Strength of British English Accents in Altered Listening Conditions

Abstract

This work is concerned with the processing or representational level at which accent forms learned early in life can change, and whether alteration to the speaker's auditory environment can elicit an original accent. In experiment one, recordings were made of an equal number of a) speakers living in the Home Counties of Britain (around the London conurbation), but who claimed to have retained the accent of the region they originally came from, b) speakers who stated they had lost their regional accent and acquired a Home Counties accent and c) native Home Counties speakers. They read two texts in a normal listening environment. Listeners rated the similarity in accent between each of these texts and all other texts. The results showed that in the normal listening conditions, speakers who had lost their accent were rated more similar to Home Counties English speakers than to those speakers from the same region who had retained their accent. In experiment two, recordings of the same speakers under frequency shifted and delayed auditory feedback, as well as the normal listening conditions used earlier, were rated in order to see whether the manipulations of listening environment elicited the speaker's original accent. Listeners rated similarity of accent in a sample of speech recorded under normal listening against a sample read by another speaker in one of the altered listening conditions. When listening condition was altered, speakers who had lost their original accent were rated as more similar to those who had retained their accent. It is concluded that accent differences can be elicited by altering listening environment because the speech systems of these speakers are more vulnerable than speakers who do not change their original accent.

The term accent referred to in this study applies to the pronunciation as opposed to the grammatical and lexical composition of utterances (Hughes, & Trudgill, 1996; Wells, 1982). In this sense, every utterance is pronounced with an accent, which may be judged to be “posh”, “standard”, “foreign”, etc. Thus, a regional accent may be discerned whether or not the choice of word forms is regional (Brennan, & Brennan, 1981a, 1981b). Accent is usually perceived and identified as such, when the pronunciation differs from what a person is used to. However, the term is commonly restricted to deviations from some defined standard. We are using it in this sense in as much as we refer to regional pronunciations as accented in contrast to a Home Counties (HC) pronunciation.

The way in which accents differ can be described at different linguistic levels. First, there are prosodic and segmental differences, the former referring to lexical or phrasal stressing or accentuation patterns and intonation, the latter referring to the realization of consonants and vowels¹. Second, there are phonological versus phonetic differences, the former involving differences in the sound systems such as missing or additional sound oppositions (such as the absence of a difference between “fair” and “fur” in Liverpool English), the latter involving the way the properties are realized within the same framework of oppositions (e.g., northern [α=⌉] vs. a more upper class [A±⌉] for /A⌉/). Third, there are differences of lexical incidence, where the same or a similar framework of oppositions exists, but the lexical items in which the equivalent sounds occur differ. (E.g. “ladder, brass, father” which have the /Θ/, /A⌉/, /A⌉/ phonemes, respectively in Home Counties British English but /Θ/, /Θ/, /A⌉/ phonemes, respectively in Northern English accents.)

This article investigates some aspects of how accent-change occurs. A speaker who was born, brought up and remains in a local language region almost always has the accent of the area of his or her own part of the country. When speakers with British regional accents relocate to different parts of the British Isles, they often acquire the accent of the adopted region. But it is common to observe that speakers who have left the region in which they grew up and acquired another accent as adults revert to their original accent when that accent is being spoken around them. How much this applies to accents lost in childhood, and at what age of accent abandonment the reversion no longer occurs has not, to our knowledge, been investigated (but see Bongaerts, Planken, & Schils, 1997; Bongaerts, Mennen, & Slik, 2000; Flege, 1999 for discussions of age effects in L2 pronunciation learning) This is one manifestation of the well established phenomenon of speakers accommodating to the communicative context, (Giles, & Smith, 1979; Beebe, & Giles, 1984), which has both influenced research methodology and defined research questions in sociolinguistics (Labov, 1986). Within the specifically British social context, changes in accent may arise out of a desire to communicate more effectively in the destination region. Regional accents of British English signal social distinctions, and although accents are becoming less closely linked with social strata, the avoidance of social stigma remains one factor behind accent change. In the British Isles, speakers who change their accent to increase their social standing frequently adopt a Home Counties (HC) English accent (the accent spoken in the region of England around to London), the form widely used in the spoken media.

There are other (involuntary) ways in which accent may change. In speakers who have changed their accent, situations may exist that cause them to revert (partially) to their original accent. Informal observation suggests that alterations to the sound of the speaker's voice can bring this about: Two such alterations are examined in the below studies (change to frequency content and to timing of speech). Telephones make such alterations to the voice (as well as change in its intensity). When a speaker uses a telephone, frequency and intensity change and these alterations may explain why an accent appears to sound stronger. Also, a speaker's original accent re-emerges in older speakers who have adopted the accent of a different part of their country. Again this may be due to frequency and intensity changes in auditory reception that occur as speakers get older (presbyacusis). The influence changing timing has, is also supported by a number of informal observations. Thus, untrained actors have problems maintaining an imitated accent in echoey or noisy auditoria. Cellular phones have appreciable delays, so the emergence of accent on this type of equipment may be a result of alteration to timing.

There has been one experimental study that examined whether altering the listening environment causes a speaker to revert to his or her original accent. Howell and Dworzynski (2001) investigated whether delayed auditory feedback (DAF) and frequency shifted feedback (FSF) elicited an accented form of English in speakers who had an original language other than English (German speakers speaking English). They showed that the German accent of the German speakers speaking English was more marked when speakers were subjected to these two forms of alteration to their listening environment. They explained their results in terms of the speaker having multiple response forms for a language (the initial one based on their native language, and one that is a closer approximation to that of the acquired language). According to Howell and Dworzynski, speakers may revert to the early form even after they have become proficient in the acquired language. A potential problem in this study is that even the speech of a fluent speaker who has not changed his or her original accent is affected when listening environment is altered. Judgement may be based on how alterations to listening environment affect these changes to speech output rather than specifically about accent changes brought about by the change in listening conditions. Howell and Dworzynski (2001) minimised this possibility by performing several analyses to verify their listeners were judging accent, not other effects that altered listening environment might have brought about.

The experiments reported in this article address two questions about accent change: 1) Is the adopted accent of speakers living in London, and who consider that they have lost their original accent, judged to be closer to HC English than the accent of speakers who claim to have retained their accent? 2) Does the accent of speakers who report having lost their accent reappear when listening conditions are altered? In addition to answering these questions, the results also have implications as to whether listeners are specifically able to judge accent change when the accent change is induced by changing the listening environment (Howell, & Dworzynski, 2001).

Experiment one

As said earlier, HC English is the reference accent for many speakers as it is widely used in the media. All the speakers involved in both the current experiments have lived in London for some time, making it probable that their adopted accent is HC English. Also, when the speakers consider they have lost their accent, they often regard themselves as using an accent approximating HC English. This experiment tests whether the speech of speakers who have lost their original accent are judged more similar to the HC English accent than the speech of those speakers who retain their accent.

The texts used for the experiment were constructed so as to provide sufficient potential differences between the accents being examined to make the measurement of a shift in accent strength viable. The texts exploit the potential differences between dialects discussed above, namely either differences in phonemic oppositions, in the lexical incidence of phonemes or in the phonetic realisation of phonemes. Clearly, the segmental features of an accent, i.e., the consonant and vowel realisations, are only part of what allows a regional or social accent to be identified. Rhythmic patterning (Low, Grabe, & Nolan, 2000), in particular intonational features (Grabe, 2002; Grabe, Post, Nolan, & Farrar, 2000), are the supra-segmental (prosodic) accompaniment to the speech-sound structure of any utterance. They may reflect the regional origin of the speaker in addition to accent carried on the segmental features of the vowels and consonants. Naturally, the prosody enters into the listeners' judgements of similarity or difference of accent, but, as argued above, it is an integral part of any realised utterance and is therefore a legitimate part of the overall impression of regional accent strength that is being investigated. Thus, though the subjects' regionally varying prosody is not controllable through the structure of these texts, this does not affect the validity of the inter-group comparisons under the different feedback conditions. The study by Howell and Dworzynski (2001), in which the foreign accent of speakers who spoke English as a second language was judged under different auditory feedback conditions, shows that segmental text definition is a satisfactory predictor of potential accent deviation.

The technique used to establish proximity to HC English accent is factorial analysis of variance, investigating listeners' ratings of accent similarity between speakers with HC accents and speakers with regional accents, as a function of regional accent group and loss/retention of that regional accent. Investigation of accent variability at the individual level is also performed, using the average linkage hierarchical clustering algorithm to illustrate the overall patterns among ratings of speaker accent similarity. This produces a tree structure (dendrogram) that reflects the proximity of different accents. If the prediction of the experiment is upheld, the speakers who have lost their original accent should lie closer to HC English speakers than those speakers who retain their accent, demonstrated by a strong main effect of accent loss/retention in the ANOVA, and illustrated by the clustering performance of individual speakers.

Method

The method of the experiment is described in two parts. The first concerns speaker selection, material used and recording details. The second gives details of the perceptual judgements made by listeners.

Speakers for recordings

Twenty speakers who live and work in London were used (four for each of five original accents, including HC English)². These were volunteers, who came from five regions with distinct accents (Wells, 1982): Northern Ireland (NI), lowland Scotland (SC), Liverpool (LI) and Newcastle (NC) as well as the Home Counties (HC). None of the speakers reported speech or hearing problems, nor had they received specialized instruction (e.g. for acting or teaching purposes) about their speech. Individual details about the speakers are given in Table 1. Accent region is indicated in column one (a subject identifier for the speaker in that accent group is also included). For each of the regional accents, two speakers reported that they had lost their original accent (coded as “-” in speaker identifier) and two reported that they had retained their original accent (coded as “+” in speaker identifier). This self-reported accent status is checked empirically in the analyses.

TABLE 1.

Details of the 20 speakers used in Experiments 1 and 2. The first column gives accent region (NI= Northern Ireland, SC=lowland Scotland, LI=Liverpool, NC= Newcastle, HC= Home counties English) and whether they had lost their accent (- signifies lost accent and + signifies retains accent, this is not applicable to HC speakers). Columns 2-6 indicate gender, age, time spent in London, time spent in another region (in years) and whether the speaker had attended a provincial university. When a speaker attended a provincial university, this was for three years. Time spent in region represents time in the accent region plus time at a provincial university (where appropriate).

ACCENT REGION	GENDER	AGE	TIME IN LONDON	TIME IN REGION	PROVINCIAL UNIVERSITY
NI-	M	23	9	11	Y
NI-	M	32	7	22	Y
NI+	M	24	6	15	Y
NI+	M	49	26	20	Y
SC-	M	24	8	13	Y
SC-	M	35	22	10	Y
SC+	M	24	6	18	N
SC+	M	21	7	14	N
LI-	F	28	8	20	N
LI-	M	56	25	18	Y
LI+	M	57	34	23	N
LI+	F	47	17	30	N
NC-	F	48	35	13	N
NC-	F	23	5	15	Y
NC+	F	25	7	18	N
NC+	M	30	14	16	N
HC	F	23	21	2	N
HC	M	34	31	0	Y
HC	M	39	36	0	Y
HC	M	52	48	1	Y

Columns two and three give gender (14 male, M, 6 female, F) and age (ranging from 21 to 57 years), respectively. Column four indicates time spent in London (in years). In all cases, number of years in London was contiguous and in the period immediately prior to when the recordings were made. Column five gives time spent in another region of the United Kingdom (in years). With one proviso, all the speakers lived exclusively in their original accent region and then in London so the “other region” in column five is the accent designation area for speakers with regional accents. The exception is where periods were spent away from London for purposes of study at provincial universities (i.e., universities outside London). These periods away were always taken between the ages 18 and 22 and the region speakers went to, varied outside those selected for the experiment. Column six indicates whether time was spent at a provincial university outside the designated accent region. So, for instance, this shows that three of the four HC English speakers have lived all their lives in London except for three-year periods at provincial universities. Most speakers with regional accents initially came to London to study and subsequently stayed there. The speakers are heterogeneous to reflect, to a limited extent, the range of accents in a region. Longer amounts of time spent in London does not necessarily lead speakers to lose their accent (for example, the third Liverpool speaker has spent 34 years in London but retains his accent).

Materials

Two texts were generated which contained words with consonants and vowels whose phonemic function and/or phonetic realisation potentially distinguish the regional accents of the speakers from one another and from HC English.

The design constraints on the texts can be illustrated by the following simplified example (Table 2) with three accent regions (Home Counties, Newcastle and Liverpool). In this example, it is assumed that HC English manifests one feature which Newcastle and Liverpool do not and which distinguishes it from them when it occurs in the text. Liverpool has a second feature that distinguishes it from HC English and Newcastle, and a third feature distinguishes Newcastle from Liverpool and HC English.

TABLE 2.

Table to illustrate idealised set of features (hc, nc and li) that separate Home counties English (HC), Newcastle (NC) and Liverpool (LI) accents respectively.

	ACCENT
	HC	NC	LI
hc	+	-	-
Feature	li	-	-	+
nc	-	+	-

It is also assumed, for this illustration, that the features are used all the time by speakers of the particular accent and are therefore reliable markers. Of course, one feature alone can rarely uniquely identify an accent; it is the combination that provides the profile. The number of accent features determines the degree of difference between accents. This leads to a third assumption, namely that the presence or absence of a feature is a defining property of an accent. Since a phonetic feature implies articulatory and acoustic properties, the absence of a particular feature implies a change of articulatory pattern and a change of acoustic identity.

Obviously, the assumptions behind this feature characterisation of regional accents is a simplification, but material that includes features that have a statistical tendency to operate in this way, can be used to differentiate accents. To illustrate the nature of the idealisation, the /Θ/-/A⌉/ opposition is considered which distinguishes “ant” (/Θντ/) and “aunt” (/A⌉ντ/) in Home counties English but not in many Northern English accents (transcriptions are given in SIL Doulos IPA 93 notation throughout this paper). A Northern English speaker may well abandon the “ant – aunt” homophony, producing a long vowel in “aunt” and other words (e.g. “glass, pass, fast, last”). However, even though the HC English distinction is adopted, it does not necessarily mean that the HC English vowel timbre is produced. There is a Northern English long A-vowel in words like “father, farther, palm, calm” etc., which is realised phonetically with a more “fronted” (palatal) quality: [φα=⌉Δ↔], [πHα==⌉μ]. Adopting the long A for words which have short A in Northern English and long A in HC English, while retaining the Northern English long A quality, is a change in the direction of an HC English accent, but it is still likely to judged as different from it.

A wide sample of accent features was used in the prepared texts (see appendix A for a phonetic description and a table showing their distribution across the accents). They were derived from the descriptions of British accents given by Wells (1982), Hughes and Trudgill (1996) and Foulkes and Docherty (1999). Two texts (given in appendix B) were generated using 17 features with the primary goal of achieving a roughly equal difference weighting between HC English and the regional accents, but with the aim also of separating the regional accents.

The features on which the composition of the texts was based, separate each of the regional accents more or less equally from HC English in terms of the number of distinguishing features, but the feature distances between the regional accents themselves vary (Table 3). This is linguistically inevitable because the accents did not arise so as to be equidistant from one another, and it is historically inevitable because some accents are more closely linked. Thus, lowland Scots and Northern Irish can be seen, as expected, to be closer than Liverpool and either lowland Scots or Northern Irish.

TABLE 3.

Distances between accents (based on the 17 features specified in the lower part of Appendix A)

	NC	SC	NI	HC
LI	9	14	16	7
NC	-	11	9	6
SC	-	-	2	8
NI	-	-	-	9

In summary, the feature-inventory defining the accents provided a basis for the generation of the accent-differentiating texts, and it provides a framework against which to evaluate the perceptual similarity of the accents to listeners. The feature distances between the accents are an abstract statement of difference potential; they cannot predict the degree to which any speaker may realise that potential. However, this is immaterial within the present study, since it is not intended to analyse the individual speakers' phonetic realisations in comparison to HC English, but merely to locate them within the accent space relative to HC English, and ascertain (in experiment two) whether their position in that space is affected by a modification of the auditory feedback conditions.

The material was, however, also subjected to auditory phonetic scrutiny which confirmed a difference in the strength of regional accent concomitant with the claims of the speakers. Residual accent in speakers who reported that they had lost their accent varied from almost non-existent to slight; it was moderate to strong in those who claimed to have retained their accent. Typical residual features were identifiable in the Scots and Northern Irish /αY9/, the Liverpool and Newcastle short central realisation of HC /A⌉/ in words such as “glass”, the Liverpool retention of /Nγ/ where HC English has /N/ (“singer”, “thing”) and the use of a closer rounded vowel (close to [Y]) for HC /℘/ as in “hut”. In only one speaker, a Northern Irish speaker who reported that he had retained his accent, were there recognizable traces of a regional intonation. This general lack of regionally differentiated intonation may be attributable to the reading task.

Procedure for recordings

The two texts were both read under normal listening conditions. (The speakers were also recorded under DAF and FSF at the same recording session. These procedures are described in the method of experiment two, as these recordings are not used in the current experiment). All recordings were made in an AVTEC amplisilence sound-treated booth. Speech was transduced with a Sennheiser K6 microphone and recorded on a DAT recorder. The recordings were transferred digitally to a PC for the perceptual tests.

Listeners

Eight listeners were recruited, four male and four female, aged between 18 and 25 (mean age 20.3 years). None reported a history of speaking or hearing problems, and none had any special speaking- or listening-training. All were native to London and normally resided in that city.

Procedure for perceptual tests

Listeners heard two texts for all 190 possible pairings of the 20 speakers. The text used was selected randomly and speakers in a test pair were also selected randomly subject to the constraint that each pair of speakers was heard once only. The selected pair of texts was played to the listener over headphones. They were instructed to listen to the text to judge the similarity of accent on a difference rating scale of 1-7 (1 – same, 7 – different). Apart from being told to rate accent similarity, the listeners were not given any further instructions about how to make the judgements. The listeners could hear the texts as often as they desired and they were self-paced to avoid fatigue. A complete assessment of the 190 pairings took approximately four hours. All ratings were made on the same day to avoid long-term changes in accent judgements.

Results

The degree of reliability of ratings between participants was high (Cronbach's alpha = 0.970 across the eight listeners), ensuring that idiosyncratic differences between listeners contributed little to their decisions about accent similarity. Similarity ratings were averaged across listeners, yielding one value per speaker pairing. All speaker pairings involving HC accents were selected for the first analysis. The four speakers with HC accents were rated as extremely similar overall (across all possible pairings, M=1.33, sd = 0.39). For each non-HC speaker, similarity to each of the four HC speakers was assessed, reported in Table 4.

TABLE 4.

Average similarity ratings between HC speakers and speakers with regional accents, as a function of regional accent group and self-reported accent loss or retention

	LI	NC	NI	SC
LOST ACCENT	3.40	2.57	4.46	4.04
RETAINED ACCENT	6.66	6.22	6.49	6.82

Factorial ANOVA was conducted upon the mean similarity ratings to HC speakers, using HC speaker as a random factor, and investigating the effects of accent group (LI, NC, NI, SC) and accent retention (lost, retained). The main effect of accent group was significant (F(3,56) = 3.330, p=.026), reflecting overall differences in the similarity between different regional accents and HC accents. Given the limited number of speakers investigated, these results could simply indicate gradations in “accent loss” rather than quantitative differences between accent groups in this task. The main effect of accent retention was also significant (F(1,56) = 113.983, p <.001), indicating that overall, speakers who self-reported as having lost their accent were indeed rated as more similar to HC speakers than were speakers who retained their accent. The interaction between accent group and accent loss was not significant (F(3,56) = 1.609, p = .198), indicating that accent loss was behaviorally comparable across regional accent groups.

Average similarity ratings between accents under normal feedback were taken, and subjected to hierarchical clustering using the average linkage algorithm, in order to establish the patterns of similarity among accents in normal listening conditions.

Figure 1 indicates the similarity properties among the accents. In the figures, speakers who reported losing their accents are marked “-” and speakers who retained their regional accents are marked “+”. Home Counties speakers are marked HC, Northern Ireland speakers NI, lowland Scotland speakers SC, Liverpool speakers LI and Newcastle speakers NC. Similarity between speakers is indicated by the point on the x-axis at which vertical connections occur. For example, the two HC speakers depicted at the bottom of the Figure are connected by a vertical marker at a value of 1.0, indicating the similarity value between these speakers. The group made up by these two very similar speakers is then connected to the next HC speaker at a value of about 1.2, indicating the average similarity between these two speakers and the third. In a similar manner the next connection (with the last HC speaker) at a value of about 1.5, indicates that this latter speaker is rated, on the average, as 1.5 units away from the other three HC speakers (but provides no further details about which of those speakers is more or less similar). Arrangement of speakers along the y-axis is determined by group average similarity as depicted by their vertical connections. The Figure illustrates not only the high similarity between the four HC speakers (clustered together at the bottom of the Figure), but also an overall correspondence between speakers' ratings of their accent loss/retention and listeners' similarity ratings. Most of the speakers who claimed to have lost their accent cluster with the HC speakers, and those who claimed to have retained their accent cluster far from the HC speakers. To summarize, speakers' judgments about whether or not they had lost their regional accents were largely reflected in the average similarity judgments.

Figure 1.

Similarity tree for participants under normal listening conditions. + depicts a speaker who has retained the regional accent, - depicts a speaker who has lost the regional accent. HC stands for Home Counties English, NC for Newcastle, NI for Northern Irish, SC for lowland Scotland and LI for Liverpool.

Discussion

The results of the ANOVA and the similarity dendrogram obtained from the hierarchical clustering analysis indicate that HC English speakers separate from other accent groups. Speakers who reported they had lost their accent tended to be rated as more similar, and thus occupy a position closer to HC English speakers than their accent counterparts who considered that they retained their accent.

The texts were clearly successful in distinguishing accents and also in differentiating perceived distances between accents. Of course, it is not possible to say which of the assumed regional features lay behind the judgements, nor what their relative perceptual effect might have been for the listeners. Graded phonetic attributes like the Northern shift of short /Θ/ and the long /A⌉/, discussed above, from a front and back vowel quality, respectively, to a more central quality, are likely to be important triggers for regional accent identification though they are not captured with the mainly categorical type of descriptor used here. There is graded exponency of accents in terms of the realization quality of the phenomenon being investigated. So, though there is a degree of latitude in the realization of e.g. /æ/ (particularly as HC English now accepts a more open, slightly centralized quality), at some point the degree of centralization would be too Northern, and would then signal non-HC English. This variable phonetic quality of the /æ/ words is, however, not captured in our features which represent the categorical presence or lack of a difference between word pairs like LASS and GLASS. In this case, either the GLASS vowel is long, or it is as short as the vowel in LASS.

It appears (given the success of the analysis) that these accompanying gradient properties varied randomly across accent types so they did not make some accents more apparent than others. The auditory-phonetic examination of the recordings reveals that the degree to which a regional accent is present varies to some extent within as well as between the unmodified and modified groups. The modified group can be described generally as retaining residual phonetic properties of the type discussed above, particularly in terms of vowel quality, having mainly adopted the standard phonemic oppositions.

These findings show that the text is suitable for revealing accent differences and that the self-report of these speakers is, in most cases, a reasonable reflection of whether accent has been lost or retained. The question of whether changing listening environment elicits original accent (operationalized as a shift in similarity patterns by speakers who retain their accent) is addressed next.

Experiment two

In experiment one, the majority of speakers who considered they had lost their accents were rated as closer to HC English than their counterparts who claimed to have retained their accent. Positioning of speakers who lost their accent close to HC English permits the effects of alteration to listening conditions on speakers' accents to be established in this experiment.

It is predicted that, when speaking under DAF and FSF conditions, speakers who have lost their original regional accent (as demonstrated in experiment one) will be rated as less similar to HC English and more similar to the speakers who have retained their regional accent when speaking under DAF and FSF conditions. This specific shift in position in perceptual space towards original regional accent when listening environment is altered, would support the view that accent re-emerges rather than speech under alteration simply being judged as sounding odd (see the discussion of Howell and Dworzynski's 2001 results in the introduction).

Method

Materials

The recordings of the Northern Irish, Scots, Liverpool, Newcastle and Home Counties English speakers made for experiment one were also used in experiment two. HC English is the accent that these speakers adopt when they give up their original regional accent. The speakers in the accent groups are the same as in the previous experiment (four speakers per group). Two speakers in each accent group considered that they had lost their accent, and two speakers that they retained their accent. The speech from the normal listening environment (used in experiment one) is used as well as speech produced in DAF and FSF environments.

In the DAF condition, subjects heard their own speech binaurally at a 66ms delay over Sennheiser HD480II headphones. Level over the headphones was set at about 70 dB SPL and periodically checked. A Digitech model studio 400 - signal processor produced the DAF delays. Subjects were told that they would hear their voice altered over the headphones, which they should ignore. A Sennheiser K6 microphone was used to record vocal responses directly onto a DAT recorder for use in the analysis, as in normal-listening conditions. The output of an additional Sennheiser K6 microphone was relayed via a Quad microphone amplifier to the Digitech model studio 400 - signal processor to produce the required signal alteration. This was then played back with a 6dB gain.

In the FSF condition, the speakers heard their own speech frequency modified. Again the Digitech signal processor was used to modify the original speech signal, shifting the whole speech spectrum down by half an octave.

Listeners

Thirty-two listeners were recruited for the experiment, none of whom had taken part in the first experiment. They were selected according to the criteria outlined in experiment one. Ages ranged from 18 to 27 and there were equal numbers of listeners of each sex. The listeners were randomly assigned to one of four conditions (eight listeners per group).

Procedure for perceptual tests

The experiment was designed to prevent listeners learning a speaker's accent under altered listening and then employing it to make decisions about that speaker's accent when in normal listening conditions. To achieve this, no listener heard a speaker under both a normal and an altered listening condition. There were four groups of listeners, each of which heard the samples of two speakers in an accent group under normal listening and the other two under one of the altered listening conditions. One speaker in each pair considered that he or she had retained their original accent, while the other considered he or she had lost their original accent. Allocation of speakers and the listening conditions they spoke under was constant for a listening group, and only one type of altered listening condition was heard by each listener group (FSF for groups one and two and DAF for groups three and four). Whether a speaker was heard in normal or altered listening was counterbalanced across groups one and two and across groups three and four to ensure that there was nothing unusual about the speakers selected to be heard under DAF or FSF. This design also ensured equal numbers of accented/unaccented samples in each listening group. Thus the four listening groups were differentiated by type of altered listening and by which speakers they heard under altered listening. The procedure for listening to accent pairs and making similarity judgements was the same as in experiment one. In order to rule out the possibility that any differences in results were due to the presence of speech under altered feedback, similarity judgements for pairs of speakers in normal feedback for each testing group were also assessed to replicate the pattern of similarity obtained in Experiment 1. This is necessary to ensure that any accent group differences are not due to the different composition of the items in the present experiment.

Results

First, as in Experiment 1, inter-rater reliability was assessed for each of the four groups of listeners in order to ensure the extent to which idiosyncratic variation among speakers affected ratings of accent similarity. Responses of listeners in each of the four groups correlated highly (Cronbach's alpha values = 0.987, 0.989, 0.991, 0.992).

Next, listeners' ratings of speakers under normal feedback conditions were assessed, in order to rule out the possibility that differences were due to distortions of speech resulting from speaking under manipulated feedback. Average similarity ratings for pairings in the different feedback conditions were combined across the four listener groups, and assessed as in Experiment one (see Table 5).

TABLE 5.

Average similarity ratings between HC speakers and speakers with regional accents, as a function of regional accent group and self-reported accent loss or retention, under different conditions of feedback

	LI	NC	NI	SC
Normal feedback
Lost accent	3.27	4.23	4.53	2.54
Retained accent	6.94	6.77	6.86	6.91
DAF
Lost accent	6.78	6.35	6.09	6.17
Retained accent	6.89	6.88	6.84	6.94
FSF
Lost accent	6.23	5.84	5.00	5.84
Retained accent	6.98	6.94	6.82	6.92

For the normal feedback conditions, as in Experiment 1, the HC speakers were rated as extremely similar (M=1.11); An ANOVA contrasting accent group and loss/retention of HC again revealed a strong effect of accent loss/retention under normal feedback conditions (F(3,128) = 333, p < .001). These results demonstrate that speakers with accent loss were indeed perceived as more like the HC speakers than those speakers who retained their accent, despite the possible differences due to the variation in rating contexts between Experiments 1 and 2.

Analyses were then conducted to investigate ratings in the altered feedback conditions. Similarity ratings were compared directly to assess the extent to which altered feedback produced more regional-sounding accents than speakers who reported losing their accents. It should first be pointed out that, even under altered feedback conditions, listeners continued to rate different HC speakers as extremely similar (average rating 1.21 in DAF condition, 1.13 in FSF condition). This is important, as it indicates that altered feedback does not universally reduce similarity among speakers.

Because “retained accent” speakers were rated as extremely dissimilar to HC speakers in all three feedback conditions, the following analyses focus only upon those speakers who had lost their regional accents. First, a 3×4 ANOVA was conducted upon similarity ratings to HC speakers, investigating the effects of feedback type (normal, DAF, FSF) and accent group (NC, NI, LI, SC). There was no main effect of accent group (F(3,116) = 1.515, p = .214. There was a main effect of feedback condition (F(2,116) = 61.589, p < .001). Further investigation showed similarity to HC accents was greatest under normal feedback, and least under DAF (all feedback conditions significantly differed, p <.05). These effects were qualified by a significant interaction (F(6,116) = 3.624, p =.002), such that the greatest effects of altered feedback were observed for the accents that were rated as most similar to HC accents under normal feedback conditions.

These results demonstrate that “lost accents” diverge from HC accents under altered feedback, but it is still necessary to demonstrate that altered feedback produces speech that is more similar to the original accent. To test this directly, similarity ratings for “lost accent” speakers were compared to: 1) the averaged judgements of similarity to the HC English speakers, and 2) his/her accented counterpart. If speakers show signs of their lost accents under altered feedback conditions, this should be revealed as less similarity to HC English speakers (as illustrated above), coupled with greater similarity to the speaker who has retained his/her accent. The different accent groups were combined for this analysis, where a two-factor repeated measures analysis of variance was used to investigate comparison group (two levels; similarity of a speaker to same regional accent speakers, similarity to HC English speakers) and feedback condition (three levels; normal feedback, DAF, FSF).

The main effect of accent group was significant overall, F(1,7)=11.571, p=.011, reflecting greater similarity between “lost accent” speakers and the corresponding regional accent than to the HC English speakers across all feedback conditions. The main effect of feedback condition was also significant, F(2,14)=7.394, p = .006, indicating differences in similarity across feedback conditions. These main effects, however, were qualified by a significant interaction, F(2,14)=12.582, p=.001. Investigation of simple main effects revealed that “lost accent” speakers were rated more similar to those speakers who had retained their accents, but only under conditions of altered feedback. This is shown in Table 6 where average similarity ratings between “lost accent” speakers and other speakers are given for each auditory feedback condition. Under normal listening conditions, the similarity rating between speakers who had lost their accent and those who had retained their accent did not differ from the rating between speakers who had lost their accent and HC English speakers, illustrating that these speakers still retained some aspects of their original accents. On the other hand, under DAF and FSF, similarity ratings between speakers who had lost their accents were more similar to those of speakers from the same region and less similar to Home Counties English. Crucially, this effect was present using speakers' own opinions of whether they had lost their regional accents or not, which was not always reflected in listeners' ratings of similarity.

TABLE 6.

Average similarity rating between “lost accent” speakers and other speakers as a function of auditory feedback condition, including all speakers who self-reported as having lost their accents: 1 = most similar, 7 = least similar.

SIMILARITY TO:	SAME REGIONAL ACCENT	HOME COUNTIES
Normal listening	3.74	3.62
DAF	2.67	5.71
FSF	2.64	6.33

Similarity dendrograms were then prepared as in Experiment 1. Because speakers were divided into two groups in order to allow the investigation of normal vs. altered feedback, two separate Figures are presented for each feedback contrast (corresponding to entirely different sets of speakers) The similarity clustering among speakers is given in Figure 2 for normal feedback, Figure 3 for the DAF condition, and Figure 4 for the FSF condition. These figures clearly indicate that with only a single exception (one lost-accent Northern Irish speaker in the DAF condition, NI- in the left panel of Figure 3), the speakers who had lost their accents now exhibited extremely strong tendencies to cluster with the accented speakers, rather than those from the HC English group under conditions of altered feedback. These results converge in indicating accent shifts away from an HC English accent towards a stronger regional accent under altered feedback; this overall trend was also confirmed in the auditory-phonetic examination. The change was manifested either in a greater frequency of features already present in the recordings without modified feedback (e.g. /Nγ/ in one Liverpool speaker) or more extreme variants of non-HC English vowel realizations). In no cases were there recognizable regional intonation patterns, presumably because all speakers raised the average pitch and loudness and spoke with an extremely restricted pitch range, in some cases almost monotonous, as a reaction to the altered feedback. Since these effects applied as much to the HC English speakers as to the other regional speakers, it must be concluded that the judgements were based primarily on the segmental changes.

Figure 2.

Similarity tree for participants under normal listening conditions. + depicts a speaker who has retained the regional accent, - depicts a speaker who has lost the regional accent. HC stands for Home Counties English, NC for Newcastle, NI for Northern Irish, SC for lowland Scotland and LI for Liverpool. Speakers were divided into two groups for the rating task, depicted in the left and right panels of the Figure.

Figure 3.

Similarity tree for participants under DAF conditions. + depicts a speaker who has retained the regional accent, - depicts a speaker who has lost the regional accent. HC stands for Home Counties English, NC for Newcastle, NI for Northern Irish, SC for lowland Scotland and LI for Liverpool. Speakers were divided into two groups for the rating task, depicted in the left and right panels of the Figure.

Figure 4.

Similarity tree for participants under FSF conditions. + depicts a speaker who has retained the regional accent, - depicts a speaker who has lost the regional accent. HC stands for Home Counties English, NC for Newcastle, NI for Northern Irish, SC for lowland Scotland and LI for Liverpool. Speakers were divided into two groups for the rating task, depicted in the left and right panels of the Figure.

The literature on foreign language learning suggests that 1) there are only weak effects of length of residence on pronunciation proficiency, but that 2) first language accent retention tends to be stronger in older than in younger second language learners. The data from experiments one and two were examined by regression analyses to check whether similar regularities are true for the acquisition of a new accent. The analyses were made irrespective of designation of participants as having lost or retained their accent. In all the regression analyses, the dependent variable was the average rating of a given speaker's similarity to the HC speakers in the corresponding feedback condition excluding HC-to-HC comparisons. The data available from experiment one were the similarity ratings under normal feedback. The data available from experiment two were the similarity rating for normal feedback, DAF and FSF (the last two allow examination of what age factors relate to degree of accent change under altered feedback conditions).

Consistent with observation 1), the length of time spent in London did not correlate significantly with similarity ratings to HC for any of the listening conditions. Thus for speakers moving to London later than in their teens, there is no greater chance of acquiring a HC accent if they spend longer in London.

There was also support for observation 2). Table 1 shows that speakers who lost their accent were younger on average when they came to London than speakers who retained their accent. This parallels the observation that first-language accent retention tends to be stronger in older than in younger foreign-language learners. To see whether older speakers are more likely to retain their accent, length of time in a non-HC accent region was correlated with similarity to the HC accent. There was evidence that the longer a person had spent in the non-HC region, the more dissimilar the accent to HC. This relationship held up under one of the feedback conditions (DAF).Thus, time spent in region was significantly correlated (negatively) with rated similarity to HC accents in the normal listening conditions of experiment one (r(16)=.475, p=.031 one tailed) and experiment two (r(16)=.538, p=.015 one tailed) and for the DAF condition of experiment two (r(16)=.573, p=.01, one tailed).

Discussion

The first experiment showed that speakers who reported that they had lost their original accent were rated more similar to HC English speakers in normal listening conditions. The speakers who had lost their accent shifted position under each of the altered listening conditions (DAF and FSF) so that they were now rated as much more similar to the speakers who retained the same original accent. This shows that original accent re-emerges (the speakers who had lost their accent now occupy a position in perceptual space closer to speakers who still have that accent form). From this, it appears that altering listening condition has a specific effect on restoring accent. This, along with the observation that the HC speakers consistently clustered very closely together under all three feedback conditions, rules out the potential problem in the Howell and Dworzynski (2001) experiment that speech under altered listening conditions is judged as generally odd-sounding rather than as reflecting the speaker's original accent.

An objection may be raised that these results show a lack of consistency in the number of features separating an accent from Home Counties English, as identified by the features in Table 3 for the different regional accents, and the relative perceptual distance of that accent from the Home Counties accent and from other regional accents. However, the complex nature of the accent-defining features as (i) phonological differences with accompanying phonetic realisation differences on the one hand and (ii) as purely phonetic realisation differences with no implications for the phonological system on the other, means that both the degree of phonetic deviation and the perceptual salience of that deviation can vary considerably. An example is the marked difference between Liverpool and Northern Irish in the number of features separating them from the Home Counties accent, while their perceptual difference from the Home Counties accent (and from each other) is much less.

General Discussion

Experiment one showed that speakers who reported they had lost their original regional accent were rated more similar to Home Counties English speakers than speakers who retained their accent. Experiment two showed that the speakers who had lost their original regional accent were rated closer to speakers who retained the same regional accent when they were speaking with DAF or FSF. The findings from the two experiments reported above have implications both for our view in general on the status of regional accent in speech production and on the function of auditory feedback in speech production.

Considering the phenomenon of accent within a general socio-psychological framework, these results provide supporting evidence for the informal observations of shifting accent as a product of ambient auditory conditions. More specifically the results throw some light on the role of auditory information during speech production. A widely accepted view of speech motor control (Perkell, Matthies, Svirsky, & Jordan, 1995; Perkell, Matthies, Lane, Guenther, Wilhelms-Tricarico, Wozniak, & Guiod, 1997; Perkell, Guenther, Lane, Matthies, Perrier, Vick, Wilhelms-Tricarico, & Zandipour, 2000) holds that the establishment of links between auditory feedback on the one hand and tactile and kinesthetic feedback patterns on the other is important at the acquisition stage (Borden, 1980; Guenther, 1995; also Perkell, Lane, Svirsky, & Webster, 1992 for learning in cochlear implant patients) but that auditory feedback cannot logically play a role in moment-to-moment control of speech-sound production, simply because it occurs too long after the event. Studies showing the very slow deterioration of segmental control after hearing loss (Hamlet, Stone, & McCarthy, 1976; Lane, & Webster 1991; Cowie, & Douglas-Cowie, 1992) provide confirmation for this assumption. In the light of this evidence it would appear surprising that speakers change their accent by generating new articulatory patterns in the short term as a result of modified auditory feedback.

The fact that speakers can change their accent long term (experiment one) but that an accent closer to their original accent can be elicited over the short term (experiment two) implies that the alternative accent forms are available to speakers who have changed their accent. It was argued in the introduction that elicitation of an accent that is not currently in use by a speaker, suggests that the response forms used in that particular accent remain available. Administration of DAF or FSF reactivates the original accent form. Similar short-term switches to earlier pronunciation patterns as a result of changes in auditory feedback conditions have been observed in deaf subjects with cochlear implants when the aid is switched off (Lane, Wozniak, Matthies, Svirsky, & Perkall, 1995; Perkell et al., 2000). The interpretation in these studies was that two different internal models for response output exist in parallel. When the auditory feedback information is not sufficient to verify the quality of the speech that is to be produced according to the more recently acquired model, the subject falls back on the more robust model acquired first.

If the argument that different models for responses are available, alteration to listening environment must also affect a low level of speech control (where speakers select alternative responses). This is consistent with a recent view about the effects of alterations to listening environment on speech control (Howell, 2002). Other authors consider, however, that alteration to listening conditions has its effect at higher planning levels in the central nervous system (e.g. Postma, 2000). It is also possible that accent influences these higher levels. Thus, people can recognise accents perceptually, and accent features interact with central parts of the cognitive system such as word-frequency effects (Foulkes, & Docherty, 1999). Though the findings that listeners can detect accents perceptually cannot be disputed, this is only relevant to production of accent assuming production is linked with perception (e.g. as in a feedback-monitoring where the speaker listens to his or her own speech to verify its output). There are many problems for such an account (Howell, 2002) such as the observation that a speaker who loses his or her hearing can continue to control speech (Borden, 1979, 1980; Cowie, & Douglas-Cowie, 1992; Hamlet et al., 1976; Lane, & Webster, 1991). The effects of word frequency on accent (Foulkes, & Docherty, 1999), nevertheless, supports the view that higher central nervous system levels are involved in accent-control. Thus, further work is needed to establish at what level or levels, accent-control occurs.

An interesting post hoc finding in the present study was the fact that not only were the subjects who had retained their regional accent the ones who had come to the London area later in life, but also that there was a link between the age of the subjects who claimed to have lost their regional accent and their distance from the HC reference accent under altered feedback conditions. This seems to parallel the findings in second and foreign-language learning research, which demonstrate the more native-like pronunciation in the foreign language and the greater resistance to interference from L1 of those learners who were exposed to and learned the foreign language at a younger age (cf., for example Olson, & Samuels, 1973, and Flege, 1999, for an overview). However, the trend in the present study is neither perfectly consistent, nor is the sample of speakers large enough to make sense of deviations from the expected pattern. However, the apparently more robust evidence from foreign-language acquisition may also not be understood as a hard and fast rule. Recent work by Bongaerts and his associates at Nijmegen (Bongaerts et al., 1997; Bongaerts et al., 2000) show that the age factor is not inevitably dominant, and that there are adult learners who are capable of acquiring native-like pronunciation. The strength of a person's motivation to be integrated into the community is apparently a further (complex) factor. How strong the parallels are between L2 pronunciation learning and the modification of an L1 accent, and how robust a native-like L2 pronunciation is to disruptive auditory conditions remain open questions for future research.