2.1 HVPT

Perceptual training is a central element of teaching pronunciation since learners rarely produce new sounds they cannot hear or distinguish from other perceptually similar sounds. Some research has demonstrated that improvements in perception lead to improvements in production (Bradlow et al. Reference Bradlow, Pisoni, Akahane-Yamada and Tohkura1997; Lambacher et al. Reference Lambacher, Martens, Kakehi, Marasinghe and Molholt2005). What kind of perceptual training is most likely to be effective in understanding natural speech? Increasingly, research suggests that input that fully represents the range of variation in normal speech leads to more robust acquisition of an L2 sound system. This is commonly known today as HVPT. Thomson (Reference Thomson2012) reviews the use of HVPT, showing that input with high variability (multiple speakers and many tokens in varied phonetic environments) leads to greater perceptual improvement, especially if it is combined with immediate information about success (i.e., feedback). This contrasts with traditional training. When first learning to teach pronunciation, I was taught that I should always use the clearest exemplars possible. For example, vowels would be taught in stressed contexts before voiced consonants (raid, lose) or in open syllables (see, go) as these provided learners with a clear model. This seems to be a mistaken approach, given how L2 learners build reliable models of L2 speech sounds. The use of HVPT training means that we now have a greater range of options for developing perceptual skills.

Lively, Logan & Pisoni (Reference Lively, Logan and Pisoni1993) and Lively et al. (Reference Lively, Pisoni, Yamada, Tokhura and Yamada1994) began testing HVPT began in the early 1990s with studies of /l/ and /r/ for Japanese learners of English. Japanese learners of English find the /l/ and /r/ contrast particularly difficult. These sounds have been widely studied because of their ability to demonstrate how new categories can be constructed and new sounds acquired. Results of early HVPT training were very promising. In a later study, Wang & Munro (Reference Wang and Munro2004) examined computer learning of three sets of vowels (some modified) in North American English and found varied but demonstrable increases in perception, a finding confirmed by Nishi & Kewley-Port (Reference Nishi and Kewley-Port2007, Reference Nishi and Kewley-Port2008). Iverson and colleagues (Iverson & Evans Reference Iverson and Evans2007, Reference Iverson and Evans2009; Iverson, Hazan & Bannister Reference Iverson, Hazan and Bannister2005) examined the effect of L1 in HVPT training, showing that the effectiveness of HVPT can be different depending on learners’ L1s. The cause and effect relationship of L1 and training is not yet clear.

Thomson (Reference Thomson2011, Reference Thomson2012) created HVPT training materials for Mandarin first language (L1) learners of English targeting ten Canadian vowels. Using a controlled experimental methodology, the training materials were designed for instructional use using computer-assisted pronunciation teaching outside of the classroom context. The HVPT materials included multiple speakers (ten male and ten female). In addition, some of the materials were acoustically manipulated, doubling the duration of the vowel nuclei to make them more noticeable. Previous research has shown that learners’ vowel categorization after the first year in an L2 environment is minimal in naturalistic development (Munro & Derwing Reference Munro and Derwing2008), so Thomson's use of HVPT to create perceptual improvement in English vowel categorization would, if successful, provide hope for pronunciation improvement.

Thomson used examples of each of the 20 speakers pronouncing the ten English vowels in open syllables with bilabial onsets ([pV] and [bV]), and found that (1) most learners increased their ability to identify the English vowels, (2) training on the bilabial CV (Consonant-Vowel) context transferred to some other environments, (3) improvement persisted on a delayed posttest, and (4) the learners became more confident in their ability to identify the vowels. The increased duration in some stimuli, however, did not help them identify the vowel categories, suggesting that this feature alone was not able to increase improvement.

These findings show that HVPT training has great promise in pronunciation training, even on features that seem particularly resistant to instruction. Thomson (Reference Thomson2011) says that ‘when the stimuli are highly variable, and when the training program promotes the orientations of learners’ attention toward the phonetic properties of the stimuli, further enhancement of the stimuli is unnecessary’ and ‘even less than ideal training items will provide the sort of input that can promote learning’ (p. 759). In an extension of his research, Thomson created a training app for mobile devices and computers, English accent coach, using the same principles.Footnote ¹ In a language learning and teaching environment in which pronunciation training materials will increasingly become digital and embedded in apps and e-books,Footnote ² this is a promising look toward the future.

It is important to carry out other studies using ‘multiple talkers’: for example, talkers all from the same dialect area, of both genders, of varied voice qualities, of different dialects, with native and nonnative models. We do not yet know what possibilities exist for multiple talkers, but decisions are likely to differ based on the goals of the L2 teaching context. Currently published pronunciation teaching materials do not typically use multiple talkers, although materials with multiple talkers are now widely available on the internet. In an analysis of commercial pronunciation software carried out by students in my graduate class, most programs had only one speaker for all activities. This should, and can, change.

2.2 Intonation

Since the late 1970s, theoretical and laboratory research into intonation has exploded with new findings, not only for English but for many other languages (e.g., Hirst & Di Cristo Reference Hirst and Di Cristo1998). Almost none of this research has found its way into teaching materials, despite excellent resource books connecting theory and practice (Wichmann Reference Wichmann2000; Wennerstrom Reference Wennerstrom2001; Chun Reference Chun2002). While some materials have been written as direct applications of models of intonation, such as Brazil's (Reference Brazil1997) discourse intonation (e.g., Bradford Reference Bradford1988; Cauldwell Reference Cauldwell2002; Gorsuch et al. 2012), intonation in most textbooks does not reflect current research. Indeed, rather than addressing intonation's role in conversational management and turn-taking (Chun Reference Chun2002), its ability to send nuances of pragmatic meaning (e.g., Thompson Reference Thompson1995), and its role in marking topic shifts through paratones (i.e., extra high tones that mark topic shifts in spoken discourse, see Wichmann Reference Wichmann2000; Levis & Pickering Reference Levis and Pickering2004), textbooks still overemphasize intonation as a reflection of grammatical form and as having a special role in expressing emotional meaning, a belief that has long been questioned (e.g., Crystal Reference Crystal1969).

What do we know about intonation from research? First, intonation functions in multiple domains within speech (Pierrehumbert Reference Pierrehumbert1980; Ladd Reference Ladd2008; Levis & Wichmann Reference Levis, Wichmann, Reed and Levis2015). Intonation in English is meaningful at the ends of phrases and sentences, at topic changes, and is especially meaningful in emphasizing words and syllables. Second, intonational meaning is primarily semantic and pragmatic, not grammatical (Levis & Wichmann Reference Levis, Wichmann, Reed and Levis2015). Any utterance with any type of grammar can be spoken with multiple intonation patterns, each of which will change the meaning of the utterance. Some of these meaning differences may be subtle, and some very obvious, but the basic effect of changes in intonation is not mispronunciation (as in saying sink for think) but a difference in how speaker intent is communicated (e.g., the question Are you going? spoken with rising intonation may sound more polite while the same question with falling intonation may sound more businesslike, see Thompson Reference Thompson1995 for more on this distinction).

Third, intonation is both holistic and atomistic. Intonation functions across the whole of a phrase, but is made up of smaller meaningful elements (e.g., described as discrete pitch levels for Americans, but preheads, heads, tonics, and tails in British treatments), facts long advocated by both American (Pike Reference Pike1945) and British (O'Connor & Arnold Reference O'Connor and Arnold1963) authors and now reflected in the work of laboratory phonologists (e.g., Ladd Reference Ladd2008) who use the autosegmental approach pioneered by Pierrehumbert (Reference Pierrehumbert1980) for intonational description and meaning (e.g., Pierrehumbert & Hirschberg Reference Pierrehumbert, Hirschberg, Cohen, Morgan and Pollack1990). In this newer approach, intonation is described in terms of High and Low tones in three domains: pitch accents (emphasizing particular syllables), phrase accents (at the end of an intermediate or utterance-final phrase) and boundary tones (at an utterance-final phrase boundary). Thus, a falling intonation contour would be described as HLL (an H pitch accent followed by an L phrase accent and boundary tone). In this description we see both the smaller meaningful elements and the overall contour. Fourth, intonation is interactional in conversation (Szczepek Reed Reference Szczepek Reed and Trillo2012) in such areas as turn-taking and regulating conversational interaction. Fifth, intonation has many more patterns than rising and falling, including falling-rising, and rising-falling levels (e.g., Brazil Reference Brazil1997), as well as de-accented (low pitch, non-highlighted) syllables following nuclear stress. In discourse, de-accenting can be a signal of given information being present (Wichmann Reference Wichmann2000: 74–75).

Although intonation is taught in almost every pronunciation book and in many general English as a second language (ESL) textbooks (Derwing, Diepenbroek & Foote 2012), few current research findings are found in teaching materials, which remain focused on intonation's role vis à vis grammar (Levis Reference Levis1997). In British materials, the discourse intonation approach of David Brazil is influential in applied circles but it has little overlap with theoretical and laboratory research into intonation. Historically, theory and practice were closely related in both North America and the UK (e.g., Pike Reference Pike1945; O'Connor & Arnold Reference O'Connor and Arnold1963). As a result, intonation teaching was descriptively and pedagogically richer in the past than today. Currently, intonation teaching tends to be limited to rising and falling patterns at the ends of sentences, with a direct connection to grammatical form. This limited account does not reflect normal spoken discourse, in which intonation continues a spoken turn or ends it, signals idea boundaries, achieves cohesion, and builds rapport (Levis & Wichmann Reference Levis, Wichmann, Reed and Levis2015). Bridging the divide between research and practice (as in Wennerstrom Reference Wennerstrom2001 or Chun Reference Chun2002) will be essential to transform intonation teaching.

3.1 Pronunciation's role in expressing information structure

While current research on intonation (in the sense of the role of pitch movement at the ends of phrases/sentences) is not well represented in teaching materials, intonation's role (through nuclear stress) in marking discourse information structure (Halliday Reference Halliday1967) and in signaling contrasts is widely found in pronunciation materials. By ‘nuclear stress’ I refer to the emphasis given to a syllable (the nucleus) in a phrase, e.g., Where are you from? The United states. This emphasis is signaled by a combination of pitch change, increased duration, change in loudness, and clear articulation of segmentals. This feature of speech has been called tonic, sentence stress, prominence, primary phrase stress, focus, etc. The term ‘nuclear stress,’ though old, seems to be reliably understood by pronunciation researchers around the world, which is not true of the other terms. Appropriate use of nuclear stress is essential in both greater understanding of discourse and in how listeners evaluate speakers (Hahn Reference Hahn2004). It also is particularly salient to listeners (Pennington & Ellis Reference Pennington and Ellis2000). Nuclear stress also calls attention to categorical contrasts (e.g., ‘that's a fabulous sportscar’ as opposed to whatever other sportscars were being discussed), another element of information structure that appears to be very learnable (Pennington & Ellis Reference Pennington and Ellis2000; Muller Levis & Levis Reference Muller Levis, Levis, Levis and LeVelle2012). Finally, nuclear stress offers a manageable and realistic connection to discourse in pronunciation teaching.

The role of nuclear stress in marking contrasts and information structure is consistently represented in textbooks. Examples of North American books presenting contrasts and/or given/new information are Reed & Michaud (Reference Reed and Michaud2005), Henrichsen et al. (Reference Henrichsen, Green, Nishitani and Bagley1999), Miller (Reference Miller2000), Grant (Reference Grant2010), and Gilbert (Reference Gilbert2012). An alternative view of information structure is found in books following the work of David Brazil, in which the nuclear tone (i.e., the nuclear syllable and its subsequent pitch movement) marks given and new information, with rising tones signaling given information and falling tones new information. Level tones, on the other hand, mark routine information and make no spoken commitment to the information structure of an utterance. Bradford (Reference Bradford1988) says that ‘speakers use falling tones in parts of utterances which contain information they think is new for their hearers – when they are telling them something they don't already know’ (p. 12) and that they use [rising] ‘tones in parts of utterances which contain ideas they think their hearers already know about or have experience of’ (p. 13). Contrasts in this approach to intonation are not tied to the given/new distinction but are a function of emphatic stress or ‘key’ in Brazil's (Reference Brazil1997) terminology.

Advocates of both approaches agree that information structure is essential in teaching materials. Each approach, however, suffers from not utilizing current research literature on contrast and given/new information. Most of this work is conducted by theoretical and laboratory phonologists (e.g., Baumann & Grice Reference Baumann and Grice2006), and by those working in pragmatics (Riesco-Bernier & Romero-Trillo Reference Riesco-Bernier and Romero-Trillo2008). It is clear that the placement of nuclear stress cannot be fully accounted for by either model of information structure, and that speakers have enormous freedom in what they call attention to or emphasize (Bolinger Reference Bolinger1972). Materials developers should consult these studies to develop more complete approaches to given/new information (Levis Reference Levis2014).

3.2 Setting instructional priorities

Most pronunciation teaching materials assume that we must set priorities. This assumption is based on simple reality (we never have enough time) but also on research into intelligibility and comprehensibility (Munro & Derwing Reference Munro and Derwing1995), functional load (Catford Reference Catford and Morley1987; Brown Reference Brown1988; Munro & Derwing Reference Munro and Derwing2006), and the Lingua Franca Core (LFC) (Jenkins Reference Jenkins2000). Setting priorities reflects a change from a focus on native-like attainment to intelligibility (Levis Reference Levis2005). Intelligibility for language teaching is not a recent concept (Abercrombie Reference Abercrombie1949) but it finally took hold with the rise of the World Englishes movement and the desire to quantify mutual intelligibility among outer and expanding circle speakers (Smith & Nelson Reference Smith and Nelson1985). Munro & Derwing (Reference Munro and Derwing1995), whose paradigm is now dominant in applied research on intelligibility and comprehensibility, use similar terms (‘intelligibility’, ‘comprehensibility’, ‘accentedness’) with slightly different meanings in a research approach based on the work of James Flege.

Although some textbooks continue to be comprehensive in their inclusion of all pronunciation targets (e.g., Orion Reference Orion1997), evidence of prioritizing is evident in the relative placement of pronunciation targets in books, and in what is not included. Grant (Reference Grant2001) makes an explicit appeal to intelligibility, saying her book ‘is designed to improve the speech intelligibility of high intermediate to advanced learners’ (p. vi). Gilbert (Reference Gilbert2012) extends intelligibility to listening comprehension as well, while stating that her book deals with the most important topics first. Intelligibility may sometimes seem to be a buzzword (Munro Reference Munro, Levis and LeVelle2011) rather than a strong principle in textbooks, though even the use as a buzzword suggests the currency of the idea among practitioners. For example, while Miller (Reference Miller2000) talks about the importance of suprasegmentals for intelligibility, she talks more about changing accents (a traditional term), giving a teacher the sense that there are priorities, but that reducing accents is the path to greater intelligibility.

A second prioritizing principle is that of functional load. Functional load is a measure of how much work two phonemes do to distinguish words in a language (Brown Reference Brown1988). For example, in English, /l/ and /r/ have many minimal pairs, both at the beginning and ends of words, while /l/ and /j/ do not; thus /l/ – /r/ is a high functional load contrast. Munro & Derwing (Reference Munro and Derwing2006) provide evidence of the validity of functional load for prioritizing instruction. Books that provide limited materials for working on segmentals show inconsistent evidence of appealing to functional load. Grant (Reference Grant2010), for example, prefers to work only on troublesome sounds, but these sounds contain both high and low functional load contrasts (e.g., /θ/ vs. /s, t, f/ and /u/ vs. /ʊ/). Prioritizing is evident, given that only five vowel and five consonant lessons are provided, but the prioritizing in this case includes not only research findings but teacher desires for certain contrasts (personal communication, June 2014).

A third prioritizing principle is Jenkins’ (Reference Jenkins2000) LFC in which only those targets likely to improve intelligibility for NNS-NNS (nonnative speaker) communication should be taught. Walker (Reference Walker2010) provides a pedagogical application of Jenkins's recommendations, while Deterding (Reference Deterding2013) has recently provided more data for NNS-NNS communication. Jenkins's LFC (based on a limited data-set) is largely segmental in nature, with some sounds left out, especially /θ/, /ð/, and [ɫ]. The only suprasegmental to be included in the LFC is nuclear stress. Jenkins’ work, though ostensibly meant only for NNS-NNS communication, was invaluable in making many researchers and teachers address the issue of prioritization (e.g., Dauer Reference Dauer2005), and its influence in making priorities a centerpiece of pronunciation discussion is unrivaled. Her recommendations were taken seriously by those operating in English as a second language (ESL) and English as a foreign language (EFL) contexts, perhaps because they understood that it is impossible to leave out native speakers (NSs) as communicative partners. They recognized that NNS-NNS communication could not exclude these other contexts, and that the LFC's recommendations spoke also to them.

Although much work remains to be done on priorities, there is explicit recognition in most materials that not all possible targets merit equal attention in classrooms. Some are not worth the effort of either teachers or learners. However, research has yet to establish definitively what will make the biggest difference to intelligibility and comprehensibility.

4.1 Connected speech processes

Connected speech processes (CSPs) reflect the pronunciation differences of words in discourse versus their citation forms (e.g., gonna vs. going to). CSPs include linking (one of), deletion (Did he go?), insertion (We are said as We yar), modification (can't you as can chew), and reductions (such as the consistent reduction of vowels in function words), all of which occur alone or in combination (Alameen & Levis in Reference Alameen, Levis, Reed and Levis2015). During the third wave revival of pronunciation teaching in the mid-1980s, research into CSPs, especially linking, was carried out by Henrichsen (Reference Henrichsen1984) and Hieke (Reference Hieke1984, Reference Hieke1987). Highly salient CSPs (e.g., gonna, wanna, whaddya) also gained pedagogical attention from Weinstein (Reference Weinstein1982). Following this there was a paucity of research into CSPs from the 1980s to the mid-2000s, when some descriptions of CSPs (Shockey Reference Shockey2003; Cauldwell Reference Cauldwell2013) were published in areas typically read by applied linguists and teachers. In speech recognition publications, however, massive phonetic changes of connected speech were addressed extensively because of their effect on recognizer accuracy rates (e.g., Greenberg Reference Greenberg1999).

CSPs in English are heavily influenced by the rhythmic timing of English syllables, especially variation in syllable length and clarity between stressed and unstressed syllables. In general, unstressed vowels and coda consonants in discourse are more prone to changes such as deletion, insertion, modification, and reduction (Greenberg Reference Greenberg1999), while all word boundaries are prone to linking. After the 1980s, the inclusion of CSPs in textbooks became nearly universal. When widespread research into pronunciation began to be evident in the mid-1990s, CSPs were largely absent, returning with an edited volume by Brown & Kondo-Brown (Reference Brown and Kondo-Brown2006) highlighting work by Japanese teachers and scholars. A practical handbook on teaching CSPs was recently published by Brown (Reference Brown2012). Another recent locus of research into CSPs is Taiwan, but this research remains little known (e.g., Lee & Kuo Reference Lee and Kuo2010; Lu & Kuo Reference Lu and Kuo2011).

CSPs are ubiquitous in published teaching materials but their importance sometimes seems overstated. ‘A few of the more useful rules for blending [modification] and linking follow. These guidelines will help your comprehension as well as your pronunciation of English’ (Grant Reference Grant1993: 157). Although CSPs are useful for both parts of the communication process (perception and production), CSPs are typically taught as especially useful for production. Indeed, this bias toward production is true of most pronunciation materials. Older research on CSPs recognized the value of CSPs for listening comprehension (Hill & Beebe Reference Hill and Beebe1980; Brown & Hilferty Reference Brown and Hilferty1986), and this focus is now evident in more recent research in which CSPs are tied to micro-listening involving bottom-up processing of speech that is critical in listening comprehension, especially in word recognition and lexical decoding. Field (Reference Field2003, Reference Field2008) persuasively argues for the importance of bottom-up processing, a goal related to CSPs but seldom reflected in pronunciation materials. This raises the question of how CSPs are related to intelligibility and comprehensibility. An answer is not yet evident. Most assertions about CSPs are based on little more than teacher intuition. Prator & Robinett (Reference Prator and Robinett1985), in an unusually ambitious treatment of sandhi (i.e., CSPs) said that ‘it is not easy to draw the line between forms you should be able to use, those you need only to understand, and those you are not likely to find useful or appropriate. NSs are far from unanimous in judging such matters, and decisions must necessarily be very subjective’ (p. 198). That they remain subjective three decades later is telling. What we know is made up of small clues collected over many years, like finding pieces of a puzzle that are never sufficient to give a sense of the whole picture.

4.2 The pedagogical superiority of suprasegmentals

The reinvigoration of pronunciation teaching in the 1980s occurred for many reasons. Not only was there recognition that pronunciation would not take care of itself in communicative language teaching (CLT) (Hinofotis & Bailey Reference Hinofotis, Bailey, Fisher, Clarke and Schachter1981), there was also growing recognition that pronunciation was central in communicative success in specialized registers such as international teaching assistants’ lectures (e.g., Tyler, Jefferies & Davies Reference Tyler, Jefferies and Davies1988). More important, however, was a greater emphasis on suprasegmentals for short-term changes to intelligibility. Suprasegmentals fit much better with the communicative and discourse-oriented goals of CLT because suprasegmentals are not constrained by the sound or word level. In perhaps the clearest statement of the superiority of suprasegmentals, McNerney & Mendelsohn (Reference McNerney and Mendelsohn1987, Reference McNerney, Mendelsohn, Avery and Ehrlich1992) claimed that suprasegmentals should be the priority in a short-term pronunciation class because they were far more likely to lead to noticeable improvement than teaching segmentals. How to make suprasegmentals a priority was not always clear. For example, Celce-Murcia (Reference Celce-Murcia and Morley1987) advocated teaching pronunciation communicatively ‘through tasks that focus on meaning’ (p. 6), but admitted that she was unsuccessful in applying her ideas to suprasegmentals, saying ‘the one glaring omission in my current approach is that I am still having problems with fully integrating stress and intonation into my teaching of English pronunciation’ (p. 11).

Nonetheless, the idea that suprasegmentals were more important, more likely to lead to quick improvement, and more central for teaching was a seed that fell onto fertile ground. During the 3rd wave of the history of pronunciation (Murphy & Baker Reference Murphy, Baker, Reed and Levis2015), textbooks and teacher training materials began to emphasize suprasegmentals in a new way (e.g., Kenworthy Reference Kenworthy1987; Wong Reference Wong1987). Although intonation had earlier received serious and extensive treatments for both American and British English (Pike Reference Pike1945; O'Connor & Arnold Reference O'Connor and Arnold1963), and even during the neglected decade of the 1970s (Allen Reference Allen1971), information about how to teach suprasegmentals was scarce and pronunciation teachers were hungry for information.

Unfortunately, the research establishing the superiority of suprasegmentals was, and is, scarce. Derwing, Munro & Wiebe (Reference Derwing, Munro and Wiebe1998) demonstrated that a pedagogical focus on suprasegmentals led to improvements in comprehensibility in extemporaneous speech and reading tasks. While a pedagogical focus on segmentals also led to significant comprehensibility improvement in sentence reading, listeners did not perceive improvement in comprehensibility in the extemporaneous speech task ‘when the nature of the linguistic task necessitated that attention be divided among multiple areas of linguistic form’ (p. 406). Saito (Reference Saito2012), in a review of 15 studies looking at the effect of instruction on pronunciation improvement, found no discernible pattern of greater improvement for suprasegmentals, although most studies did not compare improvements of the two areas. Zielinski (Reference Zielinski2008) makes it clear that deciding whether segmentals or suprasegmentals are more important is a false choice: segmentals are dependent on suprasegmentals in English, and suprasegmentals always have segmental effects. Although few experienced researchers or teachers today draw an either/or distinction for suprasegmentals and segmentals, it is my impression that many teachers and materials developers still want to draw such distinctions.

The second problem with an uncritical focus on suprasegmentals is that, as in the case of segmentals, not all suprasegmentals appear to be equally learnable or profitable. Pennington & Ellis (Reference Pennington and Ellis2000), in a study of the ability of Cantonese learners of English to perceive four different stress/intonation features (juncture, tag question intonation, compound stress, and contrastive stress), found that instruction was effective only for contrastive stress. In another study, Levis (Reference Levis1999) argued that rising and falling intonation distinctions were much more important for incomplete sentences and declaratives than for yes-no questions with inversion, where intonational distinctions were less clearly connected to meaning. Even for features that can be predicted and seem sensitive to instruction, such as nuclear stress, not all patterns seem to be equally useful. Hahn (Reference Hahn2002) found that while instruction on predictive patterns for nucleus placement demonstrated improvement, many patterns in delayed post-tests reverted to initial levels of performance or even got worse. Only a few prediction rules demonstrated long-term learning. For word stress, Field (Reference Field2005) demonstrates that rightward misstressing on 2-syllable English words (e.g., offer) seemed to affect intelligibility more than leftward misstressing (e.g., lagoon). It is clear that not all suprasegmentals are equally important.