In everyday communication, listeners encounter a multitude of variations in the speech signal, even from a single speaker, which can affect intelligibility. Accents in speech, both foreign and regional, are a source of variability that is commonly encountered, and one that can be particularly detrimental to speech perception. Indeed, accented speech has often been shown to produce speech processing costs (see e.g., Floccia et al., 2006; Adank et al., 2009; see Cristia et al., 2012, for a review). Understanding spoken language requires mapping acoustic input onto stored phonological and lexical representations. Therefore, during speech perception, listeners need to be flexible in accommodating acoustic signals that do not match perfectly with stored representations. When the mismatches are small, listeners are able to comprehend speech with minimal effort. When the variations are more substantial, however, such as with foreign-accented speech, additional cognitive resources may be required to process the speech signal (Van Engen and Peelle, 2014).

The present study investigated the potential influences of cognitive processing capacities on perception of accented speech. Specifically, executive functions, information processing speed, and working memory were assessed. For executive functions, there are varying definitions and frameworks of its components. One widely accepted account (Miyake et al., 2000) identifies three key components, namely inhibitory control, task switching, and updating. The present study will focus on the inhibition and switching components. Previous research (Adank and Janse, 2010) has shown that executive functions, particularly task switching, as measured by the Trail Making Test (TMT; which involves connecting dots in alphanumerical sequence, alternating between letters and numbers), predicts relative difficulty in understanding accented speech in a group of older adults, whereas information processing speed, as measured by the Coding test (which involves using a key to copy a series of abstract symbols paired with numerical digits), does not. The accent was an artificially created accent of Dutch, produced by reading from an adapted orthography. Adank and Janse (2010) also investigated perception of the novel accent by young adults, and found that they had significantly less difficulty in understanding the accent than the older adults. The measures of cognitive processing capacities, however, were not examined in those young adults. The present study, therefore, sought to investigate whether measures of cognitive processing capacities predict performance in perception of unfamiliar accents by young adults. When perceiving unfamiliar accented speech, older adults, given their declining cognitive functioning, may need to draw on mental resources to a greater extent than young adults (Pichora-Fuller, 2003; Adank et al., 2009). Nevertheless, cognitive processing capacities may still play a role in the perception of unfamiliar accents for young adults, and influence their performance on the speech perception tasks in the present study. In addition, a range of cognitive capacities were assessed, including two components of executive functioning, information processing speed, as well as working memory.

When processing accented speech, cognitive flexibility (i.e., “task switching” or “set-shifting”) may be required to match nonstandard pronunciations of words and speech sounds to stored standard representations. Furthermore, particularly when the accented pronunciations deviate more substantially from standard pronunciations, inhibitory control may be required to manage the activation of incorrect competing representations in order to make the correct match (Gordon-Salant and Fitzgibbons, 2004; Rudner and Lunner, 2014; Rönnberg et al., 2016). Speed of information processing, although not found to significantly predict older adults' performance in the understanding of a novel accent, may still be relevant to accent perception, as it may relate to the ability to rapidly pick up the regularities in the ways phonemes or speech sounds are replaced in accented speech (Pichora-Fuller, 2003; Adank and Janse, 2010). Therefore, the present study assessed both executive functions (including both switching and inhibition) and speed of information processing, to determine their influence on speech perception.

Working memory capacity was also assessed (using the Digit Span task, consisting of three parts—Forward, Backward, and Sequencing), since working memory has been found to influence the perception of speech, especially when of low intelligibility, such as speech-in-noise, speech produced by impaired speakers, or speech produced by a speech synthesizer (e.g., Francis and Nusbaum, 2009; Janse and Adank, 2012; see Akeroyd, 2008, for a review). Furthermore, by limiting the availability of working memory, such as by using a concurrent working memory task, speed and accuracy in the recognition of synthetic speech have been shown to be reduced, especially for low intelligibility synthetic speech (Francis and Nusbaum, 2009). The Ease of Language Understanding (ELU) model (Rönnberg, 2003; Rönnberg et al., 2008, 2013) describes the role of working memory in a wide range of speech listening conditions. In particular, if the phonological information extracted from the speech signal does not match with stored phonological representations (such as in speech-in-noise, distorted speech, or other low intelligibility speech), lexical access is then delayed, which in turn slows comprehension. Foreign accents in speech, particularly unfamiliar foreign accents, reduce intelligibility (Munro and Derwing, 1995; van Wijngaarden et al., 2002; Bradlow et al., 2010). Thus, working memory capacity may serve as a limiting factor for performance in the perception of accented speech.

Previous research has found influences of cognitive functioning on perceptual learning or adaptation to accented speech following exposure or training (Janse and Adank, 2012; Banks et al., 2015), as well as on perception performance with other types of nonstandard speech, such as speech-in-noise (Ellis et al., 2015), frequency-compressed speech (Ellis and Munro, 2013), and dysarthric speech (Baese-Berk et al., 2015). Many studies have shown that listeners can adapt to accented speech following even brief training exposures (e.g., Clarke and Garrett, 2004; Bradlow and Bent, 2008; Sidaras et al., 2009; see Samuel and Kraljic, 2009; Cristia et al., 2012, for reviews), and this adaption has further been linked with executive functions, particularly inhibitory control. Specifically, healthy young adults who performed better on the Stroop test adapted more and at a faster rate to a novel artificially created accent (Banks et al., 2015), while performance on a variation of the Flanker task was found to predict amount of adaption to an artificially created novel accent in older adults (Janse and Adank, 2012).

In regard to cognitive influences on nonstandard speech perception, Ellis and Munro (2013) showed that performance on the TMT and the Reading Span Test (RST; a measure of working memory span) were correlated with the recognition of speech-in-noise in a group of healthy adults. Ellis et al. (2015) further showed that performance on the TMT predicted recognition of frequency-compressed speech-in-noise, in both adults with and without hearing loss. TMT performance significantly predicted such speech recognition even when age (which was the strongest predictor found in that study) was partialled out. Using a different measure of task switching, namely the Intra-Extra-Dimensional Set Shift (IED), Baese-Berk et al. (2015) found that switching correlated with perception of dysarthric speech, as well as with perception of foreign-accented (Spanish-accented English) and regional-accented speech (Irish-accented English), in a group of healthy adults. However, inhibition, as measured by the Flanker task, did not have a significant influence.

The present study further investigated the influences of switching and inhibition on the perception of foreign-accented speech, in a group of healthy young adults. Unlike the study by Baese-Berk et al. (2015), this study examined multiple unfamiliar foreign accents. These were accents that would not be commonly heard in the participants' day-to-day environment, unlike Spanish- and Irish-accented English, which are comparatively more commonly heard in the USA where that study was conducted. A decrease in familiarity may lead to an increased reliance on cognitive resources for successful perception. Further, having multiple accents helps to ensure that accent perception is not influenced by relative familiarity with any particular accent. The present study also assessed other cognitive processing capacities, namely information processing speed and working memory, which may also have influences on the perception of unfamiliar foreign-accented speech, as outlined above.

When analyzing the results of the sentence verification task, in addition to overall performance across all statements, the two types of statements (i.e., true and false) were examined separately, as they may produce different patterns of results. Specifically, false statements may provide less adequate contextual cues to help identify the sentences, given their often absurd meaning. Thus, listeners may need to rely more on cognitive resources when perceiving false statements than when listening to true statements. Table 1 presents the mean performance scores on each measure. It can be seen that the accented part in the sentence verification task produced significantly slower response times than the standard part. These differences were significant for true statements, false statements, and overall, t₍₁₃₁₎ > 15, p < 0.001. However, participants made very few errors across both accented and standard sentences, and there were no significant differences across the two conditions, t₍₁₃₁₎ < 1, p > 0.3. To examine whether cognitive processing measures predicted speech perception outcomes, analyses using simple regression were conducted. Each of the cognitive processing measures was used as a predictor variable in separate simple regression analyses, predicting performance on each of the speech perception measures (see Table 2 for correlations among the variables, and Table 3 for detailed regression results).

For auditory sentence verification performance, TMT Part B, CWIT Inhibition and Inhibition/Switching conditions, and Coding each significantly predicted overall response times for the accented part, t₍₁₃₀₎ > |2.4|, p < 0.02, and overall response times for the standard part, t₍₁₃₀₎ > |2.3|, p < 0.03. These four variables also predicted response times for both true statements, t₍₁₃₀₎ > |2.3|, p < 0.03, and false statements, t₍₁₃₀₎ > |2.1|, p < 0.04, within the accented part, and both true statements, t₍₁₃₀₎ > |2.1|, p < 0.04, and false statements, t₍₁₃₀₎ > |2.1|, p < 0.04, within the standard part. That is, better performance on each of those measures (i.e., faster completion times for TMT or CWIT, or higher Coding scores) led to faster response times for both accented and standard sentence verification. For the measure of error rates, Digit Span Sequencing was a significant predictor for false statements in the accented part, b = 0.43, t₍₁₃₀₎ = 1.98, p = 0.049, where higher scores led to higher error rates. For the standard part, CWIT Inhibition Cost (Inhibition—Color Naming) was a significant predictor of overall error rates, b = 0.09, t₍₁₃₀₎ = 2.11, p = 0.037, with better performance predicting fewer errors, while TMT ratio score, CWIT Inhibition condition, Inhibition/Switching condition, and Inhibition Cost each significantly predicted error rates for non-foreign-accented false statements, t₍₁₃₀₎ > |2.2|, p < 0.03.

For accented lexical decision, CWIT Inhibition and Coding were significant predictors of response times for both words, t₍₁₃₀₎ ≥ |2.3|, p < 0.03, and nonwords, t₍₁₃₀₎ > |2.0|, p < 0.05. That is, better performance on each of those measures led to faster response times. For error rates, CWIT Inhibition condition, Inhibition/Switching condition, and combined Inhibition/Switching Cost (Inhibition/Switching—combined Color Naming and Word Reading) each significantly predicted error rates for nonwords, t₍₁₃₀₎ > |2.0|, p < 0.05, with better performance leading to fewer errors.

In summary, better performance on some measures of executive function, including TMT Part B, CWIT Inhibition and CWIT Inhibition/Switching, and better performance on the Coding measure of information processing speed predicted faster response times for both versions of the auditory sentence verification task (accented and non-foreign-accented). Better executive function performance, as indicated by CWIT Inhibition, and better performance on the Coding measure of information processing speed also predicted faster response times for the accented lexical decision task. For errors, better performance on inhibition and switching measures, particularly those in the CWIT, predicted lower error rates for non-foreign-accented sentence verification, especially for false statements, as well as lower error rates for nonwords in the accented lexical decision task. There were, however, some differences in the executive function measures that predicted response times vs. those that predicted error rates. Specifically, those that predicted error rates included measures where baseline performance has been partialled out (i.e., TMT ratio score, CWIT Inhibition Cost, and combined Inhibition/Switching Cost), whereas those that predicted response times did not (i.e., TMT Part B, CWIT Inhibition, and CWIT Inhibition/Switching). Lastly, working memory did not consistently predict performance on the speech perception tasks.

The present study investigated speech perception as a function of more general cognitive processing performance, so that potential influences of these cognitive processing capacities on speech perception could be examined. Specifically, the influences of executive functions (both switching and inhibition), information processing speed, and working memory were assessed. Previous findings with older adults have shown that executive function processes, specifically task switching as measured by the TMT, predict performance in perception of unfamiliar accented speech, while information processing speed, as measured by the Coding test, did not (Adank and Janse, 2010).

In the present study with young adults, both executive functioning and information processing speed were found to be associated with speech perception performance. Task switching, inhibitory control, and information processing speed predicted response times for both accented and standard sentence stimuli. Inhibition and information processing speed also predicted speed of responding to accented word-length stimuli. However, the measures of executive functions that predicted speech perception response times were those that did not adjust for baseline performance. Since those indicators were raw completion times on the relevant tasks (i.e., TMT Part B, CWIT Inhibition condition, and Inhibition/Switching condition), they may also reflect processing speed to some degree, rather than being pure measures of executive function processes. The influences of information processing speed on speech perception found in the present study is likely due to the nature of the speech perception tasks, which required participants to make speeded decisions and responses. The speech perception task in the study by Adank and Janse (2010) did not measure response speed, but instead measured the number of words correctly repeated after each sentence, which allowed participants to take their time in making responses.

When looking at the accuracy measure of the speech perception tasks, only executive function processes predicted performance, and information processing speed did not, consistent with Adank and Janse (2010). However, the influence of executive functions on speech perception accuracy in the present study was found only for some types of stimuli, namely false statements and nonwords. Further, unlike for speech perception response times, the measures that predicted speech perception accuracy included those that adjusted for baseline performance (i.e., TMT ratio score, CWIT Inhibition Cost, and combined Inhibition/Switching Cost), which may be more rigorous and pure indicators of executive functioning. The differences between the outcomes of the present study and those of Adank and Janse (2010) may be due to differences in functioning between young and older adults. Older adults were shown to have considerably greater difficulty understanding unfamiliar accented speech, as well as a lesser degree of perceptual adaptation following exposure, compared to young adults (Adank and Janse, 2010). Older adults may, therefore, need to draw on cognitive resources to a greater extent than young adults to perceive unfamiliar accented speech (Pichora-Fuller, 2003; Gordon-Salant and Fitzgibbons, 2004). Young adults may only need to draw on cognitive resources for items like false statements and nonwords, which do not provide adequate contextual cues to be easily perceived.

The present results are partly consistent with those of Baese-Berk et al. (2015), who found that switching predicted accented speech recognition, but inhibition did not. Compared to the study by Baese-Berk et al. (2015), the accents in the present study were less familiar to participants, and thus may require increased cognitive resources for successful perception. Specifically, the input signal of highly unfamiliar accents may differ more from expectations or stored representations than accents that are more familiar. Listeners may, therefore, need to utilize inhibitory control to a greater extent to match the incoming deviant signal to the correct representations, that is, by inhibiting the activation of multiple competing representations that are also potential matches for the input signal (Gordon-Salant and Fitzgibbons, 2004; Rudner and Lunner, 2014; Rönnberg et al., 2016).

The effortful task in the perception of nonstandard speech is overcoming mismatches between perceived phonological input and stored phonological representations, in order to comprehend the intended meaning. As such, numerous cognitive processes may be required for successful comprehension, that is, not only aspects such as inhibition, switching, and information processing speed, processes like working memory may also need to be drawn upon (Janse and Adank, 2012; Rönnberg et al., 2013). The ELU model (Rönnberg, 2003; Rönnberg et al., 2008, 2013) explicates that working memory plays a role in speech perception, under varying listening conditions, and by listeners with or without hearing impairments. In the present study, however, working memory was not found to predict perception of accented and standard speech, by young, healthy listeners. Previous research has suggested that working memory capacity may serve as a limiting factor when perceiving speech with reduced intelligibility, such as unfamiliar accented speech (e.g., Janse and Adank, 2012; see also Rönnberg et al., 2013). It is possible that working memory only plays a role when perceiving speech with severely reduced intelligibility. Indeed, evidence for an impact of working memory on speech perception has typically been found using low intelligibility speech or speech that may initially be unintelligible, such as speech-in-noise and synthetic speech (e.g., Francis and Nusbaum, 2009; Janse and Adank, 2012; see Akeroyd, 2008, for a review). Furthermore, it has been suggested that simple span tasks, such as digit span, mainly assess storage capacities in short-term memory, which may not be a good predictor of language comprehension, and that complex working memory capacity is the ability involved in understanding language (Rönnberg et al., 2013). Although, the Digit Span task used in the present study had two of three parts that entailed simultaneous storage and processing of information, a more multifaceted task may be required to assess complex working memory capacity, for example reading span and other complex span (C-span) tasks, dual n-back and recall n-back tasks, or binding tasks (see e.g., Rönnberg et al., 2013; Wilhelm et al., 2013). Such tasks require continuous mental monitoring and updating, in addition to simultaneous storage and manipulation. Finally, not all studies have found an influence of working memory on speech processing, particularly for young listeners with normal hearing. In a meta-analysis, Füllgrabe and Rosen (2016) found that working memory (as measured by the reading span task) was not consistently involved in speech perception under adverse listening conditions (i.e., comprehending speech-in-noise), for young, healthy listeners.

To conclude, the present study showed that cognitive factors play a role in determining perception of variable speech in healthy young adults. In particular, aspects of executive function predicted accuracy of perception, while both information processing speed and executive functioning predicted speed of performance in the perception tasks. These influences were observed with both sentence and word-length stimuli. Furthermore, accented speech perception was not likely to be influenced by relative familiarity with any particular accent, as multiple unfamiliar accents were included in the stimulus set. The results of the present study suggest that stronger abilities in cognitive switching and inhibition are likely to be helpful for matching unfamiliar pronunciations of speech sounds to stored representations, for example by being able to inhibit incorrect competing representations and shifting to other possible matches (Gordon-Salant and Fitzgibbons, 2004; Adank and Janse, 2010; Rudner and Lunner, 2014; Rönnberg et al., 2016). Additionally, information processing speed likely affects the ability to rapidly learn the regularities in the way speech sounds are produced, thus affecting the speed of responding to speech stimuli (Pichora-Fuller, 2003; Adank and Janse, 2010). There is scope for future work to further examine the involvement of various cognitive capacities on speech perception under different types of adverse listening conditions, for example increased difficulty in the listening conditions posed by environmental factors, such as background noise, over and above the unfamiliar foreign accent.

Conception or design of the work: LT, MT. Data collection: LT. Data analysis and interpretation: LT, MT. Drafting the work: LT. Critical revision of the work: LT, MT. Final approval of the work to be published: LT, MT.

This research was supported by an Australian Postgraduate Award from the Australian Government awarded to LT.