Christina M. Esposito

Examining a language with phonemic modal-breathy contrast (Gujarati), a language with allophonic breathiness ([West Coast American] English), and a language that is traditionally described as without significant nonmodal phonation (Mexican Spanish), Esposito investigates whether linguistic experience (familiarity) with a particular phonation type (breathiness) will cause listeners to perform better overall and/or produce more consistent judgments than listeners speaking a language with allophonic or no breathiness, in several sorting and differentiation tasks involving modal and breathy tokens from languages with which none of the listeners are familiar. As a second question, Esposito compares the acoustic correlates of breathiness in each of the stimuli with the evaluation of stimuli by each of the listener groups, to see whether judgments are more congruent with some measures than others. Her prediction in this regard is that listeners with linguistic experience of breathiness (the Gujarati group) will make their judgments based on a measure that corresponds with the way breathiness is produced anatomically in their language (in this case, H1-H2); that listeners with allophonic breathiness (the English group) will perform better than listeners with no breathiness (the Spanish group); and that listeners with no breathiness, in the absence of a pre-developed strategy to identify it, will rely on the auditorially most salient feature in any given token (which should remain constant in listeners with no hearing impairment, as opposed to perceptual salience, which will vary according to linguistic background). A third and subsidiary question is whether differences will be found between the evaluation of foreign-language stimuli and the evaluation of pathological stimuli in a language (English) with which all listeners are familiar.


Esposito notes that while prior literature had borne out Ladefoged’s (1981: 35) observation that “one person’s pathology is another’s phonemic production”, this literature had mostly examined pathological production in English (cf. e.g. Kreiman et al. 1990, 1992, 1993). Incorporating listeners from languages with different approaches to phonation should provide a comparative perspective: do people who reliably distinguish normal from pathological production also distinguish phonemic phonation change? Esposito uses modal and breathy tokens from Chong, Green Hmong, White Hmong, Fuzhou, Mon, Santa Ana del Valle Zapotec, San Lucas Quiviani Zapotec, Tlacolula de Matamoros Zapotec, Tamang, and !Xóõ, which she categorizes as more or less breathy based on seven spectral variables (H1-H2 [Open Quotient—the most common measure], H1-F1, H1-F2, H1-F3, (H1-H2)/F1, F2-F3, H2-H4[often used in pathologically disordered voices]) and cepstral peak prominence (CPP), a measurement of the prominence peaks of the harmonics in a Fourier transform of the token, where better defined peaks indicate greater periodicity. (Hillebrand et al. [1994] have found CPP to be the most reliable measure of nonmodal phonation in English).


(The same) listeners participated in three tasks: a free-sort exercise (“What does the voice sound like?”) and per-pair similarity-rating tasks with the foreign-language tokens and the pathology data (contrasted with modal English tokens). In all tasks the Gujarati group scored very high and apparently relied primarily on H1-H2 to identify breathiness, confirming all hypotheses for Gujarati. English and Spanish results were very similar across tasks and only moderately reliable, suggesting that experience with allophonic breathiness does not confer an advantage in identifying linguistically significant or pathological breathiness. Foreign-language data as organized by English and Spanish listeners most closely correlated with H1-H2, then with H1-F2; contrary to expectations, neither English nor Spanish listeners appeared to use CPP in their foreign-language evaluations. However, English speakers did appear to use CPP in evaluating pathologically disordered English stimuli. (Esposito speculates that an auditory threshold made CPP a problematic evaluative measure for the foreign-language data (167), without going into detail.) Evalautions were very consistent between and within listener for the foreign-language data, and Gujarati speakers treated the pathological data the same way, supporting Ladefoged’s statement; however, English and Spanish participants had a high level of inter-listener disagreement for the pathology data, possibly suggesting that the “one person” for whom a stimulus exhibits a voice disorder and the other for whom it is heard as phonemic simply speak different languages!


Major questions Esposito’s dissertation brings up for my project are a) what an appropriate operationalization of “creak” will be for English data that exhibit neither phonemic creaky/modal contrast nor (anticipated) pathological creakiness; b) in the perception portion, whether a free-sort or similarity task wouldn’t be more appropriate than a binary discrimination (creaky/modal) exercise, since the latter may require providing listeners with linguistic training or in another way introducing the concept of “creaky” to them in a way that may prejudice the data; and c) following Esposito’s observations on different production of breathy voice in male and female Santa Ana del Valle Zapotec speakers (7), with male speakers relying more on speed of glottal closure as reflected by H1-F2 and female speakers more on OQ (H1-H2), whether I will need to consider type as well as presence of creak in designing a perceptual study where listeners are asked to distinguish between (pitch-shifted) male and female voices.… (lisätietoja)