Reasons for an Intelligibility-Based Approach to Teaching

In 1981, when pronunciation was not being widely taught during the early days of the Communicative Approach, Hinofotis and Bailey conducted research into the effectiveness of foreign teaching assistants (FTAs) in US universities. Foreign teaching assistants (now usually called ITAs, or international teaching assistants) often taught or assisted in introductory-level classes in academic fields such as chemistry or math. It was important for these FTAs to be understandable. In their research, Hinofotis and Bailey found that although some FTAs were sufficiently expert in the course content and that their overall language proficiency was advanced, they were still not easily understood. The researchers argued that their results meant that speakers must have a threshold level of pronunciation ability to communicate and that pronunciation had a special role in promoting intelligibility.

Hinofotis and Bailey (Reference Hinofotis, Bailey, Cameron Fisher, Clarke and Schachter1981) questioned the dichotomy of native and nonnative speech, suggesting instead that we need a more nuanced continuum. Foreign language learners may not be able to achieve the pronunciation abilities of NSs, but progress in pronunciation is both possible and essential. Learners must have sufficient command of pronunciation or else their speech will not be sufficiently understood. It will not matter if they speak fluently (smoothly and without unnecessary hesitations), have sufficient vocabulary, use perfect grammar, or are pragmatically appropriate in their language use.

An intelligibility-based approach to oral communication is thus the pursuit of skilled rather than native speech. While pronunciation is important in speaking (after all, one cannot speak without pronouncing), it is not the only component of effective speech. Speakers must choose appropriate vocabulary and grammar, and must speak with sufficient fluency to communicate. All elements of speech are important in getting one’s message across, and all must combine in such a way that they are within a listener’s expected range of production. None of these elements of speech needs to be perfectly native-like, but they all need to be good enough. But what is good enough?

Since 1981, research and many teaching materials have sought to define what features are important in developing a threshold level of intelligibility, typically centered on the rough dichotomy of suprasegmentals (stress, rhythm, intonation) and segmentals (consonants, vowels). Many have advocated greater emphasis on suprasegmentals than segmentals (e.g., McNerney & Mendelsohn, Reference McNerney, Mendelsohn, Avery and Ehrlich1992) to promote intelligibility because of the belief that suprasegmentals lead to faster improvement in comprehensible speech. Because time and energy for pronunciation instruction is always limited, this faster rate of improvement is important. Derwing and Rossiter (Reference Derwing and Rossiter2003) found that comprehensibility ratings for learners who were taught with a more global approach (focusing more heavily on suprasegmentals) improved significantly more than did those for learners who were taught with a focus on consonants and vowels (a segmental approach).

Research has also found a differential importance for errors. Some errors cause greater impairments to intelligibility than others, a fact that is at the heart of the suprasegmental–segmental question. The principle of differential importance also applies within categories of errors. For example, Munro and Derwing (Reference Munro and Derwing2006) examined the principle of differential importance by looking at the comprehensibility ratings of two kinds of consonant errors: /θ/–/f/ (as in think–fink) and /l/–/n/ (as in light–night). They found not only that the /l/–/n/ errors had a greater effect than /θ/–/f/ on listeners being able to understand a sentence, but also that one /l/–/n/ error had a greater effect than a combination of two or three /θ/–/f/ errors. This is attributed to the different functional load of the two types of errors. The first has few minimal pairs, and so has a low functional load, while the second has many minimal pairs, and a high functional load with greater opportunities for confusion.

This principle of differential importance applies in many contexts outside language teaching. It is known as the Pareto principle, or the 80–20 rule. The principle may suggest, for example, that 20 percent of workers create 80 percent of the wealth, or even that 20 percent of the students take up 80 percent of a teacher’s time and energy. Applied to pronunciation teaching, then, we can guess that the largest improvement in a learner’s intelligibility may come from improvement in a minority of the learner’s errors.

If we can determine which errors constitute that important minority of errors, we can target our teaching toward the few vital errors, and deemphasize or ignore the other, less important errors, either because they will not make a difference or because many of them will be remedied in the normal course of development (as with some vowel errors or consonant errors; see Munro & Derwing, Reference Munro and Derwing2008 and Munro, Derwing, & Thomson, Reference Munro, Derwing and Thomson2015, respectively). Almost all pronunciation materials focus overwhelmingly on vowel and consonant sounds, including sound contrasts that are unlikely to be a problem for any learner or listener. For example, one can find exercises teaching the difference between the interdental fricatives /θ/ and /ð/, as in ether–either, despite the lack of evidence that it is worthwhile to teach this contrast. Most learners of English have plenty of trouble with both sounds, but the two sounds are potentially confusable with only a few infrequent words, and if learners have trouble with either sound, they are likely to confuse it with other sounds such as /s, t, f/ and /z, d, v/, not with the other interdental fricative.

What Is Intelligibility?

Intelligibility is widely agreed to be the most important goal for spoken language development in a second language – both for listening and speaking – no matter the context of communication. In ESL (English as a second language) contexts, in which learners are surrounded by the target language outside the classroom, and in English as a lingua franca (ELF) contexts, where nonnative speakers communicate primarily with other nonnative speakers, researchers have called for a focus on intelligibility.

But what is intelligibility? This is not an easy question to answer, it seems, because the word has often been used with two distinct but related meanings. Munro and Derwing suggested that “intelligibility may be broadly defined as the extent to which a speaker’s message is actually understood by a listener” (Reference Munro and Derwing1999, p. 289). This broad description of intelligibility presumes that listeners understand speech in a variety of ways: that they can identify the words that are spoken, that they understand the message, and that they understand the intent behind the message (Smith & Nelson, Reference Smith and Nelson1985). Broadly or narrowly conceived, intelligibility does not include speech that is irritating or socially unacceptable. For example, the use of taboo words in some contexts may be socially unacceptable but not unintelligible. The overuse of filler words such as like or umm may be irritating, but it does not necessarily create problems with understanding. However, intelligibility may be compromised by other features of the communicative environment, such as noise (Munro, Reference Munro1998) or a speaking rate that is challenging (Munro & Derwing, Reference Munro and Derwing1998).

Munro and Derwing (Reference Munro and Derwing1995), whose model of spoken language understanding is based on the work of James Flege, include three distinct but partially overlapping types of judgments that listeners make about L2 speech. These are intelligibility, comprehensibility, and accentedness (Figure 1.1).

Figure 1.1 Three types of spoken language understanding

(Munro & Derwing, Reference Munro and Derwing1995)

Intelligibility means both the extent to which a speaker is understandable, and whether the particular words used by a speaker are successfully decoded (the lexical level of intelligibility). In the narrow sense, intelligibility may be measured through transcription or dictation tasks, in which listeners are asked to write down exactly what they heard. If listeners cannot successfully decode particular words, the words are defined as unintelligible. In a broader sense, listeners may be asked to answer comprehension questions or provide summaries of what they understood (e.g., Hahn, Reference Hahn2004), thus demonstrating intelligibility at the semantic level.

The second type of understanding is comprehensibility, defined as the amount of work that listeners need to do in understanding a speaker. This is often measured via a Likert scale, on which 1 = extremely easy to understand and 9 = impossible to understand (Munro & Derwing, Reference Munro and Derwing1999, p. 291), or through continuous sliding scales on a computer (e.g., Crowther, Trofimovich, Saito, & Isaacs, Reference Crowther, Trofimovich, Saito and Isaacs2015). Thus comprehensibility taps into listeners’ overall sense of how easy speech is to understand and highlights difficulties in speech processing, feelings about how difficult it is to listen to a speaker, and “factors … that [do] not necessarily determine whether an utterance [is] fully understood” (Munro & Derwing, Reference Munro and Derwing1999, p. 303). Listeners are very reliable with these kinds of holistic ratings. Tapping into NSs’ intuitions also has a long history in linguistic research.

What is the relationship between intelligibility and comprehensibility? Munro and Derwing (Reference Munro and Derwing1999) found that there was a correlation between ratings of intelligibility and ratings of comprehensibility, but that the two dimensions did not measure the same things. It was not uncommon for listeners to transcribe speech perfectly (that is, to find it completely intelligible) and yet perceive it to be difficult/effortful to understand.

Finally, accentedness is important to consider because accent is commonly appealed to, both popularly and in research, for its presumed connection to whether speech is understandable and its potential usefulness in assessing spoken language (e.g., Ockey, Papageorgiou, & French, Reference Ockey, Papageorgiou and French2016). Clearly, accent can interfere with intelligibility, but research has shown that speakers can be perfectly intelligible (in that all of the words they speak are understood) while being simultaneously judged as having a very strong accent. Accentedness is also scalar, and is defined as the degree of difference between speech and a local or reference accent (Munro & Derwing, Reference Munro and Derwing1995). Because it is scalar, an assumption demonstrated by Derwing and Munro (Reference Derwing and Munro2009), it is possible to talk about listeners hearing speech as more or less accented in comparison to the reference accent or in comparison to another accent. However, describing accents as thick, heavy, or light without some kind of measured judgment task is unhelpful. Accentedness judgments of this sort seem particularly prone to bias based on stereotypes and other social judgments, perhaps leading to perceptions of accented speakers being less truthful (Lev-Ari & Keysar, Reference Lev-Ari and Keysar2010), less understandable (Rubin, Reference Rubin1992), or having poorer teaching skills (Major, Fitzmaurice, Bunta, & Balasubramanian, Reference Major, Fitzmaurice, Bunta and Balasubramanian2002).

In summary, listeners’ ability to decode words successfully does not guarantee that speech will be easy to understand. This kind of intelligibility is not only a matter of how many words are not understood, but also which words. Misunderstood function words may have less impact than misunderstood content words (Zielinski, Reference Zielinski2006). It is likely, however, that speech will be less comprehensible if words are frequently unintelligible. What we do not know is how many words can be missed by a listener before overall understanding is severely impaired. Nation (Reference Nation2002) suggested that nonnative listeners need to understand at least 95 percent (nineteen of every twenty words) or perhaps 98 percent (forty-nine of fifty words) of the content words in running speech in order to have excellent comprehension and be able to guess the meanings of unknown words. The threshold for NSs, with better guessing skills and faster processing of their first language, is likely to be lower, but we do not know how much lower.

A potential confusion of intelligibility and comprehensibility comes from Smith and Nelson (Reference Smith and Nelson1985), who divided their concept of “international intelligibility” into three different types of understanding, which they named intelligibility, comprehensibility, and interpretability (Figure 1.2). International intelligibility is listed as an overarching term based on Smith and Nelson’s attempt to describe understanding across different varieties of world Englishes. It is important to note that, despite the terms used, all three types of understanding in Smith and Nelson’s work are types of intelligibility – that is, they are all types of actual understanding. None of them is the same as comprehensibility as discussed by Munro and Derwing (Reference Munro and Derwing1995).

Figure 1.2 Three types of understanding thought to constitute international intelligibility

(Smith & Nelson, Reference Smith and Nelson1985)

Smith and Nelson (Reference Smith and Nelson1985) define intelligibility as a listener’s ability to decode the words that are spoken. In other words, this is lexical intelligibility. From the speaker’s point of view, this is the ability to say words in such a way that listeners can successfully decode them. For example, I once was listening to a talk given by the Australian applied linguist, Wilga Rivers. I had no difficulty understanding anything she said even though her accent was largely unfamiliar to me at the time. Then she used a word that I didn’t understand at all – [kəˈ ɹɑlɹi]. I spent almost a minute trying to decode the word, during which time I did not hear any of the rest of her talk. First, I searched my mind for any word that seemed to fit. Finding none, I spelled out the sounds I heard <c_rol_ry> and then visually was able to see a word that fit: corollary. Then, and only then, did I realize that I had heard that British-influenced varieties of English used this stress pattern for this word that is pronounced in North American English as [ˈk^hɔɹəˌlєɹi]. The speaker’s unexpected stress pattern and consequent vowel and consonant pronunciations of this word caused a loss of intelligibility in that I was not able to successfully decode the word in real time.

In Smith and Nelson’s work, comprehensibility involves the ability of a listener to understand a spoken message. Again, this use of comprehensibility is not the same as that of Munro and Derwing (Reference Munro and Derwing1995). Comprehensibility in this model refers to “the meaning of a word or an utterance” (Smith & Nelson, Reference Smith and Nelson1985, p. 334). A text is comprehensible if the listener can make sense of or paraphrase it. Smith and Nelson’s (Reference Smith and Nelson1985) third type of understanding, interpretability, refers to understanding the “meaning behind the word or utterance” (p. 334). The interpretability of a spoken text is high if the listener is able to understand the speaker’s intentions in uttering the message in a particular context. Interpretability refers to another type of intelligibility, pragmatic intelligibility, but this type of intelligibility can be extremely variable. It is also difficult to determine whether a listener has misunderstood the meaning behind the word or utterance. Listeners may think that they understand, and a speaker may think that a listener has understood the meaning that the speaker intended, only to find out later that the listener’s understanding was quite different from the speaker’s intended meaning. Interpretability has not played a significant role in research involving pronunciation. In one study that suggests a role for this type of intelligibility, Low (Reference Low2006) described how given information is marked as (non)prominent by British English (BE) and Singapore English (SE) speakers. It is typical for BE speakers to make given information less prominent when it is repeated (e.g., the second use of “car” in the example below). SE speakers, rather than making car less prominent, appear to make it prominent again. Low suggested that this difference caused a loss of interpretability, in that inner-circle speakers misunderstood the communicative intent of SE speakers:

This book is primarily about intelligibility, but it is also about comprehensibility. When the word intelligibility is used, it is used to mean understanding of a spoken message, either in regard to lexical identification, meaning, or intention. When I speak of comprehensibility, I mean the degree of difficulty a listener has in understanding. Loss of intelligibility can happen between two NSs, and in fact any mismatch between a listener’s expectation and the speaker’s production can cause it (Zielinski, Reference Zielinski2008). However, mismatches do not always impair understanding. They may, however, make a difference in how easily a listener understands a speaker, and consequently how likely a listener is willing to communicate with different individuals. If it takes extra effort to understand someone, interlocutors may avoid contact.

Intelligibility and Teaching Pronunciation

In teaching pronunciation in relation to discussions of intelligibility, we often make assumptions about who the listener and the speaker are in an interaction. Typically, we assume that the listener is an NS while the speaker is a nonnative speaker (NNS). Thus, we assume that the NNS bears the burden for being understood, while the NS, who has a command of the language and is able to determine whether an NNS is intelligible, bears less of a burden. Because any interaction involves both speakers and listeners playing both roles, this need not be the only, or even the most helpful, way to understand intelligibility. As I have written elsewhere (Levis, Reference Levis2005a), intelligibility can be understood through a matrix of possible relationships between listeners and speakers.

Figure 1.3 delineates possible relationships between listeners and speakers, and each of the relationships has implications for intelligibility, and in turn for pronunciation instruction.

Figure 1.3 Speaker–listener intelligibility matrix

(Levis, Reference Levis2005a)

In Quadrant A, both the speaker and listener are NSs. Implicitly, communication in this quadrant should be easy and successful since both the listener and speaker share not only the same linguistic code but many of the same rules for using it as well as intuitions about its unspoken rules. Unfortunately, we do not know how far we can expect such success to reach. Native speakers come in many flavors, with wide varieties of dialects and ways of speaking. It is not hard to recall interactions between NSs that include a loss of intelligibility. Once, while playing golf in Scotland, I played with two other golfers, one native to Scotland and the other native to the north of England but who had lived in Scotland for decades. More than once, their speech became increasingly unintelligible to me, especially as topics became more emotionally charged. I didn’t understand many words, the meaning, or the intent of the messages being communicated.

Quadrant B, with native speakers and nonnative listeners, is often found in language classrooms and when NNSs interact with NSs in second-language contexts. This is the flip side of typical intelligibility research which focuses on nonnative speakers and native listeners. In this quadrant, NNS listeners are tasked with understanding normal NS speech, with all its variations, in real time. A lack of intelligibility in this case means that NNSs cannot understand. In other words, the NS may be unintelligible to the NNS. From the viewpoint of a pronunciation teacher, the goal must be to help learners decode and process speech as quickly and easily as possible. In real-world communicative interactions, NSs must carry a large part of the burden for intelligibility, not only in phonology but also in adjustments, speech rate, etc., if some interactions are to be successful.

Quadrant C reflects the bulk of the research on intelligibility, in which NNSs are the speakers and NSs the listeners. The artificial division between Quadrant B and Quadrant C is unfortunate, because in reality any interaction between NSs and NNSs means that NSs and NNSs will play both roles. The important thing to notice, though, is that it is possible for both NSs and NNSs to be unintelligible when communicating, and conversational adjustments are common by NSs (for an early review, see Long, Reference Long1983). In Quadrants B and C, the NS often places the responsibility for success on the NNS, but the NNS cannot be fully responsible for successful communication. Success in any communicative interaction involves the ability to speak and listen and to adjust if communication is not happening. If NSs cannot adjust, they are also partly responsible for a lack of intelligibility.

Finally, Quadrant D involves speakers and listeners who are both NNSs. This is typical of the ELF setting. This quadrant describes the widespread use of English as a medium of communication for speakers who have different non-English first languages. In a smaller way, it also reflects the need for learners in a second-language classroom setting to communicate with other learners of different L1s. I have often had students ask if they could practice with me in class and not with other learners, because they believed the other learners’ grammar and pronunciation were bad. This request presumes a kind of osmosis in that other learners’ errors will somehow rub off and ruin their progress, while my native speech will rub off and lead to the kind of improvement they desire. Such osmosis is unlikely to work – either for good input or for bad. Findings about correction show that learners don’t seem to correct fellow students’ errors, and thus presumably they don’t notice them (Long & Porter, Reference Long and Porter1985).

All four quadrants reflect different intelligibility issues. Intelligibility is always a two-way street, and both NS and NNS interlocutors carry the responsibility for successful communication. It is only by recognizing the contexts in which we talk about intelligibility issues that we can achieve greater success in our teaching.

Pronunciation and Oral Communication

Although this book is primarily about the role of pronunciation in intelligibility, it is impossible to discuss pronunciation outside the larger context of oral communication. Pronunciation, with all its intricacies and interesting features, does not exist on its own. It is a servant skill. We pronounce as a byproduct of speaking, and we pronounce within the context of the communicative acts or the speaking task we are involved in. We pronounce if we are fluent, and we pronounce if we are not fluent. We pronounce with our particular lexical and grammatical skills. Likewise, we process the pronunciation features of others’ speech, presented to us within the fluency, speed, grammar, and lexis of their speech. Sometimes the speech will be more careful, and sometimes it will be more casual. Pronunciation (or its primary correlate, accent) seems to play such an outsized role in intelligibility because it is a first stop for listeners, and many of its features are those that listeners are very good at noticing. We notice small deviations of vowel and consonant sounds even when listening to other NSs, and we often make social judgments about others because of these differences. For example, the insertion of /ɹ/ into the word wash so that it is pronounced warsh is immediately noticed by most US speakers and is almost always criticized as sounding uneducated. Likewise, the deletion of [h] in words like house and hospital in British English has historically been highly stigmatized, indicating social class distinctions (Mugglestone, Reference Mugglestone1996), despite the fact that other [h] deletions in function words (e.g., Did he give him the book?) are usually unnoticeable in normal conversational speech.

Word-Based Features

Word-based features include segmentals and word stress. These features are most likely to affect intelligibility at the lexical level. In some languages (e.g., Mandarin, Vietnamese, Kikuyu), this category would also include lexical tone. First, some justification of my classification is in order. Typically, pronunciation features are described as being either segmental or suprasegmental. This classification scheme is central to teaching materials, but it creates an unnatural division in discussing intelligibility. Consonant and vowel errors primarily cause listeners to have trouble decoding individual words. This is the logic behind the use of minimal pairs in teaching pronunciation. For example, the numerous research studies that have examined the difficulty that Japanese learners of English have with /l/ and /ɹ/ all reflect the many minimal pairs that these two phonemes have in English. Brown (Reference Brown1988), extending Catford (Reference Catford and Morley1987), described one effect of the number of minimal pairs that two phonemes have as functional load. Functional load is measured by the numbers of minimal pairs that exist for a contrast (e.g., [p]/[b]) initially and finally, and the likelihood that the contrast is enforced in all varieties of English. Minimal pairs of segments with a high functional load are thought to be more likely to cause loss of intelligibility.

In addition to minimal pairs, another segmental feature that can cause difficulties in understanding words involves allophonic variations. For example, voiceless stops and affricates ([p], [t], [k]) in inner-circle English varieties are aspirated at the beginning of stressed syllables. If the sounds are not aspirated, there is a strong chance of unintelligibility because inner-circle listeners are likely to hear them as their voiced counterparts ([b], [d], [g]).

Vowel reduction is also a strong candidate for intelligibility problems, especially from the second-language learner’s point of view. Because English NS speech is full of vowel reduction (Woods, Reference Woods2005), L2 learners of English may find NS speech unintelligible when they cannot easily connect the speech signal to the written representation of the words.

Likewise, word-stress errors also cause trouble with intelligibility at the lexical level. Stressing a word incorrectly can not only mask the identity of the word (Zielinski, Reference Zielinski2008), it can also change how the vowels and consonants are pronounced. Although word stress is typically considered a suprasegmental feature, I will treat it similarly to segmentals in its effect upon intelligibility. In other words, the kinds of interpretive problems related to word stress are likely to be related primarily to intelligibility at the lexical level.

Discourse-Based Features

Most suprasegmental errors are unlikely to cause listeners to misidentify words, but may cause loss of intelligibility in understanding the message or the speaker’s intent. They may also cause decreased comprehensibility by increasing the effort listeners put forth in understanding speech. Most suprasegmental features, unlike segmental features, carry their meaning beyond the sentence level. As such, these features are part of the overall structure of spoken English, allowing it to be spoken with its characteristic rhythm and melody, and creating meaning that is pragmatic, attitudinal, and connected to the communication of information structure.

Unlike word-based features, discourse-based features in English are rarely right or wrong. If a Korean speaker who does not consistently distinguish the sounds [p] and [f] says the name Pat as fat, there is a clear sense of right or wrong. However, if the same Korean speaker asks the question, “Is Pat available to talk?” with falling rather than rising intonation, the difference in meaning is far more subtle. We know from research that both intonations are perfectly acceptable and may be equally common for yes-or-no questions (Thompson, Reference Thompson1995), so any interpretive issues are likely to be pragmatic in nature rather than related to basic speech decoding.

Intonation is thus the first suprasegmental feature that is unlikely to cause difficulties with word intelligibility. Although we do not hear differences in intonation as right or wrong as with mispronounced words or segmentals, this does not mean that intonation is unimportant since it is more likely to affect intelligibility in terms of the message or the intent. In one well-known example, Gumperz (Reference Gumperz1982) described a misunderstanding that occurred due to misperceptions of intonational meaning. British Airway employees (primarily native speakers of English) who ate at a company cafeteria thought that the servers (largely women of Indian and Pakistani descent) were rude and sullen. This perception was so strong that it caused a barely simmering revolt on both sides: The airline staff felt they were being treated badly by those they perceived as being beneath them in status, and the cafeteria workers felt those they were serving were rude and angry. Gumperz discovered that the problem had its roots in the use of intonation. When serving, the Pakistani and Indian women would offer things to the employees with a falling intonation (e.g., Gravy↘), while the employees, having English-language expectations, expected rising intonation (e.g., Gravy↗). In the Indian and Pakistani women’s L1s, the use of falling intonation was polite and appropriate, while for the native English-speaking employees it was heard as rude. In other words, the English-speaking employees did not hear the intonation as linguistically wrong, like a phoneme error; rather, they considered it to be a social failing.

In another study, Hahn (Reference Hahn2004) found that misplacement of sentence focus caused listeners difficulties in recalling information from a lecture. The study examined three conditions in which a Korean English bilingual presented the same information: correct focus placement, incorrect focus placement, and no focus placement at all. All three conditions included a text that was, technically, perfectly intelligible. However, when unexpected words were given special emphasis, or when no emphasis at all was used, listeners did not recall the information with the same degree of success. In addition, the speaker was heard as more or less likable, depending on the emphasis pattern.

Another suprasegmental feature is phrasing, or juncture, in which speakers’ pauses have more than one effect on how well listeners interpret speech. In Example (1.1), juncture plays an obvious role in differentiating between the meanings of the two sentences. There is no necessary difference in intelligibility, because both sentences use the same words. The entire difference in meaning is one of intelligibility, although phrasing differences are rarely so clear as in this example.

A third suprasegmental feature is rhythm. Although closely related to word stress in its ability to affect the pronunciation of vowels and consonants, phrase rhythm in English affects the overall structure of the discourse. Function words, such as single-syllable prepositions (e.g., in, at, on, with, by), determiners (e.g., the, a, some, an, his), and auxiliary verbs (e.g., have, can, could, would, is, did) are almost always unstressed in discourse, and thus can be difficult to process for learners of English. On the other hand, the same learners’ tendency to pronounce function words as full forms rather than reduced (e.g., can as [k^hæn] rather than [kən]) sometimes causes difficulties for native North American English listeners, who associate the full vowel pronunciation with the negative, can’t. Additionally, spoken language with too many stresses can overload the circuits of native listeners, causing them to pay attention to too much informational content, thus making it hard for them to select key words from the stream of speech. Interestingly, Jenkins (Reference Jenkins2000) argued that pronouncing all segments with full forms is unlikely to lead to loss of understanding in ELF communication, in which NNSs speak primarily to other NNSs. This remains an open question for research in ELF.

Rhythm also has the potential to affect intelligibility, especially for nonnative listeners. Rhythm in English leads to pronunciation adjustments both within and across word boundaries, especially in informal casual speech. The word polite is likely to lose its initial unstressed syllable, being pronounced p’lite, which in discourse can be misunderstood as plight. Native listeners are likely to pay attention to an additional cue, the length difference for the stressed vowel, but there is no guarantee that nonnative speakers will pay attention to or produce this cue consistently. Palatalization across word boundaries along with reductions can mask the identity of words completely, so that sentences like Did you eat yet? may sound like Didja eat yet? or even Jeechet?

Other Features

The flow of speech (fluency) appears to affect comprehensibility more than intelligibility. Fluency is a bundle of different features, and it is notoriously difficult to define precisely (Riggenbach, Reference Riggenbach2000). It includes, at the very least, speech rate, phrasing, grammatical grouping, and final lengthening, all mixed up in a soup pot with just the right amount of timing. Even NS speech may be judged as not fluent, since native speakers may have excessive and badly placed filled pauses (e.g., um) or pathological conditions such as stuttering. For listeners who are native speakers of English, disfluent speech interrupts their processing of the message. Likewise, overly fast speech, without appropriate pauses and emphasis, can also make a speaker’s message hard to understand. For listeners who are nonnative speakers of English, excessively fluent speech may be hard to process simply because of its speed. Slower speech, and speech that is less fluent, may be easier to process. From the point of view of speaking, fluency in nonnative speech is likely to develop along with proficiency and practice. A focus on language chunks may be helpful in developing fluency, but fluency is primarily a result of greater speech automaticity, in which all the elements of language do not have to be thought about equally during communication (Gatbonton & Segalowitz, Reference Gatbonton and Segalowitz1988, Reference Gatbonton and Segalowitz2005).

Speech rate is closely related to fluency. Although some L2 speakers talk at an excessively fast speech rate (Munro & Derwing, Reference Munro and Derwing2001), many more speak with rates that are too slow for native listeners. While it may appear to some North American English listeners that some second-language speakers speak too quickly (such as Indian speakers of English), the perception of a fast speech rate may also come from an unfamiliar rhythmic pattern in which more syllables are strongly stressed than expected.

Another global feature of pronunciation, voice quality (Esling & Wong, Reference Esling and Wong1983), may play a role in judgments of comprehensibility, either by facilitating understanding (if the voice quality is familiar and comfortable for a listener) or by exaggerating the effect of other speech differences. For example, Indian English is often spoken with a higher voice pitch than is American English. This difference in voice quality is part of a typical Indian English accent, and it is noticeable enough to add to a listener’s processing load, especially if the listener is not familiar with or favorably disposed toward the accent.

Another feature that may play a role in judgments of comprehensibility is loudness (or its acoustic correlate, intensity). Loudness can play an obvious role if a speaker is excessively soft, or excessively loud, in that listeners will have to work harder to listen effectively. Loudness, however, has little role in linguistic categories. It is, perhaps, a component of stress in English (Zhang & Francis, Reference 294Zhang and Francis2010), but it is not as important as length or pitch. After all, one can identify stressed vowels whether they are whispered or bellowed. In one situation I am familiar with, a female Chinese teaching assistant (TA) was teaching American undergraduate students who had complained about her English, even though her language appeared to be highly comprehensible. However, she spoke very quietly. Upon being questioned, she said that she believed that, as a woman, she was supposed to speak quietly in public situations. This appeared to be a cultural difference (and perhaps an idiosyncratic one at that), but undeniably her lack of loudness harmed both her intelligibility and her comprehensibility. In another situation, a male TA from Sri Lanka was easy to understand when speaking one-on-one, but did not pass a test for TAs because he spoke so quietly. In this case, the quiet speech was not cultural but idiosyncratic.

Finally, another area of pronunciation that may affect comprehensibility has to do with energy or “liveliness.” Hincks, in a series of articles (e.g., Hincks & Edlund, Reference Hincks and Edlund2009), has advocated teaching spoken language through using computer-assisted feedback that measures “liveliness” of speech, a complex feature that appears to be most heavily connected to variations in pitch range throughout extended spoken discourse. Hincks has found that speakers who have livelier voices (with more variation in their pitch range) are heard as more effective. In contrast, speech that sounds more monotone, with flatter pitch over long stretches of speech, may require more work by listeners. The effect of pitch variation does not appear to be connected to linguistic categories of intonation (e.g., final intonation or focus placement), but is rather an effect of how animated voices are.

Error Gravity and Intelligibility

Language errors of any kind are likely to affect the ways that listeners understand a message. Even if the errors are not serious, they may be salient, and thus the errors become an extra part of the message being communicated. For example, children’s speech is noted for its tendency to overgeneralize patterns to exceptions. English-speaking children may say “they flied” rather than “they flew” or “two mans” rather than “two men.” Although this process may seem ubiquitous in children’s speech production, one study that examined the speech of 83 children found that only a median of 2.5 percent of potential errors were actually regularized for all the children studied (Marcus et al., Reference Marcus, Pinker, Ullman, Hollander, Rosen, Xu and Clahsen1992). When I have asked students in introductory linguistics classes how often children make these kinds of errors, and given them options of 5, 25, 45, and 65 percent, they invariably choose 45 or 65 percent, and are shocked when they hear that 5 percent is closer to the actual occurrence. This finding regarding the perception of the frequency of L1 errors suggests a principle. When we consider our ability to understand a speaker with a foreign accent, we are likely to overstate the role of accent and to single-out errors that we know how to name, leading us to attribute to those errors an oversized role in intelligibility. Segmental errors, especially obvious ones like substitutions for /θ/, are noticeable and we can name them. While linguists know that such errors, even when they occur more than once in a sentence, rarely affect the most basic level of understanding – that is, intelligibility (Munro & Derwing, Reference Munro and Derwing2006) – the sounds continue to be taught first in many pronunciation classes. This is the equivalent of putting a bandage on a scratch when the accident victim is having trouble breathing.

Research into errors in written and spoken language has shown over and over again that certain errors are more serious than others. In an early study for English NSs, Hairston (Reference Hairston1981) asked readers in various occupations to judge the seriousness of errors in grammar and usage that are typically addressed in college writing classes and writing handbooks. A surprising number of errors were not considered serious, while those that were considered most serious included those Hairston called status markers (i.e., glaring examples of nonstandard usage, such as double negatives) and errors that caused sentences to seem incompletely formed, such as sentence fragments. In addition, Hairston found differences between the judgments of male and female respondents. Females were stricter in their judgments.

Other studies have looked at judgments of spoken nonnative errors. Fayer and Krasinski (Reference Fayer and Krasinski1987) asked English- and Spanish-speaking judges to evaluate the seriousness of errors from Puerto Rican speakers’ speech. Spanish judges were stricter and more irritated than were native judges, suggesting that the intelligibility of spoken language is also dependent on the listeners and their connection to the language of the speakers. In another study, Derwing, Rossiter, and Ehrensberger-Dow (Reference Derwing, Rossiter and Ehrensberger-Dow2002) found that NSs were better than nonnative speakers at identifying errors in aural than in written form, but that nonnative judges were stricter about the gravity of errors and were more annoyed by the errors.

Other Factors Affecting Judgments of Spoken Language

Judgments of speech may also be negatively influenced by factors unrelated to understanding. Some errors are more irritating than others, even though the message may be perfectly understood. For example, Hairston’s (Reference Hairston1981) status markers, her most negatively rated category, included errors such as “He don’t” and “We was,” both of which are well-known deviations from standard English found in nonstandard dialects. Lippi-Green (Reference Lippi-Green1997) described errors of native and nonnative speakers that are stigmatized even though the speech is perfectly understood. If speakers come from a social group that listeners consider less desirable, then the way they speak may also be considered less desirable.

There is also some evidence that accented speech may be more intelligible if the listeners become familiar with the features of the speech. Derwing, Rossiter, and Munro (Reference Derwing, Rossiter and Munro2002) provided training for social work students who needed to interact with Vietnamese learners of English. The students were given training in the typical features of Vietnamese accents before interacting with Vietnamese immigrants to Canada. The training was minimal (20-minute sessions integrated with other classroom work over eight weeks). Their confidence increased and they reported being more successful in their interactions than those who did not do the training.

Firm answers about the differential effects of errors on understanding have been elusive. While we believe it is possible to identify which errors make a difference and which ones do not, even errors that are not usually harmful to intelligibility may sometimes be more serious, and errors that might be expected to cause difficulties may not. For example, errors in the pronunciation of /θ/ have been almost uniformly considered as not having a serious effect on intelligibility. Jenkins (Reference Jenkins2000) went so far as to include all consonants as important for international intelligibility except dark [ɫ] (as in all) and the two “th” sounds. The case of the “th” sounds is particularly interesting in making recommendations about how much effect errors have on communication. Jenkins (Reference Jenkins2000) and Jenner (Reference Jenner1989) both say that these sounds are relatively unimportant. Brown (Reference Brown1988) found that errors involving “th” typically have a low functional load. However, Deterding (Reference Deterding2005) found that the actual pronunciations of /θ/ as /f/ by Estuary English speakers was very harmful to intelligibility for Singaporean English listeners who listened to British English speakers. The lack of intelligibility was especially noticeable in the words three and free.

Similarly, word-stress errors can easily mask the identity of a word and cause serious difficulties in comprehension, but Cutler, Dahan, and van Donselaar (Reference Cutler, Dahan and van Donselaar1997) found that NSs were rarely troubled by mis-stressing of words for which the stress was a key clue to lexical category. Word pairs like ˈpermit/perˈmit and ˈinsult/inˈsult differ only in stress, yet native listeners do not seem to rely on stress to interpret the words (Cutler, Reference Cutler1986).

These findings suggest that the gravity of errors in a language not only differ between categories, but that they also differ within categories. Thus, it is unlikely that suprasegmental errors are uniformly more serious than segmental errors or that consonant errors are uniformly more serious than vowel errors. Rather, it is likely that each subset of errors has members that are sometimes more serious and sometimes less so. While Gumperz’s (Reference Gumperz1982) examination of Indian and Pakistani women’s intonation shows that unexpected intonation can sometimes cause serious problems in intelligibility (as well as the consequent irritation), it is likely that the use of single-word utterances (Gravy↗ rather than Would you like gravy↘) also played a role in the depth of the misunderstanding. The same unexpected intonation in a syntactically complete question with the politeness marker “would” may be less likely to be considered rude than would the single-word utterance heard by the English NSs.

We need a nuanced approach to intelligibility, an approach that examines in detail what we know of intelligibility, an approach that specifies research priorities that will help tease apart those features that are important for those helping L2 students become more successful in communicating. While it is unlikely that we will ever know perfectly which errors always cause the greatest difficulty, we can know which are likely to do so. But we must examine pronunciation and other errors in greater detail before our language classes can truly reflect an approach that prioritizes intelligibility.

What Is Involved in Pronunciation?

An intelligibility-based approach to teaching pronunciation requires a clear description of the key elements of pronunciation and how they relate to one another. This description is the goal of the following section.

When we use the term pronunciation, we are talking about an interrelated system of sounds and prosody that communicates meaning through categorical contrasts (e.g., phonemes), systematic variations (e.g., allophones), and individual variations that may mark gradient differences such as gender, age, origin, etc. As such, the system of pronunciation can be divided in a variety of ways, but the divisions are merely a way to understand pieces of the system, even though all parts of the system interact with each other in ways that often make it impossible to separate their effects upon understanding.

The classification of the features that I use here is somewhat unusual (Figure 2.1). Typically, pronunciation is divided into segmentals and suprasegmentals, but from the viewpoint of spoken language understanding, word-level features are those most likely to impact intelligibility at the lexical level, and discourse-level features are likely to affect intelligibility at the semantic and pragmatic levels, and they are also more likely to impact comprehensibility. There are also spoken language elements related to pronunciation that are nonetheless not included in my categories for pronunciation. Any of the categories (word-level, discourse-level, and related areas) may impact perceptions of speech. In addition, although Figure 2.1 divides features into separate categories, this is a failure of the visual in representing the ways in which the categories interact. For example, word stress and rhythm both affect the pronunciation of vowels, with unstressed syllables strongly leaning toward schwa. Schwa is clearly a frequent segmental, but it is not a phoneme of English, but rather a variant of other vowels when they occur in unstressed syllables (Ladefoged, Reference Ladefoged1980). Stressed syllables in any part of a word are also the location for aspirated voiceless stop consonants, as in pill, repeat, till, return, kill, and acute, creating a dependency of some consonant allophones and stress. The rhythm of the phrase level and the stress patterns of words affect the ways sounds change in connected speech (e.g., the phoneme /t/ as realized in city, nature, button, and “can’t you’” are not the phonetic sound [t] in North American English). Pronunciation of the individual segments, in other words, is dependent upon where they occur in words, which are in turn affected by where they occur within a phrase. Prominence occurs on syllables that are emphasized through a combination of pitch, syllable duration (a rhythmic feature), and loudness. Prominence typically (but not always) occurs on the stressed syllable of the prominent word, connecting prominence to word stress. Prominence results in segments that are pronounced with particular clarity and precision. The prominent syllable in a phrase is often marked by pitch, and is the beginning of the final pitch contour in the phrase – that is, its intonation.

Figure 2.1 Pronunciation features related to intelligibility

Word-Level Features

Word-level features include segmentals (vowels and consonants) and word stress (Figure 2.2). They also include consonant clusters, a type of segmental which may, when mispronounced, have an effect on syllable structure. This introduction is provided to define how segmentals and word stress are related to intelligibility and comprehensibility.

Figure 2.2 Word-based features important to intelligibility

Segmentals include approximately forty phonemes for most varieties of English (around twenty-four consonant phonemes and fourteen or more vowel phonemes). The phonemes mask the number of sounds that English uses, since phonemes often have multiple well-known allophones that are important for pronunciation teaching. For example, English regularly employs a glottal stop ([ʔ]) before vowel initial words (e.g., I → [ʔaɪ]) or as an allophone for /t/ before a final nasal (e.g., button –[bʌʔn̩]). Other well-known allophones include aspirated voiceless stops and affricates in pill/till/kill/chill, the dark (velarized) /l/ in all (as opposed to the light /l/ in word initial position, e.g., lap), the flapped /t/ in city, the now increasingly rare voiceless labial-velar fricative [ʍ] in which (sounds like [hw]), and many others. For vowels, allophones are almost too numerous to count, as vowel quality often shifts noticeably in the presence of nasals, before /ɹ/ or dark /l/, before /g/, and in unstressed syllables.

Regarding word stress, English is a free-stress language in that stress is not fixed to a particular syllable. Stress can occur on first, second, or third syllables, but the placement is fixed for individual words. For example, the main stress for a word may fall on the first syllable (COMfort, BEAUtiful), the second (caNOE, rePULsive), the third (referENdum, questionnAIRE), etc. Stressed syllables typically have greater segmental clarity, greater syllable duration, and greater intensity than unstressed syllables. When they are in particular discourse contexts, they may also be marked with pitch movement (Ladd & Cutler, Reference Ladd, Cutler, Cutler and Ladd1983)

Both segmentals and word stress are likely to impact intelligibility at the lexical level in that mispronunciations may lead listeners to fail to decode the intended words. This failure may come from identifying other possible words (as happens with minimal pairs) or failing to identify any word that matches the speech signal (as with segments that are distorted). Both consonants and vowels are affected by the stress patterns of words. Mis-stressed words may especially affect the ways that vowels are pronounced in English because of the ubiquity of the unstressed vowel schwa (33 percent of all vowels according to Woods, Reference Woods2005). Schwa is a key perceptual clue to lack of stress in English, and listeners tend to classify full vowels, even in unstressed syllables, as stressed (Fear, Cutler, & Butterfield, Reference Fear, Cutler and Butterfield1995). Because of these interactions, word stress and segmentals are inseparable in their impact on intelligibility (Zielinski, Reference Zielinski2008). An unexpected stress pattern on a word (e.g., FORtune → forTUNE) also may affect the ways that vowels and consonants are produced (e.g., [ˈfɔɹtʃən] versus [fɚˈtʰun]).

Discourse-Level Features

Discourse-level features (Figure 2.3) include suprasegmental features that carry categorical (i.e., phonological) meaning differences. In relation to how listeners may (mis)understand speakers, these suprasegmentals are not likely to cause listeners to misunderstand individual words (making words unintelligible) but are likely to cause listeners to process meaning with greater difficulty (making speech more effortful to understand). Another suprasegmental, word stress, as discussed above, is included as a word-level feature because it is more likely to impact intelligibility (though it may also impact comprehensibility, or ability to process speech, even without a change in vowel quality, as in Slowiaczek [Reference 290Slowiaczek1990]).

Figure 2.3 Discourse-based pronunciation features important to intelligibility

Rhythm, the first discourse-level feature, involves at the very least the relative durations of syllables and the timing of syllabic beats. The constructed sentence in (2.1), made up of all single-syllable words, varies between longer, stressed syllables (in CAPS) and shorter, unstressed syllables (in lower case). The stressed words have a pronunciation that will be closer to the citation form, while the unstressed words are prone to simplification (e.g., has is likely to have the [h] deleted or even to be contracted with John).

In pronunciation teaching, rhythm has often been described in terms of stress timing and syllable timing (e.g., Pike, Reference Pike1945). Stress-timed languages (like English) are asserted to have large durational differences between stressed and unstressed syllables, with relatively equal timing between stresses. Syllable-timed languages are considered to have quite similar durations between syllables and thus timing that is at the level of the syllable rather than at the level of stressed syllables. This well-known formulation is overly simplistic, however, and stress timing and syllable timing are tendencies rather than absolutes (Dauer, Reference Dauer1983). The rhythmic characteristics of many languages remain of interest to researchers, and a wide variety of rhythm metrics have been tested for different L1 speakers (e.g., Low, Grabe, & Nolan, Reference Low, Grabe and Nolan2000), L2 speakers (Yang & Chu, Reference Yang and Chu2016), and L1–L2 comparisons (White & Mattys, Reference White and Mattys2007). However, there is great uncertainty about how well various rhythm metrics actually capture perceived rhythmic differences between languages (Arvaniti, Reference Arvaniti2012).

In English, rhythm and word stress are similar. The discourse level for rhythm in many ways mirrors the word-level rhythm of lexical stress. A major difference is that word stress typically is limited to multi-syllabic words, whereas rhythm includes stress for single-syllable words. In English, for example, content words (e.g., nouns, verbs, adjectives, adverbs, negatives), including those of one syllable, are normally stressed in discourse. Single-syllable function words (e.g., prepositions, auxiliary verbs, pronouns, determiners) are typically unstressed in discourse. Many single-syllable function words are also among the most frequent words in English, helping to contribute to the perception of stress timing.

Intonation, the second suprasegmental, includes at least three distinct ways in which meaning is communicated: prominence, tune, and range. The example in (2.2) illustrates these three. For context, imagine that the sentence is spoken in the middle of a lecture.

(2.2)

The initial extra-high pitch range in the example is meant to signal a topic shift, or what has been called a paratone (paragraph tone, see Wichmann, Reference Wichmann2000). Pitch range may also signal gradient meanings related to emotional engagement. NEXT is a prominent syllable with a jump up in pitch to call attention to the importance of the information (in this case, NEXT is likely related to PREVIOUS, the topic(s) that came before.) The last use of pitch is the drop in pitch from NEXT to the end of the utterance. This is the tune. Each of these uses is part of how intonation works in English.

Intonation is the system that uses voice pitch changes to communicate meaning. However, intonation may include more than voice pitch. Prominent syllables are not only higher or lower in pitch than the syllables that precede them, they also have greater duration (a rhythmic feature) and more clearly enunciated segmentals. In addition, the varied intonational categories are closely related. The final prominent syllable in a phrase (the nucleus) is also the beginning of the tune, and both may be pronounced with greater or lesser pitch range.

In regard to comprehensibility, the specific contribution of these suprasegmental features is understudied and needs greater attention. Isaacs and Trofimovich (Reference Isaacs and Trofimovich2012) found that more native-like vowel reduction and pitch contours correlated with better comprehensibility ratings, while pitch range was not significantly related to comprehensibility. Kang (Reference Kang2010) found that pitch range was instead associated with accentedness, but not comprehensibility. Tune, on the other hand, has also been suggested to have an effect on comprehensibility. Pickering (Reference Pickering2001), for example, found that Koreans teaching in English used a greater number of falling tunes than would be expected, and that the relative numbers of rising and falling tunes made their speech more challenging for listeners. (This is an interpretation of Pickering, given that she worked within a model that considers tunes to communicate differences in information structure.)

Related Areas

Comprehensibility and intelligibility are not only associated with pronunciation, but also with other characteristics of spoken language (Figure 2.4) that have an indirect connection to pronunciation. These areas include fluency (Derwing, Munro, & Thomson, Reference Derwing, Munro and Thomson2007), speech rate (Kang, Reference Kang2010), loudness (not typically addressed for L2 pronunciation research), and voice quality (Esling & Wong, Reference Esling and Wong1983; Ladd, Silverman, Tolkmitt, Bergmann, & Scherer, Reference Ladd, Silverman, Tolkmitt, Bergmann and Scherer1985; Munro, Derwing, & Burgess, Reference Munro, Derwing and Burgess2010). Generally, this book will not address these characteristics in detail, because other than fluency and speech rate (which is a component of fluency), these things are more idiosyncratic than the other features.

Figure 2.4 Spoken language features sometimes associated with pronunciation

In particular, research on fluency and speech rate seems to have significant effects on judgments of comprehensibility and accentedness. Pronunciation research on voice quality and its inclusion in teaching materials has never been common, despite its seeming promise in pedagogy (Jones & Evans, Reference Jones and Evans1995; Pennington, Reference Pennington1989). Loudness is an especially important issue in regard to hearing loss, hearing in noise, and the intelligibility of speech for those with cochlear implants. Anecdotally, some L2 learners can become more intelligible simply by speaking at a volume more appropriate to the context (e.g., in a large classroom), but this has not been typically considered important for an L2 pronunciation syllabus.

Fluency is sometimes associated with general proficiency (Fillmore, Reference Fillmore, Fillmore, Kempler and Wang1979) or smoothness of speech (Lennon, Reference Lennon1990). I use fluency to refer to smoothness of speech. However, judgments of fluency are also closely tied to many discrete elements of speech, including the numbers of silent pauses and filled pauses, whether junctures in spoken phrases are grammatically logical, mean length of run, repetitions in speech, and more (Lennon, Reference Lennon1990), as well as the level of automaticity in speaking (Segalowitz, Reference Segalowitz and Riggenbach2000), phonological memory (O’Brien, Segalowitz, Freed, & Collentine, Reference O’Brien, Segalowitz, Freed and Collentine2007), and attention control (Segalowitz, Reference Segalowitz2007). Remedial attention to pronunciation is more likely to be successful when learners are relatively comfortable speaking and listening in the L2 – that is, when they are sufficiently fluent. The development of fluency is clearly an important part of the big picture of L2 speaking development and instruction (Firth, Reference Firth, Avery and Ehrlich1992), and comfortable fluency can help give a global structure to other elements of spoken language, but fluency is not, by itself, part of L2 pronunciation. The fact that it impacts comprehensibility and is critical in communicative language teaching (Rossiter, Derwing, Manimtim, & Thomson, Reference Rossiter, Derwing, Manimtim and Thomson2010) means that, like pronunciation, it should be prioritized in teaching speaking and listening.

Speech rate is predictive of fluency judgments (Cucchiarini, Strik, & Boves, Reference Cucchiarini, Strik and Boves2000; Kormos & Dénes, Reference Kormos and Dénes2004). Fluent speakers can speak at different rates, so a fluent speaker could speak at a slower rate than a speaker who is judged less fluent. Speech rate is typically measured in syllables per second or words per minute. This rate may include all silences and filled pauses, or they may be removed, providing a measure of articulation rate. L2 speakers tend to speak more slowly than L1 natives, and their comprehensibility may be helped by faster speech. However, excessively fast or slow speech is more likely to be rated as less comprehensible and more accented (Derwing & Munro, Reference 273Derwing and Munro2001; Munro & Derwing, Reference Munro and Derwing1998).

Prioritizing: A Summary and Critique of Recommendations

Various writers have made recommendations about priorities for pronunciation teaching. These recommendations go from very general (learners should try for “listener-friendly pronunciation,” in the words of Olle Kjellin) to more detailed descriptions of what might be included in instruction. Kenworthy’s (Reference Kenworthy1987) early approach to intelligibility was based on describing learner pronunciation issues that may result in unintelligibility. She lists both segmental and suprasegmental issues, including substituting one sound for another, deleting or adding sounds, connecting one word to another, mis-stressing words, not using stress-based rhythm, and the misuse of intonation or use of unfamiliar intonation. Kenworthy does not further prioritize these potential sources of unintelligibility.

In another early attempt to prioritize, Jenner described a “common core for pronunciation” that would “guarantee intelligibility and acceptability anywhere in the world” by specifying “what all native speakers of all native varieties have in common which enables them to communicate effectively with native speakers of varieties other than their own” (Reference Jenner1989, p. 2). Rather than focusing on the differences between varieties, or indeed ruling out certain native varieties a priori while elevating others as models, Jenner thought it essential to look at commonalities, recognizing that speakers of native varieties have a better chance of being intelligible to each other because of what they share. Jenner suggested that the commonalities included vowel quantity (phonetic length differences), most consonants, syllable structure, stress-based rhythm, and varied commonalities of intonation, including tonic syllables and final movements of pitch. Jenner’s recommendations refer to segmental and suprasegmental features, and they often distinguish between a category (e.g., consonants) and secondary features that are not essential in the category (e.g., distinctions between [l] and [ɫ]).

In another analysis focused on segmental pronunciation, Brown (Reference Brown1988) proposed functional load as a way to determine pronunciation priorities. Functional load, a topic that had been put forth long before Brown’s application to pronunciation teaching, measures the “frequency with which two phonemes contrast in all environments” (Brown, Reference Brown1988, p. 591). Functional load is therefore inherently contrastive and the use of minimal pairs gives a way to quantify priorities. Phoneme contrasts that have a higher functional load are more likely to cause confusion if mispronounced, and they should be given priority over those with lower functional loads (assuming, of course, that students have difficulty with the contrast). Brown’s proposal for measuring functional load takes into account not only the number of minimal pairs for a contrast, but other issues such as the number of minimal pairs for the same part of speech, the extent to which a mispronunciation is stigmatized in native accents, and the acoustic similarity of the sounds involved in the contrast (not all minimal pairs are likely to be confused). For example, Brown’s analysis lists contrasts such as /p, b/, /p, f/, /l, ɹ/, and /l, n/ as of the highest importance, and contrasts such as /f, θ/, /ð, d/, and /ʤ, j/ as very low-priority contrasts. The proposal quantifies importance, but Brown recognizes that quantification alone cannot completely determine priorities. In addition, functional load is a measure of segmental importance only and cannot be applied to suprasegmental features, even to relatively uncomplicated ones like word stress.

Firth prioritizes differently by proposing a “zoom principle” for teaching in which “a pronunciation syllabus should begin with the widest possible focus and move gradually in on specific problems” (Reference Firth, Avery and Ehrlich1992, p. 173). By this, she means that pronunciation instruction should start with general speaking abilities and ability to communicate before moving on to phonetic details. This is put forth as the most likely way to promote comprehensible speech. Priorities beyond general speaking ability (including volume and clarity) include intonation and stress/rhythm (features that are more related to general speaking ability) before considering consonants and vowels. Interestingly, Firth’s recommendations led to some paradoxical suggestions that seem to have little to do with communicative importance. Despite its not being seen as critically important for understanding, /θ/ is given high priority within the consonantal system because students perceive it to be important and because it is relatively easy to teach effectively, giving students confidence to try for more difficult sounds. Practically speaking, this may also be a feature to spend modest time on if it leads to greater commitment to features more likely to improve intelligibility (Derwing & Munro, Reference Derwing and Munro2015).

Evidence for Firth’s recommendations comes from a recent study by Isaacs and Trofimovich (Reference Isaacs and Trofimovich2012). With a goal of making explicit the issues involved in comprehensibility ratings used in a variety of spoken assessment tools, the researchers examined which factors were salient in ratings of comprehensibility. Using criteria collected from research studies, the researchers identified nineteen quantitatively scored speech measures, including segmentals, suprasegmentals, fluency, vocabulary, grammar, and discourse. Speech samples of forty French learners of English were analyzed, and the scores derived from the analysis were then correlated with naive NS raters’ comprehensibility ratings (based on Munro & Derwing, Reference Munro and Derwing1995). Following this, three experienced ESL (English as a second language) teachers with minimal training in pronunciation listened to the speech samples and identified the factors they noticed when evaluating the speech of the French-speaking learners. Five factors were identified as important: fluency, breadth of vocabulary, grammatical control, construction of the discourse, and word-stress accuracy. Of these five measures, only one involves pronunciation in an explicit way, while the others are more global measures. The study concludes that expert raters do not focus overtly on pronunciation when evaluating speech comprehensibility, but that instead they take account of features that are not often included in pronunciation-oriented instruction. Other evidence for Firth’s contention can be seen from the work of Derwing, Munro, and Wiebe (Reference Derwing, Munro and Wiebe1998), who found that listeners rated L2 learners’ comprehensibility higher after a pronunciation course focusing on global skills and suprasegmentals than after a course focusing on segmental improvement. Similar results for other groups of learners can be found in the work of Gordon and Darcy (Reference Gordon and Darcy2016).

McNerney and Mendelsohn (Reference McNerney, Mendelsohn, Avery and Ehrlich1992) argue for suprasegmentals to be given top priority in pronunciation instruction. Suprasegmentals control how information is related, and they also are said to have a special role in conveying attitudinal meaning (cf. Pike, Reference Pike1945). McNerney and Mendelsohn say that “a short-term pronunciation course should focus first and foremost on suprasegmentals as they have the greatest impact on the comprehensibility of learners’ English” and because through suprasegmentals “greater change can be effected in a short time” (Reference McNerney, Mendelsohn, Avery and Ehrlich1992, p. 186), although they provide no empirical evidence for their confident assertion. Suprasegmental features include:

• word stress and rhythm;

• major sentence stress (or focus);

• intonation;

• linking and pausing;

• palatalization in rhythmically related contexts such as “can’t you” and “did you.”

In another attempt to specify priorities for a short pronunciation course, Henderson (Reference Henderson2008) identified pacing (stressed words per minute), speech rate (syllables per second), and word stress as important in promoting more comprehensible speech in spontaneous and prepared speaking tasks. The author argued that these three areas were most amenable to changes in the short term, and that learners of English in this university-level planned speaking course were most likely to be successful by learning to vary their pacing and speech rate. Word stress, which was asserted to be important in promoting understanding, was also presented as a feature that may not be easy to change in the short run. It may be significant that both Isaacs and Trofimovich (Reference Isaacs and Trofimovich2012) and Henderson (Reference Henderson2008) studied French learners of English, a group for whom word stress may be particularly important.

Gilbert (Reference Gilbert2001) suggests priorities for beginning learners based on her experience and repeated attempts to distill that experience into pronunciation features that are likely to be learnable and to make a difference in comprehensibility. She lists the following as essential: sound–symbol correspondences for the spelling of key vowel sounds, consonant sounds (mostly final) that serve as signals of grammatical meaning, linking between words, epenthesis and deletion of syllables, word stress, differences between weak and strong syllables (stress timing), and emphasis (prominence or nuclear stress). This set of priorities emphasizes suprasegmentals, but also says that certain segmental targets are both important enough to pay attention to and likely to be learnable even at the beginning levels of proficiency.

Not all writers who agree that a greater emphasis on suprasegmentals is important for pronunciation instruction prioritize what should be taught. Morley (Reference Morley1991), in her historical review of the evolution of pronunciation in TESOL (teaching English to speakers of other languages), described many of the changes from the traditional, segmental-based approach of the 1950s and 1960s to the more communicatively oriented approach becoming evident in the 1980s and 1990s. However, Morley’s recommendations called for “an expanded concept of what constitutes the domain of pronunciation” (Reference Morley1991, p. 493) rather than setting clear priorities. This expanded domain saw pronunciation’s proper sphere of influence as encompassing communication skills, suprasegmentals, segmentals, voice quality, body language, greater learner and teacher involvement in developing self-monitoring skills, contextualization, linking listening to speaking, greater attention to sound–spelling relationships, and attention to individual differences among ESL learners. These recommendations cannot be faulted in and of themselves, but taken together their approach to teaching pronunciation suggests that previous lack of success came not because of misplaced priorities or goals, but because the program of instruction was too limited to work. While Gilbert’s triage metaphor suggests that certain pronunciation needs are medically critical, identifiable, and should be dealt with immediately, Morley’s recommendations sound like an extended stay at a luxury pronunciation spa with personal accent trainers.

Prioritizing for Nonnative Speaker–Nonnative Speaker Communication

All these approaches assume that native listeners are the appropriate audience for determining what is intelligible and what is not. This assumption is called into question by Jenkins (Reference Jenkins2000) in her proposal for a prioritized set of features for pronunciation teaching, the Lingua Franca Core (LFC). Jenkins recognizes that most nonnative speakers (NNS) of English around the world interact in English not with NSs but with other NNSs. This difference in audience prompted Jenkins to consider changes in how priorities are determined, with mutual intelligibility being the standard by which pronunciation features are to be judged for importance. Jenkins studied the interactive interlanguage talk of NNS dyads doing communicative tasks. She analyzed where communication failed in these communicative tasks and determined the causes of the forty instances in which communication broke down. Of these, twenty-seven were related to pronunciation deviations. The deviations that caused loss of intelligibility were candidates for the LFC. Those features that did not cause a loss of intelligibility were typically excluded from the core. The core included most consonant sounds, some consonant cluster simplifications, vowel length, and nuclear stress (i.e., prominence). Consonants that were excluded were the interdental fricatives and [ɫ], the velarized, or dark, allophone of /l/. Consonant cluster deletions were included in the LFC because loss of sounds was argued to be more likely to impact intelligibility than was epenthesis. Perhaps the place in which the LFC departs from the other recommendations most radically is in its treatment of suprasegmentals. Only one, nuclear stress, is included in the core. Others, such as stress-based rhythm, intonation, and word stress are all excluded based on Jenkins’ data, her appeal to teachability/learnability, and the impact of universals. Some of these decisions have been criticized, especially regarding word stress (Dauer, Reference Dauer2005; McCrocklin, Reference McCrocklin, Levis and LeVelle2012). In one description of pronunciation teaching in China that used the LFC as a rubric, Deterding (Reference 274Deterding2010) showed what most experienced teachers know very well: Pronunciation difficulties are varied, and they include errors that reflect both core and non-core features.

In a replication of Jenkins’ (Reference Jenkins2000) work, Kennedy (Reference Kennedy2012) found that the most common sources of unintelligibility were vowel and consonant segments, either individually or in combination. The only suprasegmental feature implicated in unintelligibility was word stress. Nuclear stress (prominence) was not a source of unintelligibility. Kennedy also suggested that learners may not always indicate that they do not understand a speaker and that researchers and teachers may not realize that pronunciation is a factor. This problem may be connected to the types of interactive tasks that are used to collect data, such that both listeners and speakers have to demonstrate understanding.

Walker (Reference Walker2010) applies the LFC extensively to pronunciation teaching, giving the original recommendations a classroom teacher’s perspective. His defense of ELF (English as a lingua franca) priorities describes why he thinks the LFC is appropriate, including issues related to bottom-up rather than top-down processing, mutual intelligibility, speaker identity, and teachability. One of the benefits of the LFC, Walker argues, is that it recognizes that ELF speakers make greater use of bottom-up processing in their interactions. This means that they are much more reliant on the details of the acoustic signal in interactions, and thus more likely to be affected by unexpected pronunciations of segmentals. NSs, in contrast, make greater use of top-down processing, in that they can more easily guess at the content of a spoken message even when the segmentals deviate from what is expected. The LFC, however, remains controversial in many of its recommendations. It lacks robust empirical support, assumes that all NNS contexts are similar, and does not take into account the importance of stigma associated with otherwise intelligible pronunciations (LeVelle & Levis, Reference LeVelle, Levis, Levis and Moyer2014). The actual details of what should and should not be included thus have found uncertain acceptance despite its appeal, and influential researchers have criticized its consistency, calling for empirical verification of its recommendations (e.g., Szpyra-Kozłowska, Reference Szpyra-Kozłowska2015, pp. 77–84).

The second area discussed is mutual intelligibility. Intelligibility is a context-sensitive feature of spoken discourse, a characterization that will find little disagreement from almost anyone interested in teaching pronunciation. Walker is more nuanced than Jenkins, recognizing that the LFC may need to be adjusted in some ways because interlocutors may also include NSs. For example, the teaching of weak forms and vowel reduction may be strictly non-core as far as production, but mutual intelligibility suggests that it is a core feature for perception. NS interlocutors will reduce vowels, and it is important for ELF listeners to be able to understand such speech. For communication to happen, both speakers and listeners must be intelligible to each other, and learning materials typically make use of native speech.

The third positive aspect of an LFC approach is that it recognizes the importance of speaker identity. The LFC recognizes that achieving a native accent is not necessary, and that the influence of the speaker’s L1 should be accepted, as long as intelligibility is not compromised. This is clearly not just part of NNS–NNS communication. Most people in inner-circle communities, especially in larger cities with significant immigrant communities, also think that an NS accent is not needed.

Finally, Walker discusses Jenkins’ concept of teachability. Walker says that “many features that are essential in a traditional EFL syllabus are largely unteachable. This was the case with tone and stress-timing, and with the use of weak forms and certain connected speech changes. In contrast, most of the items in the LFC are teachable, with classroom teaching leading to learning” (Reference Walker2010, p. 63). While it makes little sense to teach things that our learners cannot learn, little evidence is provided regarding teachability. In fact, many of the features that are said to be unteachable are ones that Judy Gilbert (Reference Gilbert2001), who is a notorious stickler for teaching only those things that can be learned, provides as priorities for beginning learners: linking, word stress, and distinguishing strong and weak syllables.

What does it mean for a feature to be teachable? Walker (Reference Walker2010) says this about nuclear stress placement:

Teachability thus seems to mean a topic whose rules can be learned and applied by learners, leading to acquisition. It does not mean teachable. Any topic can be taught. What matters is the extent to which the teaching, the input, becomes learning, or intake. The principle of teachability/learnability will be discussed in detail in Principle 6 in Chapter 8.

Walker suggests ways the LFC might apply to speakers of different languages. Following Jenkins, he includes in the core rhotic /ɹ/ in all positions (characteristic of most North American English speakers), the non-flapped /t/ characteristic of British English, and word stress (an admitted gray area in Jenkins, Reference Jenkins2000) because of its impact on the core feature of nuclear stress, as well as vowel reduction and weak forms for receptive competence. In addition, certain errors (such as final glottal stops) that are common among certain users of English should be addressed because they may cause loss of intelligibility by masking the character of final stop consonants (see Walker, Reference Walker2010, p. 44; cf. Gilbert, Reference Gilbert2001).

The LFC’s recommendations have been used to examine features that promote the mutual intelligibility of emerging South-East Asian Englishes and the international intelligibility of Hong Kong English. Deterding and Kirkpatrick (Reference Deterding and Kirkpatrick2006) examined conversational interactions in English among speakers from ten South East Asian countries and identified features of speech that seem to form the basis of a developing regional variety. Features of this variety shared by speakers from at least four countries were the use of a stop for the voiceless dental fricative (dis for this), reduced initial aspiration of voiceless stops (pill sounds like bill), monophthongal mid-front and back vowels (take and goat do not have the extra glide typical of inner-circle varieties, so that they may sound like tech and gut), a lack of reduced vowels, stressed pronouns, and phrase-final discourse stress (e.g., Give it to HIM). Of these, several are features that Jenkins suggests should be treated as part of the LFC core (aspiration and nuclear stress) and others are part of her non-core features (dental fricatives and reduced vowels). Others are less obvious and may take finer analysis to determine whether they should be seen as core or non-core. Monophthongal vowels, however, may violate the quantity criterion for vowels while keeping the quality intact.

In the work of Kirkpatrick, Deterding, and Wong (Reference Kirkpatrick, Deterding and Wong2008), the intelligibility of the Hong Kong English of highly educated students was rated by university students in Singapore (a transitioning outer-circle country where English has an official role but is not the native language of all) and Australia (an inner-circle country in which English is the native language of most people). The students in Australia were both NSs and NNSs of English. Recordings of their speech were played for the subjects, who did a listening comprehension task about the content of the speech. In addition to this measure of intelligibility, raters considered the speakers in terms of intelligence and likeability, two concepts well-attested in other studies to be associated with speech, and in our terms, with the potential for irritation. In an interesting finding, the speakers who were the most intelligible were also seen as less intelligent and less likeable, often based on things they said, but sometimes on the basis of the speech being too good, suggesting that the raters thought the speaker was showy and proud. Clearly, intelligibility, a good thing in itself, may sometimes be judged negatively in some contexts based on unforeseen social values. Overall, Hong Kong English was widely intelligible in this area of the world, where it is likely to be a familiar variety of English. However, not all speakers were equally intelligible.

These different attempts to specify priorities for intelligibility-based instruction are interesting both in what they agree on and also in what they do not agree on. The variety found in the recommendations comes primarily from a heavy reliance on reasoning and a paucity of empirical evidence. Table 2.1 provides a summary of the recommendations.

Critiquing the Recommendations

The different attempts to specify priorities are a mishmash of incomplete and contradictory recommendations. Some of the studies offer recommendations based on experience (Firth, Reference Firth, Avery and Ehrlich1992; Gilbert, Reference Gilbert2001; Kenworthy, Reference Kenworthy1987; McNerney & Mendelsohn, Reference McNerney, Mendelsohn, Avery and Ehrlich1992), others provide priorities based on analysis of similarities and differences between English varieties (Jenner, Reference Jenner1989) or careful experimental evidence (Isaacs & Trofimovich, Reference Isaacs and Trofimovich2012) or other objective analyses based on models of intelligibility (Brown, Reference Brown1991; Jenkins, Reference Jenkins2000); some study intelligibility without any intention of recommending pronunciation priorities (Deterding & Kirkpatrick, Reference Deterding and Kirkpatrick2006; Kirkpatrick et al., Reference Kirkpatrick, Deterding and Wong2008), and others describe not only what should be included but also what should be excluded (Brown, Reference Brown1991; Jenkins, Reference Jenkins2000; Jenner, Reference Jenner1989). Features such as word stress are seen to be essential in some research (Isaacs & Trofimovich, Reference Isaacs and Trofimovich2012), while they are seen as relatively unimportant in other recommendations (Jenkins, Reference Jenkins2000) or potentially important in relation to other features (Walker, Reference Walker2010). A focus on suprasegmentals is encouraged by some authors (McNerney & Mendelsohn, Reference McNerney, Mendelsohn, Avery and Ehrlich1992), while it is largely bypassed in favor of segmentals in other accounts (Jenkins, Reference Jenkins2000). Some writers seek to achieve quicker rates of improvement in intelligibility by focusing first and foremost on features that are not usually part of pronunciation instruction (Firth, Reference Firth, Avery and Ehrlich1992), while other recommendations read like a pronunciation wish-list with no attempts to prioritize (Morley, Reference Morley1991). Sounds such as /θ/ are left off many lists, including Jenkins’ influential LFC, but other writers long to keep /θ/ because its supposed teachability may make it easier for learners to feel success and thus try harder sounds (Firth, Reference Firth, Avery and Ehrlich1992) or because there are situations in which /θ/ can affect intelligibility (Deterding, Reference Deterding2005; Henderson, Reference Henderson2008). Some of these seemingly contradictory recommendations are likely due to different L1 learner groups, limited numbers in each study, and the context in which the study took place.

It is clear that there is a further need to examine principles that help teachers decide on priorities based on context, allowing a finer-grained analysis than any single study can provide. All teachers have to prioritize, and it is best to have explicit, research-based support for setting priorities (Derwing & Munro, Reference Derwing and Munro2005). Much that has been written about priorities cannot be called research-based, and the articles that are based on empirical data should therefore have greater weight. Jenkins (Reference Jenkins2000), for example, has been much discussed, much praised, and much derided, but her recommendations are valuable because they are based upon evidence. However, the amount of evidence is small and some of the recommendations have been called into question (Dauer, Reference Dauer2005; McCrocklin, Reference McCrocklin, Levis and LeVelle2012). Her twenty-seven pronunciation errors (of forty total errors impacting intelligibility) allow us to make suggestive recommendations about what should and should not be included in the core features. /θ/, for example, did not lead to loss of intelligibility in her data. Clearly, there must have been many other phonemes that likewise did not lead to loss of intelligibility, but Jenkins made recommendations against /θ/ as a core item, and for other sounds not only based on her evidence but also on her view of English’s role in the world. This indicates that decisions about priorities must be made not only on explicit evidence, but on how implicit evidence is interpreted regarding pronunciation’s role in communicative success or failure in particular communicative contexts.

Context and Intelligibility

Finally, intelligibility is sensitive to the context in which communication takes place. What this means is that the degree of accuracy that determines intelligibility changes from one situation to the next, depending on the type of language use required. Intelligibility can be seen as the lowest possible standard that a speaker has to meet in order to get by. Hinofotis and Bailey (Reference Hinofotis, Bailey, Cameron Fisher, Clarke and Schachter1981), in a now famous phrase, talk about “an intelligibility threshold” that speakers must meet in order to communicate effectively. The intelligibility threshold, rather than being an objective criterion, is actually a moving target that includes much more than pronunciation (Tyler, Reference Tyler1992). Much of what causes the target to move is the context in which speech takes place. The influence of context is understudied, but it is likely to be an important determinant of how intelligible a particular speech sample is.

It should be obvious that certain contexts of use have higher stakes for both the speaker and the listener. For example, if your job involves staffing a cash register in an area in which ethnic shops are the norm (e.g., Chinatown), your needs for understandable pronunciation in English may be relatively low. Most of the customers are likely to either be from the same speech community or outsiders who have decided to take the extra step to shop there rather than somewhere else. The people the clerk interacts with are either sympathetic or are unlikely to come back regularly (tourists). Contrast this context with an instructor in a university class. Interactions are regular and required (students have to come to the lab or breakout session), high-stakes (performance in the class is highly dependent on being able to understand material through the mediation of the instructor), and subject to significant cross-cultural conflict (misunderstanding may be likely to be seen as caused by the inability of the other to play the expected role; cf. Rubin, Reference Rubin1992). High-stakes contexts include education, health, and translation, all areas in which speakers and listeners have to negotiate language and culture barriers and where the cost for failure is very real.

In North American higher education, many basic classes in the natural sciences, mathematics, and engineering are taught by NNSs of English. While some of these teachers are regular faculty, many are graduate teaching assistants who help fund their education through teaching. The classes they teach include not only majors in the field, but also students from other fields who are required to take courses that they may not feel comfortable with or enjoy. Their overall proficiency in English is very high. These NNSs have, after all, been admitted to demanding graduate degree programs in a foreign-language setting, and their needs for language support are targeted to specific areas such as speaking or writing.

Having these students teach sets up a natural context in which undergraduates’ stress from learning that demands new content (e.g., organic chemistry) can interact with stress from the way the content is presented (which may be due to the graduate students being inexperienced teachers or due to cultural views of appropriate teacher and student behavior), mixed in with unfamiliar or hard-to-understand accents. Pedagogical effectiveness and unfamiliar or inadequate pronunciation or language skills may both be implicated in lack of achievement, but inadequate language skills are most likely to be blamed for ineffective teaching. Many international teaching assistants (ITAs) and faculty development programs recognize that working on teaching and presentation skills can lead to greater success (and presumably, a better perception of comprehensibility) even without extensive work on language skills.

Other studies also make clear that comprehensibility is not based only on pronunciation. Tyler (Reference Tyler1992) asked raters to listen to two presentations, one given by an ITA and one by an NTA (native teaching assistant). Both presentations were then transcribed and read aloud by an NS of English so that pronunciation would not be a factor in how the presentations were rated. Raters evaluated the ITA as being less effective and less easy to follow than the NTA. The researcher argued that the ITA’s use of unexpected, nonparallel discourse markers (e.g., “the first one” followed by “and then” and “after that”), not establishing clear synonyms or clearly linking pronominal forms to the original noun phrases, and overuse of coordination and underuse of subordination, caused a loss of understanding.

The use of discourse markers may also be involved in how easy it is to understand speech. Williams (Reference Williams1992) found that when discourse moves were explicitly marked, ITA presentations were rated as being more comprehensible. Tyler and Bro (Reference 292Tyler and Bro1992), however, found that ITAs overused simple additive connectors that were ambiguous in the connections between ideas in the discourse. Liao (Reference 283Liao2009) examined Chinese ITAs’ use of common English spoken discourse markers in interactions with an interviewer and found that the ITAs overused some (especially yeah) and underused others (e.g., well, I mean). Overall, their markers were more restricted in range than for NTAs and included innovations that were not likely to be understood easily by NS interlocutors. At the level of grammatical competence, lexico-grammatical features may hamper ITAs’ ability to communicate information clearly. Tyler, Jefferies, and Davies (Reference Tyler, Jefferies and Davies1988) found that NTAs often used strategies to focus listener attention on information to be foregrounded and backgrounded, but ITAs did not.

In addition, even speech that is completely intelligible may be heard as heavily accented or take more effort to understand (Munro & Derwing, Reference Munro and Derwing1995). Expectations and implicit stereotypes may affect how well listeners understand a speaker, despite the speaker being intelligible. Rubin (Reference Rubin1992) played a short lecture given by a female speaker of General American English to undergraduate students under two guises: a Caucasian guise and an Asian guise. In the study, some listeners heard the lecture while looking at a picture of a blond Caucasian woman, while other listeners heard the same lecture (spoken by the same voice) while looking at a picture of an equally attractive Asian woman. When asked to demonstrate their understanding of the lecture, the listeners who heard the lecture in the Asian guise understood significantly less well than those who heard the lecture in the Caucasian guise. Comprehension was measured via a cloze of the passage with every seventh word deleted. In addition, listeners completed a semantic differential instrument with scales measuring their attitudes, issues related to background, values, and appearance, as well as items related to accent, ethnicity, and teaching qualifications. While the Munro and Derwing (Reference Munro and Derwing1995) study demonstrated that listeners can decode speech with 100 percent intelligibility yet find it heavily accented, this study suggests that lack of understanding can be affected by seemingly unrelated nonlanguage factors, in this case the unconscious biases that listeners bring with them to interactions. More recent research suggests that this bias may also be connected to congruence between the visual and the aural. McGowan (Reference McGowan2015) used transcription accuracy in noise to examine whether listeners would be more accurate in transcribing Chinese-accented speech when presented with a Chinese face, a Caucasian face, or an unspecified silhouette. Listeners transcribed more successfully when presented with a congruent face (a Chinese face), a finding that was consistent despite differences in experience with listening to Chinese-accented English.

Conclusion

Many recommendations about priorities for pronunciation teaching are painted in broad brushstrokes that probably mask distinctions between important and unimportant features within the same category. For example, Isaacs and Trofimovich (Reference Isaacs and Trofimovich2012) found that pitch movement correlated with comprehensibility ratings at a relatively high rate, whereas pitch range showed no correlation. Yet both are considered to be part of intonation. Levis (Reference Levis1999a) suggests that final intonation may be important for certain grammatical structures (e.g., declaratives), while the same pitch movements are relatively unimportant for others (e.g., yes–no questions). Syllable structure modifications are part of Jenner’s (Reference Jenner1989) core and Gilbert’s (Reference Gilbert2001) recommendations, while Jenkins (Reference Jenkins2000) distinguishes between initial deletions (core) and some medial and final deletions and epenthesis (non-core).

It seems clear from the often conflicting, and sometimes contradictory, recommendations that the criteria we use to set priorities are often themselves unclear, based on (un)informed intuition, unsupported by research findings, perhaps because research findings themselves have many gaps. We are still trying to understand the picture in which too many elements are missing, like trying to understand the picture on a jigsaw puzzle with only half the pieces available. A first step toward understanding the bigger picture is to try to specify and justify guidelines that may help us describe what an intelligibility-based approach might look like in the classroom (as in Chapters 8 and 9). It may also be that we need a better picture of how pronunciation not only affects understanding in a vacuum (i.e., via the speech signal alone), but also how it affects understanding in social and communicative contexts.