explicit production practice. Second, instructors can ensure that multiple talkers are used to provide perceptual input. This could include the teacher in addition to recorded materials. Note that this is different from the common practice of providing multiple English accents and dialects in listening materials (e.g., British, American, Indian). In perception-focused pronunciation instruction, including talkers from different target-language varieties is disadvantageous, since the goal is to learn a particular model as opposed to learning to comprehend other-accented speech; however, the practice of including multiple accented voices is most useful for the teaching of listening. Third, appropriate types of corrective feedback should be employed (Lee & Lyster, 2016, 2017). Learners should not only be told that they made a mistake in perception but should also be given an opportunity to hear the item again, or hear an example of the non-target item they incorrectly selected. They should not receive feedback that includes replays of both target and non-target items, as this appears to lead to confusion. Fourth, training should incorporate both nonsense syllables/words and real words. The use of nonsense words seems to make phonetic information more salient, while teaching the perception of the same sounds in real words will promote transfer to the real world (Thomson & Derwing, 2016).
Since L2 pronunciation develops slowly and requires substantial amounts of input, it is important to focus perceptual training on those sounds that have the greatest Functional Load (FL). FL refers to the relative contribution of particular sound categories to the communication of meaning in a given language (Derwing & Munro, 2015; Munro & Derwing, 2006). This can be calculated on the basis of the frequency with which sounds contrast with other sounds, and may also incorporate information on grammatical categories within which minimal pairs are found, since within-category confusion is more problematic to communicating meaning than between-category confusions. For example, an error between “lock” and “rock” would be more confusing to a listener, since both are nouns, than an error between “laughed” and “raft,” where a mispronunciation is more likely to be successfully decoded by listeners, since one is a verb and the other a noun. Some particularly salient sound contrasts, such as those involving the English “th” sounds – /ð/ and /θ/ – have a very low functional load because despite being high frequency, they mainly occur in function words. Brown (1991) and Catford (1987) provide detailed FL information for English. In addition to overall FL, the location of particular errors within words has an impact on whether mispronunciations will lead to a breakdown in communication. In general, vowels contribute more to intelligibility than do consonants (Bent et al., 2007). Consonant errors at the beginning of syllables affect communication more than consonant errors at the ends of syllables. Taking these facts into account, pronunciation teachers can prioritize what sounds to teach and in what context. This is important given that learning in one phonetic context or word does not easily transfer to new contexts (Levis, 2018).
While most of the research on perceptual training has focused on segmentals, the same principles can be applied to determine which suprasegmental features warrant instruction. Derwing et al. (2012) demonstrated that some L2 English prosodic patterns can develop without the need for explicit instruction (e.g., word stress), while other patterns may benefit from instruction (e.g., sentence stress). There are also individual differences related to the L1 background of the listeners.
Lee and Lyster’s (2016) study is a great starting point for envisioning how perception-oriented training can be incorporated into the classroom in an evidence-based manner. They used several activities in their training sessions. One was a Pick-a-Card game, where each half of a minimal pair was written on either side of the same card (e.g., “lip” versus “rip”). Instructors produced target words and individual learners had to show the side of the card that matched the word they perceived. The participants then received immediate feedback on the accuracy of their response. This is different from a traditional minimal-pair activity in two ways. First, learners heard more than one talker produce the same sounds, since three teachers were involved in the training. Second, they received immediate corrective feedback, which is often not the case in a paper-and-pencil minimal pair task. Lee and Lyster also used Word Bingo and Fill-in-the-Blank activities to teach speech perception by incorporating minimal pairs that were known to cause perceptual confusion to this group of learners. As with the Pick-a-Card game, input was provided by more than one teacher and corrective feedback was given immediately after every response. These same types of activities could easily incorporate nonsense syllables/words in place of real words. This would require learners to know a phonetic alphabet, or to use key words containing the target sounds as labels for them to indicate what they had perceived. While Lee and Lyster (2016) provide several examples of perception activities, many traditional communicative language tasks could be adapted with similar results, such as information gap activities or find the difference tasks, in which minimal pairs of interest can be targeted.
The most obvious challenge to implementing these evidence-based activities in a classroom is the need for multiple instructors to provide the necessary input variability. It would be more practical to utilize recorded speech of multiple talkers, with immediate feedback still provided by the instructor. In mixed L1 classrooms, it might be possible to allow learners to provide input to each other, if they can be paired in such a way that members of each pair don’t have the same L2 pronunciation difficulties. For example, a Spanish L1 speaker could provide /l/-/ɹ/ pairs to a Japanese L1 speaker for whom this pair is problematic.
Practical Resources for Pedagogy
While many HVPT studies have examined this technique’s efficacy in the lab, their focus tends to be very narrow, investigating only a small number of sounds in a small number of phonetic or word contexts. As such, they do not comprise complete training systems which can be turned public-facing for consumption by a wide range of learners. Nor do they typically provide the opportunity for scalable research, since they are largely designed for one-off studies investigating a specific pronunciation issue faced by a particular group of learners. Even platforms that have been used to conduct research in the context of real language programs suffer from similar limitations (e.g., Thomson, 2011, 2012a; Wang & Munro, 2004.
A handful of HVPT researchers have attempted to make their platforms available to the public. Iverson and Evans’ (2009) British English Vowel Trainer was a mobile HVPT application, but it no longer seems to exist. Linguatorium Auris (LA) (Linguatorium Smart Language Systems, 2021), is an empirically validated HVPT system (Qian et al., 2018), bundled with a vocabulary training application. One of LA’s strengths is that it includes a variety of learning tasks beyond those normally associated with HVPT. This should encourage learner engagement. LA is also adaptive, targeting only those sounds that the system determines a particular user needs to target. A limitation of LA is that it only utilizes two training talkers, both of which are artificial. Thus, despite constituting HVPT in terms of its use of variable phonetic contexts, it only marginally qualifies as HVPT with respect to its number of talkers. English Accent Coach (EAC) (Thomson, 2018b) is largely a traditional listen-and-click version of HVPT training for vowels and consonants. It also includes a game called Echo (reminiscent of the 1980s Simon game, but with speech sounds instead of musical tones). EAC utilizes 30 talkers, producing North American English sounds in nearly every phonetic context. It also allows selection of sounds in either nonsense syllables or in real words. Unlike LA, EAC is not adaptive, requiring that users determine their own focus of instruction.
Both LA and EAC allow learners to create personal accounts to track their progress over time. LA is offered on a per semester basis, for a nominal fee, whereas EAC is currently free. Thomson (2012b) provides a detailed teacher-oriented description of EAC, along with some suggestions for activities to extend learning from the computer to the classroom. LA has less public documentation. Ultimately, both applications would benefit from built-in modules to encourage learners to practice the production of sounds that they are learning to perceive. This may facilitate faster transfer to production.
While it may be theoretically possible for instructors to build their own HVPT systems, if the goal is to include many talkers and sounds in many contexts, the recording process alone is a massive obstacle to overcome. Ideally, publicly available
