One 
important context independent contour that has been studied is the 
rise-fall-rise (Ward and Hirschberg, 1985, Pierrehumbert and Hirschberg, 1990) 
which is related to perceived meaning of an utterance.  A subsequent study by 
Hirschberg and Ward (1992) additionally showed the central role of pitch range 
in conveying meaning.
 	The current study expands on the work of Beckman and Pierrehumbert (in 
preparation) who found that two cues appeared to account for the high success 
rate (90%) of subjects capturing a "confirming" versus "counter" meaning: 
intonation peak delay and intensity.  An early/low peak produced a "confirming" 
response, while a late/high peak produced a "counter" response.  While both 
cues apparently worked together to produce meaning distinctions, their separate 
contributions were not assessed.
Subjects made meaning judgments on pairs of utterances that were identical in 
their segments, but differed in intonation patterns (having been extracted from 
different discourses).  The current study used naturally produced stimuli along 
with their contour-manipulated (hybrid) counterparts.  By reversing the natural 
pitch displacements, hybrid stimuli allowed assessment of individual 
contributions of pitch range and peak delay.  Forty-two subjects responded to 
40 short dialogues in which speaker A makes an assertion, and speaker B, 
accepting the truth of the assertion responds with a confirming or counter 
response.  For example, speaker A asserts "Only crazy people study prosody." to 
which two response recordings were made:

Confirming: "Indeed, [Melanie studies prosody], and she's crazy".
Counter: "Really? [Melanie studies prosody], and she's not crazy".

After hearing the assertion, subjects heard only the extract [in brackets] from 
one of these two responses, deciding whether it was confirming or counter. 
Natural and hybridized recordings were included, with the contour for "Melanie" 
used for manipulations.
 	The results show that an early/low or early/high peak elicits a strong 
confirming response (.73 and .69 respectively) while a late/high peak elicits a 
counter response (.83).  Subjects did not make a clear distinction in meaning 
for the late/low peak (counter at .56).  This indicates that these cues are 
independent yet interactive components, a result which is not apparent when 
they coincide in natural utterances. Subjects apparently required a larger 
pitch range in the late peak to make a clear choice on the more marked 
interpretation.