The Role of Linguistic Knowledge in the Encoding of Words and Voices in Memory

There is a longstanding tradition in phonetics to conceive of speech as containing both “indexical” and “linguistic” components (Abercrombie, 1967). The linguistic properties of speech support the identification of linguistic (phonemic, lexical, etc.) contrasts, while the indexical properties of speech “reveal personal characteristics of the speaker”, such as their dialect or group membership, their gender, their age, their emotional or mental state, or particular vocal idiosyncrasies unique to that speaker. In Abercrombie's conception, speakers provide a personal "medium" for linguistic messages; this means that certain indexical properties in this medium may be "extra-linguistic". Exemplar theories of speech perception (Johnson, 2007) have emphasized the cognitive utility of not breaking down the signal into independent linguistic and indexical streams. Instead, exemplar theories conjecture that listeners store in memory unanalyzed representations of speech that are “rich” with informative detail. Linguistic representations might, for instance, include indexical or talkerspecific information, just as indexical representations might include linguistic information. Generalizations are then calculated on the fly, from summed activations of similar exemplars. In its strong form, exemplar theory claims that listeners store in memory every detail of every speech event they experience in their lifetimes. The core evidence for the exemplar view comes from known interactions between indexical and linguistic properties in speech perception. Linguistic information affects the identification of the indexical properties of speech in that listeners can more easily identify someone speaking a language that they know (Goggin et al., 1991; Perrachione et al., 2009). The reverse of this equation is also true--indexical information affects the processing of the linguistic content of speech. Palmeri et al. (1993), for instance, demonstrated that voice information influences the recognition of repeated words in a "continuous word recognition" experiment. Listeners in this experiment heard a series of words, spoken in different voices, and identified whether or not each word was “old” or “new” in the list. It was easier for listeners to recognize repeated words if they were spoken in the same voice as on the first presentation, suggesting that listeners were storing word and voice information together in memory. Nygaard et al. (1994) also showed that there was a "familiar talker advantage" in word recognition. In this study, listeners learned to identify talkers over a series of nine days. On the tenth day of the study, the listeners transcribed words spoken by both trained (familiar) talkers and untrained (unfamiliar) talkers. The listeners identified a significantly higher percentage of words produced by the familiar talkers, indicating that knowledge of the familiar talkers' voices in memory contributed to the processing of novel stimuli. However, such evidence of indexical influences on linguistic processing primarily comes from English-only language studies. Do listeners encode the same amount of indexical detail in memory from exemplars of second language (L2) speech? There is some reason to suspect that they might not, since listeners often exhibit perceptual insensitivity to fine-grained phonetic details of L2 speech. For