Perceptual features of unknown foreign languages as revealed by multi-dimensional scaling

Abstract

Adult listeners are able to discriminate between and often identify spoken samples of languages that are unknown to them. Two studies were designed to explore which perceptual properties inherent within the phonological structure of languages are salient to foreign language listeners.In study one, fifteen subjects were asked to judge whether pairs of spoken foreign language sentences were selected from same or different languages and to explain how they had made the judgement. A multi-dimensional scaling (MDS) was conducted on the subject responses for the 'same language' condition. The resulting map revealed that responses could be characterized along two dimensions: phonologically based psychoacoustic properties and talker specific characteristics. The two dimensions define the distinctiveness of the languages and elicited different perceptual feature relationships in subjects.In study two, this perceptual feature relationship was tested using similarity judgements. Thirty subjects rated similarity on a sevenpoint scale for the same set of sentence pairs that had been judged in study one. MDS analysis revealed that the 'different language' condition yielded a map in which the language proximities closely approximated those which had been derived by focusing on phonological properties. This finding suggests that since analysis of both 'different language' and 'same language' sentence pairs produced similar maps, perceived language similarity among foreign languages depends upon the listeners' salient organizational categories inherent within the phonological structure of language and the talker specific characteristics of voice quality and speech rate.

Processing Relationships Between Language-Being-Spoken and Other Speech Dimensions in Monolingual and Bilingual Listeners

“…In line with previous work demonstrating a language familiarity benefit for talker identification (e.g., Bregman & Creel, 2014; Goggin et al, 1991; Köster et al, 1995; Perrachione et al, 2009; Perrachione & Wong, 2007; Thompson, 1987; Wester, 2012; Winters et al, 2008), it was found that language-being-spoken interfered with the classification of talker. Further, the fact that talker also interfered with the classification of language-being-spoken is consistent with previous work suggesting that listeners can use talker information when classifying languages (e.g., Muthusamy et al, 1994; Stockmal, 1995; Stockmal et al, 1996). Moreover, the asymmetry found in this experiment is consistent with the idea that talker-general cues may be more important than talker-specific cues in language identification; listeners abstract over talker information to form representations of languages, and phonetic information is organized language-specifically.…”

“…Our finding that language-being-spoken interfered with talker supports prior work demonstrating a language familiarity benefit for talker identification (e.g., Bregman & Creel, 2014;Goggin et al, 1991;Köster et al, 1995;Perrachione et al, 2009;Perrachione & Wong, 2007;Thompson, 1987;Wester, 2012;Winters et al, 2008). Also, previous research suggesting that listeners can use talker information when classifying languages (e.g., Muthusamy et al, 1994;Stockmal, 1995;Stockmal et al, 1996) is bolstered here by the result that talker also interfered with language-being-spoken. The asymmetry observed in Experiment 1 is further supported by several observations discussed in the introduction to Experiment 1, where it was suggested that talker processing is more language-specific than language-being-spoken processing is talker-specific; the fact that listeners abstract over talker information to form representations of languages emphasizes the importance of talker-general information, and the fact that phonetic information is organized language specifically (Bradlow, 1996;Flege & Eefting, 1988;Lisker & Abramson, 1964) emphasizes the importance of language-specific information.…”

“…For language classification, the role of talker and other indexical information has been less systematically studied, but there are suggestions that the identification of language-being-spoken may also be at least partially reliant on talker processing. Although there is evidence that listeners use talker-general acoustic information when identifying languages unfamiliar to them (e.g., pitch patterns, rhythm, and segment; Stockmal, Muljani, & Bond, 1996), they may also be making use of talker-specific information (e.g., voice quality, speaking rate; Muthusamy, Jain, & Cole, 1994; Stockmal, 1995; Stockmal et al, 1996) in language identification decisions.…”

Rhythmic unit extraction and modelling for automatic language identification

“…Following Muthusamy, several researchers have tried to quantify these effects. Stockmal, Bond and their colleagues (Stockmal et al, 1996;Stockmal et al, 2000;Bond & Stockmal, 2002) have investigated several socio-linguistic factors (geographical origin of the speakers, languages known by the subjects, etc.) and linguistic factors (especially rhythmic characteristics of languages).…”

Automatic language identification with recurrent neural networks

“…LID is of interest as an attempt to automate one o f the capabilities of the human brain ([8], [13]). Practical applications of LID include systems that iden* the speaker's language as their primary functionality (e.g., multi-lingual telephone-call routing), and system augmenting speech recognition ( e g , vocabulary selection in multi-lingual multimedia and translation systems).…”