Asymmetries in consonant confusion

Abstract

Both historical sound change and laboratory confusion studies show strong asymmetries of consonant confusions. Historically [ki] commonly changes to [t∫i] (e.g., English chill, cognates with cool), but not the reverse. Similarly, Winitiz et al. [J. Acoust. Soc. Am. 51, 1309–1317 (1972)] in a consonant confusion study, found [ki] confused with [ti] more often than the reverse. It is hypothesized that such asymmetries arise when two sounds are acoustically similar except for one or more differentiating cues, which cues are subject to a highly directional perceptual error. For example, if soundx possesses a cue that y lacks, listeners are more likely to miss that cue than introduce it spuriously. /k/ and /t/ before /i/ have similar formant transitions but differ in their burst spectra: /k/ has a sharp mid-frequency peak that /t/ lacks. Listeners are more likely to miss the spectral peak for /k/ than introduce it in the burst of /t/. These consonant confusion studies of Italian syllables support this hypothesis: Italian listeners confused /ki/ with /ti/ with increasing asymmetry when the S/N ratio increased (where noise masks the burst more than the formant transitions) and when the burst was excised completely. Implications for phonetic theory and speech technology will be discussed.

Learning Phonology With Substantive Bias: An Experimental and Computational Study of Velar Palatalization

“…The confusion rates for these sounds are asymmetric; for example, [ki] was identified as [t ʃi] 3.5 times more frequently than [t ʃi] was identified as [ki]. Such asymmetries are commonly found in identification experiments (Nosofsky, 1991;Ohala, 1997;Plauché, Delogu, & Ohala, 1997;Tversky, 1977). Although not of central interest for this article, the observed asymmetries do have some consequences, discussed in section 3, for formal modeling of the confusion data.…”

“…2. I take the absolute limits on human phonologies to be not substantive, but rather set by formal properties of the type investigated within optimality theory by Albro (2005), Eisner (1997), and Riggle (2004) and within rule-based phonology by, for example, Reiss (2003 (Nosofsky, 1991;Ohala, 1997;Plauché, Delogu, & Ohala, 1997;Tversky, 1977). Although not of central interest for this article, the observed asymmetries do have some consequences, discussed in section 3, for formal modeling of the confusion data.…”

Diachronic Explanations of Sound Patterns

“…However, as has been amply revealed in perception studies, perceptual confusion is very often asymmetric. For example, Guion (1998) found that when listening conditions were artificially degraded so as to increase confusion rates, misidentifications of /ki/ as /i/ occurred three to four times as often as the converse confusions (/i/ heard as /ki/), which were quite rare (see also Plauché et al. 1997; Winitz et al.…”

“…For example, in visual letter recognition, one is far more likely to misperceive a ‘Q’ as ‘O’ (by failing to notice the distinctive tail that sets the former apart from the latter) than to misperceive ‘O’ as ‘Q’ (by spuriously introducing that same characteristic feature). Plauché et al. (1997) suggest that the /ki/ vs. asymmetry is due to precisely such an all‐or‐nothing feature in the acoustic signal, namely, the spectral peak in the 3–4 kHz range that is characteristic of velars.…”

Publications Received for Review

“…Without these unique acoustic-phonetic cues, /k/ perceptually resembles /p/ and /t/, but /p/ and /t/ in the presence of noise would not be mistaken for /k/. Perceptual studies show support for such an explanation in the case of the observed asymmetric pattern favoring /ti/ over /ki/ [Plauché et al, 1997;Chang et al, 2001;Plauché, 2001]. In the account given in those reports, listeners may occasionally fail to perceive a midfrequency burst when presented with /ki/, resulting in the perception of /ti/, but listeners will rarely if ever hallucinate a mid-frequency burst when presented with /ti/, and will therefore rarely report /ki/.…”