Thomas R. Sawallis
IASCP, 63 Dauer, Univ. of Florida, Gainesville, FL 32611
Previous research [T. R. Sawallis, ICSLP 92 Proc. 2, 1067--1073 (1992)] has shown the feasibility of using techniques from signal detection theory to quantify perceptual weights of acoustic cues. Their abstract nature makes these weights autonomous and generalizable across cues and conditions, unlike trading relations which use proportions to bind physical scales. These weights are also environmentally valid because the experimental design avoids multiply-edited ``prototypes'' in favor of multiple singly edited natural tokens. This preserves the behavior of the nontarget cues against which the experimental cues are balanced. Using the French intervocalic /t-d/ contrast, this measure has been shown to be influenced by the listener's linguistic background (both native language and linguistic training) [T. R. Sawallis, J. Acoust. Soc. Am. 93, 2390(A) (1993)]. However, these results reflected only the subjects' first sitting of the perceptual test. This paper reports the influence of multiple sittings of the test on both accuracy (i.e., error rates for edited and unedited tokens) and perceptual weight (i.e., sensitivity to the edited cue).