Correlational data
The data linking perception and production within individuals are also surprisingly sparse. Most of the data show that talkers’ perception and production categories are somewhat similar. For example, Newman (2003) found small correlations between the VOT prototypes of listeners and their production VOT values (accounting for approximately 27 percent of the variance). However, Frieda et al. (2000) did not find such a correlation for the perceptual prototype for the vowel /i/ and production values. Fox (1982) showed that the factor analysis dimensions derived from listeners’ judgments of similarity between vowels could be predicted by the acoustics of vowels produced by the participants, but only by the corner vowels /i, u, A/. Bell‐Berti et al. (1979) categorized the manner in which participants produced the tense/lax distinction in front vowels based on their examination of electromyographic recordings. They later found that those participants who used a tongue‐height production strategy showed larger boundary shifts in an anchoring condition in a vowel‐perception test than those who used a muscle tension implementation of tense/lax. Perkell et al. (2004) also grouped participants on the basis of measurements of production data, and found that these groups performed differently in perception tests. The more distinct the production contrasts between two vowels that talkers produced the more likely those subjects were able to distinguish tokens in a continuum of those vowels.
The most recent evidence in support of this correlational relation between perception and production abilities comes from Franken et al. (2017). In this study, production variability for vowel formant values was measured and the ability to discriminate between vowel tokens assessed. These two variables were found to correlate in their data. However, the correlations are modest and smaller than those reported by Perkell et al. (2004). The argument put forward in Franken et al. (2017) is that talkers with better perceptual acuity are less variable in production and that these talkers are more sensitive to feedback discrepancies. Indeed, Villacorta, Perkell, and Guenther (2007) showed a greater response to formant perturbation in subjects who had greater acoustic acuity. However, this finding is inconsistent with MacDonald, Purcell, and Munhall’s (2011) meta‐analysis of the variability of production and compensation magnitude in F1 and F2 for 116 subjects. The lack of relationship between variability and compensation observed by MacDonald et al. is important given the large sample size considered in their analysis.
Interference effects
If the two processes of perception and production of speech share a common representation or resource, there should be evidence that the performance of one process can interfere with or in some cases enhance the performance of the other process. There are tantalizing findings of this kind in the speech perception–production literature. At the adult level, there are a series of imaging studies and transcranial magnetic stimulation (TMS) studies that document possible interference or enhancement effects (see Skipper, Devlin, & Lametti, 2017, for a review; cf. Hickok, 2014, for a critical review of findings that relate to the mirror neuron hypothesis; Hickok, Holt, & Lotto, 2009, for criticism of motor influences on perception). Janet Werker and colleagues have reported a series of intriguing studies on early speech development that relate production to perceptual abilities (e.g. Bruderer et al., 2015). As has previously been shown by Werker and others, young infants are capable of perceptually making nonnative phonetic category distinctions. Using unique methods, they have shown that the ability to make these perceptual distinctions is reduced when the effector responsible for producing the distinction (i.e. the tongue in Bruderer et al.’s 2015 study) is interfered with by a custom‐designed soother. When a different soother, which did not interfere with the tongue’s movement, was used during the perceptual phase of the study, no perceptual interference was observed. Infants could perceive the lingual nonnative contrast if the tongue was not constrained by a soother. These six‐month‐old infants had not learned their perceptual or productive native sound inventory at the point of testing. Yet, there appears to be a sensorimotor linkage between phonetic perception and the articulatory system. In a follow‐up study, Choi, Bruderer, and Werker (2019) showed that the perception of nonnative contrasts as shown by Bruderer et al., and native contrasts (e.g. a /b/–/d/ distinction) can be interfered with by soothers that block the movement of key articulators in the production of the distinction (e.g. tongue tip, lips). These studies suggest a direct linkage between the speech motor and perception systems (Bruderer et al., 2015). Further, the results suggest an auditory–effector mapping that is available very early in development.
Influences in early infant speech behavior have also been shown from the other direction. Speech‐production tendencies can be correlated with developing perceptual abilities (e.g. Majorano, Vihman, & DePaolis, 2014; DePaolis, Vihman, & Keren‐Portnoy, 2011). Majorano, Vihman, and DePaolis (2014) tested children learning Italian at 6, 12, and 18 months. At the end of the first year, children whose production favored a single vocal motor pattern, showed a perceptual preference for sounds resulting from those speech movement patterns.
Conclusion
None of these findings, however, elucidate how auditory feedback processing develops, nor what role hearing your own productions plays in learning to pronounce words. The findings of MacDonald et al. (2012) raise the possibility that early productions by the child are not tuning the word‐formation system on the basis of auditory‐error corrections. Instead, the early focus may be on the adult models. Cooper, Fecher, and Johnson (2018) recently showed that two‐and‐a‐half‐year‐old children preferred recordings of adults over recordings of their own and other toddlers’ speech. In fact, the children showed no familiarity effect and thus did not prefer their mother’s speech from another adult’s. As Cooper, Fecher, and Johnson (2018) suggest, the driving force in lexical acquisition may be the adult targets and not the successive shaping of children’s targets by corrective auditory feedback.
Howard and Messum (2011, 2014, Messum & Howard, 2015) have proposed a model of early word pronunciation that is consistent with this idea. According to their view, infants develop a set of vocal motor schemes (Vihman, 1991, 1993, 1996) during babbling. When early word learning takes place, infants are producing sounds that parents reinforce in the context. By this account, it is the adult who is initially solving the correspondence problem of perceiving the match between the children’s production and the adult target. Adult modeling and reinforcement guide the infants’ early lexical acquisition. Using their computational model called Elija, Howard and Messum have shown that early productions can be acquired by the model learning the associations between “infant” motor patterns and the acoustics of the adult produced in that context.
One of the key requirements for successful reinforcement learning is exploration. Sampling the control space allows the organism to learn the value of a range of different actions. In vocal learning, variability of productions allows the learner to produce a broad range of activities that can be selectively reinforced. In adult movement, it may be desirable to reduce or control peripheral or execution variability (Harris & Wolpert, 1998). However, variability such as that generated in the birdsong circuitry indicates that this type of “noise” is an essential part of the sensorimotor system (Bertram et al., 2014). Motor learning may require this variability (Dhawale, Smith, & Ölveczky, 2017) and the variable behavior of toddlers learning to produce speech may be showing exactly the prerequisite talking required to be vocal learners.
In summary, it is clear that hearing yourself speak is important to speech motor control. The precision and regularity of articulation seem to depend on auditory feedback. However, how this process works is less clear. The considerable hysteresis of degradation shown when profound hearing loss occurs, or when artificial modifications of speech feedback are introduced, suggests that auditory feedback may be an important stabilizing force in the long term but it is not essential for moment‐to‐moment control. Further, the data show that hearing others speak involves a different process from the perception of the sound of your own speech. The loose relationship shown between perceptual and production individual differences
