Spoken word identification involves accessing position invariant phoneme representations

Abstract

Abstract:In two adaptation experiments we investigated the role of phonemes in speech perception. Participants repeatedly categorized an ambiguous test word that started with a blended /f/-/s/ fricative (?ail can be perceived as /fail/ or /sail/) or a blended /d/-/b/ stop (?ump can be perceived as /bump/ or /dump/) after exposure to a set of adaptor words. The adaptors all included unambiguous /f/ or /s/ fricatives, or alternatively, /d/ or /b/ stops. In Experiment 1 we manipulated the position of the adaptor phonemes so that they occurred at the start of the word (e.g., farm), at the start of the second syllable (e.g., tofu), or the end of the word (e.g., leaf). We found that adaptation effects occurred across positions: Participants were less likely to categorize the ambiguous test stimulus as if it contained the adapted phoneme. For example, after exposure to the adaptors leaf, golf... etc., participants were more likely to categorize the ambiguous test word ?ail as 'sail'. In Experiment 2 we also varied the voice of the speaker: Words with unambiguous final phoneme adaptors were spoken by a female while the ambiguous initial test phonemes were spoken by a male.Again robust adaptation effects occurred. Critically, in both experiments, similar adaptation effects were obtained for the fricatives and stops despite the fact that the acoustics of stops vary more as a function of position. We take these findings to support the claim that position independent phonemes play a role in spoken word identification. 3Traditional linguistic theory postulates a small set of phonemes that can be sequenced in various ways in order to represent thousands of words in a language (Chomsky & Halle, 1968;Trubetzkoy, 1969). Phonemes are the smallest linguistic unit that can distinguish word meanings and usually are of a size of a single consonant or vowel, e.g., the consonants /b/ and /p/ are phonemes in English because they differentiate the words "bark" and "park". Phonemes are critically distinguished from speech sounds (i.e. phones) in their level of abstractness. Phones are acoustically defined units that are often context-dependent, i.e. in a given language a certain phone may be bound to a specific syllable position, or require a certain stress pattern, or occur within the context of specific surrounding sounds. Although phonemes are widely assumed in linguistic theory, the psychological evidence in support of phonemes, at least in the domain of speech perception, is scant.This has given rise to various models that abandon phonemes as a functional unit in are used to activate dog and god representations, respectively. These time-bound segments can be seen as analogous to Pierrehumbert's position-specific phones (in that the segments do not abstract across position) although the input units in these models are often labeled phonemes.The common rejection of position invariant phonemes in psychological theories and models of word perception is a fundamental claim, and we explore this issue here. First we review the current empir...

Allophones, not phonemes in spoken-word recognition

“…The previous data with voiceless fricatives, both from perceptual learning and selective adaptation (Bowers et al, 2016;Jesse & McQueen, 2011) are consistent with both hypotheses. Even in the perceptual-learning paradigm, there is generalization of learning across position for fricatives (Jesse & McQueen, 2011), and there is no moderation in terms of effect size in the amount of learning from codato-coda position to coda-to-onset position.…”

“…Importantly, data from a perceptual-learning paradigm showed that some form of prelexical unit has to be assumed to allow learning to generalize from one set of words to another (McQueen, Cutler, & Norris, 2006;Mitterer, Chen, & Zhou, 2011;Sjerps & McQueen, 2010). Regarding the size of the units, data using this perceptual-learning paradigm supports the hypothesis that there are allophonic units ALLOPHONES IN SPOKEN-WORD RECOGNITION (in press, Journal of Memory and Language) 3 (Mitterer, Scharenborg & McQueen, 2013), while data using a selective-adaptation paradigm supports the additional hypothesis that there are also phonemic units (Bowers, Kazanina, & Andermane, 2016). The present study tests these two representational hypotheses.…”

“…The critical finding apparently arguing against the model without phonemes in perception (i.e., the model in the right panel in Figure 1) is the generalization of selective adaptation across position for place of articulation in stop consonants (the /b/-/d/ contrast tested by Bowers et al, 2016 (Stevens & Blumstein, 1978). There was therefore acoustic overlap between the exposure material and the test stimuli in the critical generalization condition.…”

“…Adaptation to initial segments appears to lead to adaptation on allophonic and phonemic levels, while adaptation of stops in other positions appears to lead only to adaptation at the phonemic level. Bowers et al (2016) highlight the issue that competent speakers of a language must know about phonemes. In the case of the Dutch word zwaar, a speaker needs to know that the underlying phoneme is /r/, because the approximant cannot be used for the inflected form zware (*[zʋa.ɹə]; the approximant can only be used in coda position, in syllable onset position a trill or tap has to be used (e.g., [zʋa.rə] not the case that this knowledge must reside in the pre-lexical perceptual system.…”

“…Even in the perceptual-learning paradigm, there is generalization of learning across position for fricatives (Jesse & McQueen, 2011), and there is no moderation in terms of effect size in the amount of learning from codato-coda position to coda-to-onset position. This is presumably because voiceless fricatives are physically very similar across positions (Bowers et al, 2016;Jesse & McQueen, 2011;. If voiceless fricatives therefore do not have different allophones for different positions, generalization of learning or adaptation across position with these segments can be explained at the allophonic level in both models.…”