• No se han encontrado resultados

ALUMNOS NO INSERTADOS

The ORE has a long and rich history, similar to the LFE in the sense that it has occupied thought in both forensic and cognitive psychology (for a meta-analytic review, see Meissner & Brigham, 2001). I will direct the reader to a particularly influential account of the ORE, for which empirical evidence has recently accumulated, and which might

provide a theoretical basis for discussion of the mechanistic underpinnings of the LFE. The similarity-based “face-space” account, proposed by Valentine (Valentine & Endo, 1992; Valentine, 1991) holds that faces are represented as points in a hypothetical

multidimensional space. When Valentine initially proposed this model, he considered it both in terms of a norm-based account and an exemplar-based account. In the norm- based version, faces are encoded by reference to the mean of the face space. The mean of the space represents an average of all faces encountered by the observer over the course of their perceptual experience. The exemplar-based version posits that individual faces are encoded with respect to their relative (dis)similarity to each other, so the centre of the space is irrelevant.

The dimensions of the face-space could represent features which are useful for identity diagnostics. For example, one dimension might encode the position of the eyes, or the shape of the nose, or the colour of hair. Importantly, regardless of what these dimensions might represent, they are proposed to be elaborated through perceptual experience; a consequence of this, is that the dimensions may be developed according to the faces to

which an individual is most exposed over the course of his or her life. Therefore, if a Western-Caucasian perceiver receives more exposure to other Western-Caucasian faces, then the dimensions of his hypothetical face encoding space will be attuned to those features which best aid in the differentiation of Western-Caucasian faces. For example, hair and eye colours span a relatively wide-range in Caucasians, as compared with, say, African or East Asian ethnicities. As such, an identity representation scheme which contains dimensions coding for these features will serve a perceiver well in recognition tasks. However, when a perceiver attempts to recognize faces from a group to which they have little exposure, then they may experience confusions between different individuals which could lead to recognition errors. As in our earlier example, consider the case of a Caucasian perceiver and a set of East Asian faces. If the observer has had little exposure to East Asian faces, then his encoding scheme will be asymmetrically optimised for the storage and retrieval of Caucasian faces. Consequently, when relying on diagnostic features which may help to individuate Caucasians – such as hair and eye colour –

recognition of Asian faces will be impaired, as different Asian identities may not be as well differentiated as a function of variability in those diagnostic features.

By extension, we can now consider the LFE within this framework. This analogy has been drawn by others previously (e.g., Perrachione et al., 2007), focusing on representations of language phonology. Taking the many apparent similarities between face and voice processing into account (Yovel and Belin, 2013), we may posit a multidimensional voice space which resembles that which has been adduced for faces. Here, dimensions may include speaking fundamental frequency (f0) or the formant frequencies, or harmonicity (“smoothness”) of a voice. Indeed, as was discussed earlier, this issue has already been examined in some depth (Baumann and Belin, 2010; Latinus et al., 2013). Importantly, the

manner in which a voice is encoded could be influenced by early linguistic experience. For example, as we have seen, infants display an early sensitivity to speaker changes in their native language only, which might follow on from perceptual narrowing to the sounds of their burgeoning mother-tongue (Kuhl, 2004; Kuhl and Rivera-Gaxiola, 2008). In the voice space, dimensions might be developed based upon the sounds an individual is most accustomed to hearing from other speakers in their environment; in other words, the native-language speech sounds most commonly heard. Under this conception, incoming voices will be assessed by reference to the listener’s stored representations of the sounds of their mother tongue. The listener can bring their phonological knowledge to bear in computing the speaker’s pitch, accent, or other features which might vary according to the speaker’s anatomy or geographical background. After this evaluation, the voice will be stored in the space accordingly, with reference to an individual’s internal voice

“prototype”. However, if they are tasked with evaluating voices which do not speak their native language, their voice-space dimensions (which have been shaped according to the phonology of that language) will be of lesser use for the encoding of those voices, which may result in a perception of increased inter-talker similarity, leading to a failure to tell- apart and recognize foreign voices.

While the framework proposed above describes the LFE with reference to stored

representations of voices in a multidimensional space which has been shaped according to linguistic input, an alternative conception concerns the “abstract” or “prototypical” representation of familiar speech sounds and language-specific acoustics themselves, as has been alluded to above and by others (e.g., Perrachione et al., 2007). In spite of the interactions between “speech” and “voice” information described earlier, human

native-language speech perception tasks; a process referred to as talker “normalization” (e.g., Johnson, 2008; Nygaard & Pisoni, 1998). For example, the word “dog” is easily recognizable to a native English speaker, whether it is produced by a female or male speaker, despite what may be considerable differences in the acoustical profiles of the two signals (Johnson, 2008). The listener may take advantage of stored “prototypical” representations of familiar speech sounds (such as phoneme combinations and words) and language-specific acoustics (e.g., in Mandarin, a prototypical representation of the characteristic f0 pattern of the rising tone) during speech processing in order to allow them to disregard pronunciation differences between speakers, which may result from such properties as accent or gender (Belin et al., 2004). Conversely, in the case of the LFE, just as this variability can be “tuned out” in speech perception tasks, it may in fact be brought to bear in a manner which benefits the listener when performing a voice recognition task in their native language. Rather than discarding variability, the listener could take advantage of inter-speaker differences in the pronunciation of familiar words and use these as cues for identity differentiation. In other words, the ability to process non-linguistic variability around robust, stored prototypical representations of native language speech sounds may facilitate voice recognition in a listener’s native language. On the other hand, as in the case of an unknown language, impoverished (or non- existent) representations of the speech sounds which are used by a talker will leave the listener reliant purely upon para-linguistic cues to voice identity, which, while useful, do not enable successful recognition to the same degree as cases where both linguistic and para-linguistic information is available (Perrachione et al., 2011; Zarate et al 2015). As efficient voice individuation is impeded, different foreign voices might therefore sound highly similar to a listener, leading to recognition failure.

With regards to the ORE, the “multidimensional space” model has garnered empirical support from computational modelling studies (Caldara and Abdi, 2006) and in recent electro-encephalography (EEG) and functional Magnetic Resonance Imaging (fMRI) studies showing enhanced individuation of own-race faces in putatively face-preferential neurophysiological responses (Vizioli, Rousselet, & Caldara, 2010; Vizioli, 2012). Given that a norm-based version of this account may be plausibly extended to voices (Bruckert et al., 2010; Latinus and Belin, 2011; Andics et al., 2013; Latinus et al., 2013; Yovel and Belin, 2013), it might also represent a useful basis for interpreting the LFE. Behavioural studies provide some support for voice recognition failure based on degraded

phonological representations (Perrachione et al., 2011), but do not explicitly evaluate listeners’ perceptions of similarity among different native and foreign talkers.

Furthermore, while the ORE has enjoyed sustained interest from the neuroimaging community (for a review, see Natu, Raboy, & O’Toole, 2011), no explicit examination of the neural basis of the LFE has been conducted, to my knowledge. The neural bases of speech perception (Hickok & Poeppel, 2007; McGettigan & Scott, 2012; Price, 2010, 2012; Scott & McGettigan, 2013) and voice recognition (Belin et al., 2004; Belin et al., 2011; Yovel and Belin, 2013; Schweinberger et al., 2014) have been, respectively, relatively well- studied, but most investigations evaluate these phenomena separately. However, neuro- cognitive models (Belin et al., 2004; 2011) and behavioural findings – such as the LFE – strongly suggest that the two systems are coupled in realizing successful voice

recognition. Indeed, a handful of recent neuroimaging works have begun to investigate this interaction. For example, it was recently shown that speech perception areas in the posterior left superior temporal cortex (STC) responded to speaker-related changes in vocal tract parameters (von Kriegstein, Smith, Patterson, Kiebel, & Griffiths, 2010). Furthermore, right posterior STC processed speaker-related information during a speech

recognition task as contrasted with a voice recognition task, and these left and right posterior superior-temporal regions were functionally connected. More recently,

Chandrasekaran and colleagues (Chandrasekaran, Chan, & Wong, 2011) showed that the left posterior middle temporal gyrus (MTG) exhibited reduced BOLD-signal adaptation to variations in both indexical (speaker-related) and lexical (word) information, as

contrasted with a condition where all stimulus content was repeated (i.e., same speaker identity and same word repeated). The authors interpreted this as evidence of neural integration of “what” and “who” information in this region. Taking these findings into account, therefore, it is important to determine the extent to which such putative

integrative regions are perturbed under foreign speech conditions, as contrasted with the familiar speech conditions previously used.

Documento similar