21
LANGUAGE AND SPEECH, Vol. 23, Part 3, 1980 213 STYLISTIC VARIATION AND EVALUATIVE REACTIONS TO SPEECH: PROBLEMS IN THE INVESTIGATION OF LINGUISTIC ATTITUDES IN SCOTLAND SUZANNE ROMAINE University of Birmingham This paper describes some of the difficulties involved in conducting language evaluation tests in Edinburgh and reports some results of a pilot study. Samples of speech were obtained from six different speakers, each one reading a text and speaking casually. These were presented to 10 subjects using Lambert's matched-guise technique. Subjects were asked to evaluate the speakers in terms of paired characteristics on a semantic differential scale. As in other evaluation experiments, the responses to linguistic behavior appeared to be mediated through the reaction to social groups. In this case, two dimensions of evaluation were highly salient: perceived identity of the speaker and the way in which the speech style of the speaker was evaluated in two different contexts. INTRODUCTION The investigation of linguistic attitudes is a crucial part of an ongoing sociolinguistic survey of Edinburgh speech, whose ultimate aim is to examine both actual speech and informants' attitudes towards it.^ This paper describes some of the difficulties involved in conducting evaluation tests in a Scottish linguistic situation, and reports some of the results of a pilot study. There are a number of problems in obtaining information about attitudes towards languages or language varieties which are common to all linguistic situations. The transla- tion of attitude from the subjective domain into something objectively measurable is a common problem in any research that involves social categorization and/or perceptual judgments. Labov's (1966) investigation of New York City speech points out some of the more obvious difficulties that arise in dealing specifically with linguistic attitudes. Labov comments that although most informants have very strong opinions about language and are able to detect the presence or absence of certain stigmatized or socially diagnostic features and evaluate these in social terms with great regularity, the reactions to many linguistic variables are inarticulate responses below the level of conscious aware- ness. Conscious attention, it seems, focuses only on those items that have risen to the level of social awareness and have become part of the accepted folklore about language. This is why Labov (1966, p. 482) says that most perception of language is, in fact, perception of social experience and socially accepted statements about language. Further- ^ The sociolinguistic investigation of Edinburgh speech is a research project iu the Department of Linguistics at the University of Edinburgh supported by a grant from the Social Science Research Council (cf. Romaine, 19 78).

Romaine 1980 Stylistic Variation and Evaluative Reactions to Speech

Embed Size (px)

Citation preview

LANGUAGE AND SPEECH, Vol. 23, Part 3, 1980 213

STYLISTIC VARIATION AND EVALUATIVE REACTIONS TO SPEECH:PROBLEMS IN THE INVESTIGATION OFLINGUISTIC ATTITUDES IN SCOTLAND

SUZANNE ROMAINE

University of Birmingham

This paper describes some of the difficulties involved in conducting language evaluationtests in Edinburgh and reports some results of a pilot study. Samples of speech were obtainedfrom six different speakers, each one reading a text and speaking casually. These werepresented to 10 subjects using Lambert's matched-guise technique. Subjects were asked toevaluate the speakers in terms of paired characteristics on a semantic differential scale. As inother evaluation experiments, the responses to linguistic behavior appeared to be mediatedthrough the reaction to social groups. In this case, two dimensions of evaluation were highlysalient: perceived identity of the speaker and the way in which the speech style of thespeaker was evaluated in two different contexts.

INTRODUCTION

The investigation of linguistic attitudes is a crucial part of an ongoing sociolinguisticsurvey of Edinburgh speech, whose ultimate aim is to examine both actual speech andinformants' attitudes towards it.^ This paper describes some of the difficulties involvedin conducting evaluation tests in a Scottish linguistic situation, and reports some of theresults of a pilot study.

There are a number of problems in obtaining information about attitudes towardslanguages or language varieties which are common to all linguistic situations. The transla-tion of attitude from the subjective domain into something objectively measurable is acommon problem in any research that involves social categorization and/or perceptualjudgments. Labov's (1966) investigation of New York City speech points out some of themore obvious difficulties that arise in dealing specifically with linguistic attitudes.

Labov comments that although most informants have very strong opinions aboutlanguage and are able to detect the presence or absence of certain stigmatized or sociallydiagnostic features and evaluate these in social terms with great regularity, the reactionsto many linguistic variables are inarticulate responses below the level of conscious aware-ness. Conscious attention, it seems, focuses only on those items that have risen to thelevel of social awareness and have become part of the accepted folklore about language.This is why Labov (1966, p. 482) says that most perception of language is, in fact,perception of social experience and socially accepted statements about language. Further-

^ The sociolinguistic investigation of Edinburgh speech is a research project iu theDepartment of Linguistics at the University of Edinburgh supported by a grant fromthe Social Science Research Council (cf. Romaine, 19 78).

214 Linguistic A ttitudes in Scotland

more, this perception is mediated through a stereotyped view of a group which is believedto speak in a given way.

It is thus common, both in New York City and in Scotland (cf. especially Romaineand Reid, 1976; Romaine, 1975; Macaulay and Trevelyan, 1973), for informants tocondemn the language of a person or even a whole group as "sloppy" or "rough" becausethis is the way in which the individual or group is perceived. Language serves as a symbolof a particular social identity. Most informants, if pressed as to which particular featuresthey find offensive, are able to mention only a few stigmatized and much talked-aboutstereotypes such as glottal stops, to take a Scottish example, although it is likely that theunfavorable reaction is the result of the co-occurrence of a number of features.

In Scotland there appears to be an added dimension of variability in lexis, whichcan combine in particular with phonetic or phonological variation to modify the evalu-ational reaction to speech. There is a kind of "middle-class folklore" about what con-stitutes acceptable Scottishness in speech, so that a middle class speaker can "get awaywith," so to speak, the use of a number of marked Scotticisms, provided they occuragainst the background of a middle-class and not a working-class accent (cf. Romaine,1975). Another problem in dealing with linguistic attitudes is that most informants donot have a vocabulary of socially and linguistically meaningful, precise terms with whichthey can evaluate speech.

A consideration of these and other factors indicated that an investigation of linguisticattitudes in Scotland required several different approaches. Attitude is a more generalconcept than can be accurately determined from the answer to a specific question, orfrom the responses of an informant in a carefully controlled experimental situation. Itwas therefore decided that this problem should be explored more fully through twomethods: experiments that attempted to quantify certain dimensions of informants'linguistic attitudes, and a questionnaire to be administered during a later stage of thesurvey, which would complement the information obtained in the experiments.

Since it was expected that single responses obtained to specific direct questions wouldbe a poor indication of a persons's attitude (and would not, in any case, carry muchweight unless they were seen as part of the larger picture), a means of sampling a range ofaspects of the attitudes towards language had to be devised, so that a number of dimen-sions could be used simultaneously. The basis of attitude measurement is that there areunderlying dimensions along which individual attitudes can be ranged. By using anattitude-scaling procedure, an informant's response can be assigned a numerical scoreto indicate his position on a particular dimension of the attitude that is being investigated.Osgood's (Osgood et al, 1957) semantic differential was used here as a means ofquantifying informants' reactions to samples of spoken language.

The semantic differential is a technique of measurement consisting of a seven-pointscale whose endpoints are described by adjectives that are polar opposites. Osgood firstused the technique to measure the correlations given to concepts on different bipolarscales by factor analysis. The results suggested that attitude could be identified withevaluation, and that an attitude scale could be constructed from a series of bipolar ratingscales that measure the evaluative factor. To fon-n an attitude scale then, the investigatorhas to decide on the dimensions of the attitude to be evaluated and find suitable adjective

S. Romaine 215

pairs to describe them.The semantic differential has a number of advantages which made it the most appro-

priate and easily implemented technique for this survey. Perhaps the most important ofthese is that it has been used quite successfully in the investigation of evaluationalreactions to language (cf., e.g., Uldall's 1960 work on intonation contours). Even morerelevant to this study, however, is the work of Lambert (cf. especially Lambert et ai,1960; Lambert, 1967,1969,1971) concerning evaluational reactions to spoken languagesusing this technique. This will be discussed more fully in later sections. It provides us witha large body of data which can be compared with those obtained in this study. Apartfrom the technique's successful'application elsewhere, it has been shown that there arehigh correlations between the semantic differential and the Thurstone and Guttmanscales. These correlations provide additional support for the use of the evaluativedimension of the semantic differential as an attitude-scaling technique (cf. Moser andKalton, 1958).

Lambert has done a number of experiments using the so-called "matched guise"technique, in which judges were asked to evaluate bilingual speakers on the basis of asample of speech in each of two languages (French and English in most cases), withoutknowing that the same speakers were being evaluated twice. At first blush, it appearsthat this technique, though with some modification, has a great deal of relevance to theScottish situation; while not bilingualism in its strictest sense, the linguistic situation issimilar to it, so that a person's set of speaking styles can be seen as spanning a linguisticcontinuum ranging from Scots (i.e., the most fully local variety of Scottish English) toEnglish. Variation occurs in response to a number of social factors which are only justnow being examined in some detail in the ongoing survey of Edinburgh speech (cf.Romaine, 1978).

Neither Lambert's matched-guise nor Labov's subjective-reaction test has been appliedin any sustained and systematically controlled manner in Scotland.^ However, it seemedreasonable to expect two things: 1) It should be possible to obtain widely divergentreactions to the same speakers speaking in different styles, as did Lambert with French-and English-speaking Canadians and Giles (1970) with speakers of different accents ofEnglish. 2) Differing reactions should be isolable to contrastive uses of different variantsof a variable, as Labov managed to demonstrate in New York.

The present study was designed to test these expectations, as well as a number ofevaluative dimensions, to see which were the most relevant. It was hypothesized beforeconducting the experiment that differences in reactions to various speakers would dependon what style the person was using, and that these differences would reflect the listener'sattitudes towards a group that had similar characteristics to those perceived of a particularspeaker, and his attitudes towards members of the listener's own group.

Cf. however, Bratt (1974) for an attempt to apply Lambert's matched-guise techniqueto test evaluational reactions of five-year-old children to different varieties of spokenEnglish. Some critical comments on her approach can be found in Romaine (1975).

216 Linguistic A ttitudes in Scotland

METHOD

Two samples of speech were obtained from six different speakers, four males and twofemales, speaking in two different situations, reading and conversation. These werepresented to 10 subjects, who were asked to evaluate the speakers in terms of sets ofpaired characteristics on a semantic differential scale. The subjects who acted as judgeswere told they were taking part in an experiment to study what could be discoveredabout a person on the basis of a sample of speech.

The subjects were given a total of 20 samples of speech, but they were not told thatsome of the same speakers would appear twice. "Filler voices," i.e., speakers for whomno evaluation was being sought, were added to make up the 20 samples. There weretwo parts to the experiment. In the first part there were 10 speakers, four female and sixmale (i.e., the six matched-guise speakers plus filler voices); each read the same fewsentences from a text, which lasted about 3040 seconds. The subjects were given asample speaker before the actual experiment to familiarize themselves with the pro-cedure.

In the second part of the experiment, subjects were told that they would hear"another" set of 10 speakers, again six males and four females, but that this time thesample of speech from each speaker would be taken from a natural conversation recordedwith that speaker. Samples of speech of about 3040 seconds in duration were thenplayed for each of the "matched guise" speakers along with samples from filler voices.Different speakers were used as filler voices in parts 1 and 2 to confuse the subjects andminimize the possibility of speaker recognition. In the second part of the experifnent,subjects were allowed to hear each sample of speech twice, since in each case somethingdifferent was being said. In part 1 the subjects had been given a sample speaker and knewbeforehand what each speaker was going to say; therefore, they were allowed to hear the"reading guises" only once.

There was a break between the two parts of the experiment to relieve the tedium ofthe evaluation process, and more importantly, to minimize the likelihood of auditoryretention of material from the first part, so that subjects would not be aware that some ofthe speakers appeared twice. If the subjects had suspicions, the break would make itdifficult or nearly impossible to remember the order of the speakers to compare theratings given in part 1 with those in part 2.

The speech samples

The selection of the two types of speech samples was made from a collection of tape-recorded interviews; these contained a reading passage as well as free conversation fromwhich two samples of speech for each matched-guise speaker could be obtained.^

The reading passage was included to give the subjects a standardized fonnat for makingthe evaluation of the matched-guise speakers. The passage was taken from a fairy tale and

•̂ These interviews were collected as part of an SSRC-sponsored project on the intonationof Scottish English. I would like to thank Karen Currie for making these tapes availableto me.

S. Romaine 217

was neutral in terms of content, and since the content was the same for all speakers, itwas assumed that there was minimal extraneous variation along this dimension whichmight have influenced judgment. In linguistic terms all of the speakers were on theirbest behavior, as might be expected from the context of the situation.

The second sample of speech from conversation was included so that some of themore marked variables characteristic of Scottish speech, which do not normally occurin reading, would appear. Variables that were either of interest to me and/or had beenpreviously studied (cf. for example, Macaulay and Trevelyan, 1973 and Romaine, 1975)were represented in the speech samples of part 2. Obviously, only a limited number ofvariables which will ultimately b*e studied as part of the larger investigation of Edinburghspeech were included, since there is a limit to the number of samples of speech thatinformants can reasonably be expected to evaluate. This paper will report some of theresults obtained for the evaluation of speakers' use of glottal stops.* In addition, onespeaker had two guises in part 2 (plus one guise in part 1): one in which no glottal stopsoccurred, and another in which virtually every voiceless plosive was replaced by orco-articulated with a glottal stop.^ This was included to see whether the reaction to aspecific variable might be isolated in a way similar to Labov's subjective reaction testin New York. The samples of speech were chosen not only for the specific instances ofthe variable in question, but also with a view to obtaining neutral subject matter, whichwould not reveal anything about the speaker's background.

Judges and speakers

Since the purpose of the large investigation is to describe Edinburgh speech, all of thespeakers who appeared in matched guises had been born, raised and educated in Edin-burgh. The fact that after the experiment several judges expressed surprise at this isillustrative of the range of variation represented in the speech samples. In fact, oneperson who took part in the experiment expressed profound dismay that she had failedto recognize one of the speakers, DR, as an Edinburgh person, or indeed as a Scot. DR,of course, was chosen precisely for the reason that he is not perceived for all practicalpurposes as an Edinburgh person on the basis of his speech. He speaks RP (i.e., "receivedpronunciation"), which he acquired in a fee-paying school in Edinburgh; nevertheless, asa native of Edinburgh, he is part of the Unguistic continuum. Interestingly enough, bothDR and the woman who failed to identify him as an Edinburgh person were graduates ofthe same fee-paying school. Speaker SM, on the other hand, was selected because sherepresented what might be called middle-class, Scottish Standard English (SSE). There isnot much difference in SM's speaking and reading styles. Speaker EF, however, althoughstill middle class (or perhaps lower middle class), displays quite a contrast in reading and

^ The other samples of speech contained instances of other phonological, lexical andsyntactic variables which will not be discussed here (cf. Romaine. 1978).

5 In Scots, glottal stops may replace voiceless plosives, namely, /t/ (and to a lesserextent in some accents /k/ and /p/) in certain positions. This occurs in medial posi-tion, e.g., bottle; word-final position, e.g., cut; in certain final clusters, e.g., he\i\andinitially in the word to in going to future forms.

218 Linguistic A ttitudes in Scotland

speaking. And finally, speaker DC, for whom three samples of speech are presented, isan upper working-class male. The results for two other speakers (both working-class men)are not presented here, as their samples of speech involved instances of other variables,but the results from these speakers conform to the general trends outlined later.

The subjects who acted as judges were fairly homogeneous in terms of social back-ground; all were connected with the University of Edinburgh, either as students orresearchers in the Departments of Linguistics and English Language. No attempt wasmade to find a stratified random sample of subjects owing to the preliminary nature ofthe pilot study; there is no intention to make any claims about the representativeness ofthe sample for any larger group. The pilot study was being conducted primarily as a testof methodology; for this reason a small non-random sample served the purpose. Apartfrom time considerations and the preliminary nature of the experiment, there is somejustification for the small number of subjects used, which has to do with the allegedregularity of certain forms of social behavior. Labov (1966) has claimed that of all formsof social behavior, language is exceptional in that the pervasive patterns of linguisticstructure are manifested just as reliably in a few tokens and speakers as they are in amuch larger number. The assumption here is that the reaction to linguistic behavior isas regularly structured as the linguistic behavior itself.^ Each subject who acted as a judgewas asked to answer a few brief questions about his or her background so that if therewere any noticeable deviation in response between individuals, an explanation might befound in external factors. In fact, all of the subjects, eight females and two males,responded quite consistently.

The semantic differential

Each subject was given 20 copies of a form on which to make the evaluation of thespeakers. The ratings were made on a seven-point scale with each position closer to orfurther away from either end point of the scale represented by opposite characteristicsindicating the degree to which the judges associated the speaker with a particular category.There were 20 such pairs of opposites listed on each form. Some of these had been usedbefore in evaluational tests, and some were being tried for the first time here (cf.Appendix).

The paired items were chosen because they represent aspects related to language atti-tudes or are dimensions along which evaluation may be seen to take place. The individualitems have been grouped under the following larger headings in the analysis of the results.

I. Personality traits (Items 2 ,3 ,5 ,8)

II. Status characteristics (Items 1, 7)

III. Absolute characteristics (Items 4, 6)

IV. Speech or code characteristics (Items 11, 14, 17, 18, 19, 20)

V. Identity (Items 9, 10, 12)

VL Solidarity and linguistic security (Items 13, 15, 16)

^ This seemed to be the case in the pilot study: for further details see Romaine (1978).

S. Romaine 219

As can be seen on the actual scoring sheet given to the subjects (cf Appendix), theitems were not presented in these groupings, or with all the positively scored attributeson one side, but in a mixed order. The category headings serve to focus on particulardimensions of attitudes.

Scoring

Scores range from one to seven for each item. A score of seven was assigned to onemember of each pair and a score of one to the opposite end as given in the Appendix. Ahigh score does not always imply a value judgment. For example, whether a personsounds old or young does not imply the same type of value judgment involved in apair such as "educated/uneducated," although in each case we are interested in a relativeand not an absolute judgment. To take another example, in the case of the item "mas-culine/feminine" it was made clear to the judges that the object was not to decidewhether the speaker was a man or woman (which seemed obvious in the samples anyway),but to estimate how masculine or feminine a speaker sounded. This of course implies avalue judgment, but not a "constant" one. The implicit assumption is that it is positivefor a man to sound masculine, but negative for a woman to sound masculine, and viceversa.

In some cases there is external reference for the characteristic in question; in otherwords, the characteristics could be compared with objective or physical facts; e.g., thespeakers were all from Edinburgh whether they sounded like it or not. However, in mostof the paired items this is not the case, e.g., we have no means of estimating the honestyor friendliness of the speakers. This reflects an important difference between attitude-measurement experiments and certain psychophysical ones using similar scales. Thelatter rely on some objective, external physical dimension as a point of reference forcomparison with the experimental stimuli and the responses to them. In this case,however, the emphasis is on the perceived picture of each speaker and the comparisonof each speaker with himself and with other speakers in different controlled contexts.

All items do not necessarily imply a choice between good or bad qualities; a highscore is not always the "best." Average scores were figured for the 20 items for each ofthe matched-guise speakers and presented under the appropriate category heading.

RESULTS

The results in general indicate the existence of patterns similar to those found in othevexperiments involving evaluational reactions to the spoken language. These are summarizedbelow. The figures and tables show the mean ratings assigned to each of the matched-guise speakers in two situations, reading and speaking. The speakers are referred to asEF (Fig. 1, Table 1), DR (Fig. 2, Table 2), SM (Fig. 3, Table 3), and DC (Fig. 4, Table 4).1 will confine my comments to the most important trends that can be observed.

1. The reactions to the different traits of the speakers change depending on the guise ofthe speaker, although these changes are not always in the same direction or of the same

220 Linguistic Attitudes in Scotland

6 . 0

5.5

5.0

4 . 5

4.0

5.5.

3 .0"coO p n

CO "^'^

2 . 0

1.5

1 . 0

.5

0

\\

\

-

1 1

y1

_i

1

Fig. 1. Profile of speaker EF. SoUd line = reading; dashed line = speaking.

1 = mean personality rating2 = mean status rating3 = mean code rating4 = mean identity rating

TABLE l

Profile of speaker EF; mean ratings for personality,status, code and identity

Characteristic

Personality

Status

Code

Identity

MeanReading

4.6

4.1

4.5

5.5

RatingSpeaking

4.5

3.1

3.1

5.9

Direction offrom part

——

change1 to 2

Size ofchange

0.1

1.0

1,4

0.4

S. Romaine 221

CDOin

6.0

5.5U

5.0

4.0

5.0

2 . 5

2 . 0

1.5

1.0

.5

0

Fig. 2. Profile of speaker DR. Solid line = reading; dashed line = speaking.

1 = mean personality rating2 = mean status rating3 = mean code rating4 = mean identity rating

TABLE 2

Profile of speaker DR: mean ratings for personality,> status, code and identity

Characteristic

Personality

Status

Code

Identity

MeanReading

4.5

5.7

5.1

4.2

RatingSpeaking

4.1

3.4

2.9

3.2

Direction of changefrom part 1 to 2

Size ofchange

0.4

2.3

2 2

1.0

222 Linguistic Attitudes in Scotland

CtJo

C/5

6.0

5.0

i-. 5

4.0

^.5

3.0

2 .5

2 . 0

• 1 . 5

1.0

.5

0

Fig. 3. Profile of speaker SM. Solid line = reading; dashed line = speaking.

1 = mean personality rating2 = mean status rating3 = mean code rating4 = mean identity rating

TABLE 3

Profile of speaker SM: mean ratings for personality,status, code and identity

Characteristic

Personality

Status

Code

Identity

MeanReading

4.5

4.9

5.4

5.0

RatingSpeaking

5.1

5.0

4.9

5.8

Direction offrom part

-»-

+

+

change1 to 2

Size ofchange

0.6

0.1

0.5

0.8

S. Romaine 223

magnitude. In general, most of the speakers, with the exception of SM (Fig. 3), the SSEspeaker, are evaluated less favorably on the basis of a sample of conversation than theyare on the basis of a sample of reading. In other words, the speakers are perceived asgenerally being less well-educated, of lower social status, sloppier etc. Not surprisingly,all the the evaluational dimensions relating to speech characteristics show some shift in anegative direction from reading to speaking, since it is a commonly held belief that muchof conversational speech is ungrammatical, sloppy etc.

2. All of the speakers except DR, the RP speaker (Fig. 2), are perceived as sounding moreScottish and more like EdinburgTi people when speaking than reading. This is also not anunexpected result, since the features that are associated with the Scots end of the linguisticcontinuum ranging from Scots to SSE have been functionally restricted to the morecasual and familiar domains of speaking.

3. When the judges considered the question of whether they thought they spoke like anyof the speakers or would like to, all of the Scottish judges (N =8) reported that notonly did they not think they spoke like any of the speakers, but also that they would notwant to, with the exception of SM, the SSE stereotype. DR, the RP speaker, on the otherhand, was emphatically rejected, i.e., none of the judges thought they spoke Hke him,nor did they want to. Two of the subjects were not Scots, and their answers to the samequestions are particularly revealing. DR fared favorably with this group; in fact he is theonly speaker to receive a positive rating from the non-Scottish group in part 2; therest of the more Scottish-sounding speakers are given considerably lower ratings.

Isolated reactions to a variable

As an additional "experiment within an experiment," one speaker, DC, was presentedto the judges in three guises or "styles"; one in which he read the same passage as theother speakers, and two other guises of conversation in which the use of glottal stopswas contrastively controlled. It was hypothesized that different reactions would beobtained on the basis of these samples of speech, and that it was probably likely that anydifferences in evaluation from one conversational style (which contained a high frequencyof glottal stops) to another (which contained no glottal stops) would more or less rep-resent the isolation of a subjective reaction to a specific linguistic variable, in so far asthis is possible (Fig. 4).

This general prediction is confirmed; DC is evaluated differently in almost everyinstance, depending on which guise he appears in. However, the hypothesis that theratings would shift unidirectionally from most positive in reading to least positive inspeaking style 2, which contained a high frequency of glottal stops, with the otherconversational sample of speech representing an intermediate position, was not borne out.A clue to the somewhat skewed distribution of responses can, I think, be found in a closerexamination of DCs "way of speaking." It becomes clear that DC is perceived as beingclearer, more correct, more polite, more careful and less broad in the intermediatespeaking style, rather than in the reading passage. More will be said about this interestingpattern in the final section.

224 Linguistic Attitudes in Scotland

6.0 ^

oCO

4.0

.̂"̂

3.0

2.0

1.5

1.0

.5

0

Fig. 4. Profile of speaker DC in three styles. Solid Une = reading; dashed line = speaking1; dotted line = speaking 2.

1 = mean personality rating2 = mean status rating3 = mean code rating4 = mean identity rating

TABLE 4

Profile of speaker DC: mean ratings for personahty,status, code and identity in three styles

Characteristic

Personality

Status

Code

Identity

Reading

4.8

2.5

3.7

4.8

Mean RatingSpeaking 1

4.2

3.4

4,1

5.8

Speaking 2

4.4

2.4

3.3

5.9

S. Romaine 225

Reactions to Momingside speech

As an interesting sidelight, an example of a "Morningside" accent was included in theexperiment as one of the filler voices in part 2P Although much has been said aboutthis type of accent in impressionistic terms, no one has really described it or attemptedto elicit systematic reactions to it. Yet it is a stereotype which is well-grounded inScottish folklinguistic belief. The great deal of laughter that ensued after the subjectslistened to the Morningside speaker was evidence that they had quickly recognized thelinguistic hallmark of this stereotype, the raised /a/ in the word actually. The resultsshowed that the Momingside sp^ker compared favorably with the other speakers interms of personality and status characteristics, but the judges indicated they would notlike to speak in this way; nor did they consider it likely that the speaker would be afriend of theirs. One other result is very revealing; the Morningside speaker received thelowest rating of all the speakers on the item honest/dishonest.

DISCUSSION

The results of this experiment and others like it are perhaps the most graphic illustra-tion that can be offered in demonstration of the fact that different ways of speaking havedifferent social values and meanings attached to them. As in the case of other evaluationalexperiments that make use of language as stimuU, the responses to linguistic behaviorappear to be mediated through the reaction to different social groups, in so far as theseare characteristically identifiable by speech. In this case, the key to the interpretationof the results lies in two areas of evaluation, perceived identity of the speaker and theway in which the speech style of the speaker is evaluated in two different contexts. Thequestion of identity will be considered first.

Lambert's work with English- and French-speaking Canadians elicited a general stereo-typed reaction to two languages, French and English, in a situation where language as awhole is symbolic of community and national identity. As Lambert expected, thedifferences in the favorableness of the subjects' reactions to the speakers depending onwhich language was used refiected their attitudes towards members of their own groupand members of the other language group. Thus, a low evaluation of a French Canadianguise tells us more about group biases and the social context in which French exists inCanada today than it does about language.

In the Scottish experiment we can see a very strong demonstration of Le Page's(1978) view of speech as an "act of identity." The split between the evaluation ofpersonality and status underlines this. The speaker with the highest perceived statusscores well on attributes linked with socio-economic success, but the more Scottish-sounding speakers, who do not score as well on status, are still very well-liked, i.e., theyreceive favorable personality ratings, in some cases more so than the RP speaker. This

'̂ The term Morningside comes from a residential district in Edinburgh that contains ahigh proportion of the professional and middle classes. It is used to refer to a typeof accent that is posh and hypercorrect (compare "Kelvinside speech " in Glasgow).

226 Linguistic A ttitudes in Scotland

result is not a novel one. Sometimes minority groups evaluate the speech of their owngroup higher than that of groups that are dominant in socio-economic terms. For example,Lambert found that there was a tendency for French Canadian guises to be evaluatedmore favorably than English guises on certain stereotyped characteristics associated withFrench Canadians, e.g., religiousness; although on all the other personality characteristicsboth the English and French judges evaluated the French-speaking guises lower than theEnglish-speaking ones.

Cheyne (1970) found in an evaluative experiment that Scottish subjects rated theirown group as mote generous, friendly, good-hearted, humorous and likeable than Englishspeakers, thus displaying "accent loyalty." Milroy and McClenaghan (1977) report asimilar result for the evaluation of Belfast speech; a southern Irish speaker and an RPspeaker were rated higher on characteristics connected with socio-economic success, butmuch lower than a Belfast speaker on personality characteristics. In fact, the Scottishand Ulster speakers were evaluated similarly in this experiment, thus demonstrating thatthe hnguistic affinity between the two speech communities has a perceptual correlate.The authors conclude that the results of their experiment reflect very well the socio-political circumstances of Ulster today.

Nationalistic feeling can be expected to be very strong at the moment in Scotland, andspeech is certainly one of the most overt markers of in-group identity. The most positiveexpression of linguistic solidarity can be found when, in spite of any socio-economicsuperiority of the dominant group or implied denigration of the subordinate group, aspeaker affirms the linguistic norms of the community in which he lives. This is a keyissue. It "explains," for example, Labov's seemingly paradoxical discovery that peoplemay condemn others in their own group on the basis of the way they speak, but yet theydo not want to change (or do not change) the way they speak themselves (cf. alsoTrudgill, 1974). From what we now know about covert and overt reactions to speech andhow these are interrelated, the conclusion that people speak the way they do becausethey want to must stand as a basic socio-linguistic principle. The acceptance of thisprinciple depends on the recognition that each variety has its own prestige, i.e., eitherovert or covert. More importantly perhaps, people do not dislike the speech of those theyrecognize or perceive as similar to themselves, even though on some occasions they maysay so.

This notion of identity must underlie the interpretation of the results obtained inthis experiment. It explains to a great extent why SM receives a positive rating, as do theother Scottish-sounding speakers on most personality traits, and DR does not. AlthoughDR is a native Scot who was born, raised and educated in Edinburgh, as were the otherspeakers, his speech does not betray his local identity. There is no way on the basis ofsuch a short sample of speech that the judges could be sure that DR was a Scot. SM, onthe other hand, is recognizable as a local Edinburgh product; moreover, she is mostsimilar to most of the subjects who evaluated her, i.e., middle-class women. It might beinteresting to repeat the same experiment with a different group of judges, this timetelling them that all of the speakers are from Edinburgh to see if DR receives a morefavorable rating.

Not surprisingly, social processes that are refiected in linguistic structure are also

S. Romaine 227

manifested in the evaluation of linguistic structure. The way in which linguistic groups aredistributed in society and the attitudes which one group has toward another are importantfactors in predicting language maintenance and shift. This has been demonstrated quiteconvincingly in an experiment by Gatbonton-Segalowitz (1975) in which FrenchCanadians were asked to react to a number of matched-guise speakers and some mono-lingual English and French speakers. The French Canadians were divided into threegroups on the basis of their answers to questions concerning French separatism forQuebec and related socio-political issues: nationaHstic, liberal and non-nationalistic. Thejudges were asked whether each matched-guise speaker would be acceptable as a leaderof a group consisting only of French Canadians. Both pro-French groups, i.e., thenationalistic and liberal, reacted more favorably to the French-only guises than either thedouble-language guises or the English-only guises, while the non-nationalistic groupshowed the reverse pattem.

These findings have considerable relevance in Scotland at the moment, where in themidst of ongoing socio-political change, attitudes towards Scottish nationalism will nodoubt be reflected in the way in which different speakers are evaluated.^ It can beexpected that nationalistic feelings will be manifested in a desire to preserve and cultivatethings that are perceived as Scottish, and there are certainly ample linguistic means fordemonstrating Scottish identity. These and other questions will be investigated in furtherwork by direct questioning of informants and more experiments.

The experiment also indicated that the reactions to speakers were not mediated solelythrough the perceived degree of formality, politeness, correctness etc. of the speaker ineach context. The results reinforce my belief that Labov's notion of the structure of astyUstic continuum is not one that can be applied in any consistent way in Scotland, orindeed in any fairly complex linguistic situation. The reactions obtained to DC in threestyles of speaking demonstrate this. If a stylistic continuum of the type postulated forNew York City existed (which has been taken largely for granted elsewhere, for example,in Norwich, cf. Trudgill, 1974), then we would have expected DC to be perceived as mostformal in reading, less formal in the unmarked conversational style (Sj) and least fomialin the marked conversational style (S2). The fact that this result was not obtained is anindication, I think, of my own, Macaulay's and the Milroys' belief that reading andspeaking are two very different types of speech behavior which are not in the same linearcontinuum, at least not for all speakers.

The gap between reading and speaking is greater for some speakers than for others;in addition, the range of variation appropriate to reading a text can be considered quitenarrow. Both the Milroys (1977) and I (cf. Romaine, 1975, 1979) have cited the failureof certain significant phonological variables to pattern along a continuum from leastformal (spoken) to most formal (reading) in Belfast and Edinburgh respectively assufficient justification for regarding conversation and reading as separate parts of a

^ ft is important to note that this experiment was conducted in late 19 78 before theso-called Scotland Bill, which would have granted Scotland considerable autonomyand provided for the establishment of a Scottish Parliament, was put to Scottish votersin 1979. This referendum was defeated for lack of sufficient support.

228 Linguistic A ttitudes in Scotland

speaker's linguistic repertoire. A greater styHstic difference is often characteristic of manylower middle-class rather than of middle-class speakers. This can be seen in the differingevaluations of the speakers in parts 1 and 2. This stylistic "gap" can be demonstratedimpressionistically if we listen to the matched guises for all the speakers; some speakersreally do sound like different people.

This suggests that Lambert's findings need to be re-examined in view of this result;namely, to what extent are judgments about languages and the groups who speak themmediated through judgments of stylistic propriety? In other words, to what extent was"French/English" perceived as an appropriate or inappropriate vehicle for the speechstimuli in the experiment (cf. also Taylor and Clement, 1974)?

It is not simply the case that differences in evaluation are related monotonically tostyle, so that judges think reading sounds better than conversation, and hence speakersare on the whole perceived more favorably in a reading guise. If this were true, then wewould have expected the shift in evaluation from part 1 to part 2 to be unidirectional forall speakers. Yet all of the speakers (except for SM) are evaluated less favorably on otherdimensions seemingly unrelated to style, when they are speaking in conversation. SM,however, does not "shift" stylistically to the same degree as the others; this is in turnrefiected in the ratings that the judges assign to her; i.e., both guises are similarlyevaluated. This emphasizes the point that reading and speaking do not differ to the sameextent for all speakers.

This point should be viewed again with respect to a more general phenomenon whichLabov (1966) has referred to as "hypercorrection by the lower middle class.',' Thisterm refers to the observation that lower middle-class speakers may typically go beyondor exceed the highest status groups in their tendency to use forms considered appro-priate and correct for more formal styles; hence, there is often a greater gap betweenthe most and least formal styles of these speakers than there is for middle and upperclass speakers.

At first glance, this "crossover pattern" appears to be a deviation from the classicsociolinguistic finding that socially diagnostic variables will exhibit parallel behavior ona stylistic continuum; that is to say, if a feature is found to be more common in the lowerclasses than in the upper classes, it will also be more common in the less formal than themost formal styles, with each social group occupying a similar position in each continuum.When this apparent deviation is viewed in the larger context of the social structuresthat the linguistic structures reflect, this is not really so unexpected. The lower middleclass typically has a high tendency for upward social mobility, and one of the ways inwhich this aspiration is manifested is linguistically.

CONCLUSION

There are still a number of problems or uncontrolled variables in experimental design,which relate not only to the present experiment but also to the larger body of workdone on evaluational reactions to the spoken language. These deserve additional con-sideration here.^

^ A more detailed discussion of these and other issues relating to experimental designcan be found in the larger report.

iS". Romaine 22*̂ ^

In choosing samples of natural speech (rather than synthetic speech, which was ruledout as being too artificial) of two different types, the problem of obtaining reactionsto idiosyncratic voice qualities has not been solved. This difficulty was also encounteredby Milroy and McClenaghan (1977) , who noted that great care was taken to avoid highlysalient and idiosyncratic voice settings. A similar claim could be made for the speechsamples in this experiment , although in further work it might be preferable to ask otherphoneticians to evaluate the speakers in terms of dimensions relating to voice quality.It is no t wi thout considerable importance to examine the effect that voice quality mighthave in influencing judgments of this type , but I think this is a more serious problem forsubjective reaction tests of the type Labov (1966) conducted, where the claim was madethat the reaction to a specific variant of a variable was being isolated. Strictly speaking,this claim is, of course, inflated; even though Labov used carefully controlled readingpassages, the specific variants occurred in context with other speech.

It is no t unreasonable to assume that all available levels of linguistic expression can andwill be used by hearers as diagnostic of the speaker's identi ty. This will include, amongother things, lexical choice, morphological and syntactic variation, phonetic/phonologicalvariation and t empo , in addition to parahnguistic features such as voice quality. The useof natural speech in evaluational experiments does of course permit all this variationto enter into the judgmental process. It is paradoxical, however, that this is perhaps thebest argument we can cite both in favor of and against using natural speech stimuli.Stereotyped features do not occur in a linguistic vacuum; they occur in context with awhole range of other features.

In this experiment and others, e.g., Lambert 's and Milroy and McClenaghan's, reactionswere obtained to speakers/groups as a whole; since certain voice quahties or componentsof these may be characteristic of these groups, then these should occur along withsegmental features. The important thing is, as Milroy and McClenaghan point out , toavoid idiosyncratic voice quahties, which may be perceived as unpleasant, in so far asthese are identifiable. Giles' work (cf. especially Giles, 1970) has demonstrated that thesame pat tern of prestige evaluation could be produced when different stimulus speakerswere used to illustrate various accents as well as when one speaker was used in differentmatched guises. It seems then that without varying the task, there is no easy way toseparate paralinguistic features (of which voice quality is only one), from the segmentalcontext in which they are embedded. If we do , then the experimental situation is verymuch removed from reality, since people never hear speech of this type, let alone evaluateit.

The results of this s tudy did suggest some modifications in the experimental methodthat might make it more easily administered. For example, it seems that certain charac-teristics can be bet ter , i.e., more consistently, evaluated on the basis of speech thanothers , and that certain characteristics that pat tern together might be represented by oneitem that produced a strong response. Fur thennore , some items can be eliminated entirely,since they either produced ambiguous results or did not seem to be relevant to linguisticat t i tudes in this particular case. For example, the dimension "mi ld /b road" was not inter-preted in the same way by all the judges, so that , in fact, both the RP speaker and themost Scottish-sounding speaker were classified as very broad by some judges. This

230 Linguistic Attitudes in Scotland

suggests that any accent that is greatly different from one's own is likely to be perceivedas broad.

An example of an item that did not pattern as predicted, or seem relevant, was thedimension "tough/gentle." Earlier sociolinguistic work both in Britain and the UnitedStates (cf., in particular, Trudgill, 1974; Labov, 1966) indicated that working-classspeech had the connotation of "toughness" and "masculinity," but there was no realevidence of this in these results (cf. also Edwards, 1979a).

Finally, simplification of the task itself might also be profitably considered. Thereis no reason for assuming that there are a given number of degrees of meaningfulness,in this case seven, along which subjects can evaluate a characteristic. A three-point scale,for example, in which a dimension might be evaluated as very, average or not very, wouldyield less finely discriminated judgments, but it is just as likely that it will produceclearer results. One of the drawbacks in the use of the semantic differential with sevenpoints is that judges differ somewhat in their utilization of the entire range of the scale;there is a tendency to avoid the ends.

One final comment is relevant. What is perhaps very remarkable about experiments ofthis kind, which no one to my knowledge has mentioned, is the finding that subjectsundertook the task so willingly; i.e., they were quite willing to make these judgmentsabout others (cf. also Edwards, 1979b). Of course, in an experiment the appropriatebehavior is for people to cooperate. Only one subject mentioned afterwards that he hadexperienced some degree of discomfort in being asked to evaluate other people in thisway. Perhaps this finding is just as (if not more) revealing about human nature than theresults of the experiment.

APPENDIX

Scoring Sheet for the Evaluation of Speakers

Sppaker nn

This

1.

2.

3.

4.

5.

6.

7.

8.

9.

1 person sounds:

educated

negative

friendly

masculine

honest

young

lower class

tough

urban

Points

7

1

7

7

7

1

1

1

7

uneducated

positive

unfriendly

feminine

dishonest

old

upper class

gentle

rural

S. Romaine 231

not Scottish

broad

unlike an Edinburgh person

not like the way I think Ispeak at all

unclear

as if I wouldn't want toknow him/her at all

not Hke the way I wouldlike to speak

informal

correct

rough

careful

The points assigned to each member of a pair of adjectives are given here for reference;they did not appear on the forms given to the judges.

10.

11.

12.

13.

14.

15.

16.

17.

18.

19.

20.

Scottish

mild

like an Edinburgh person

like the way I think I speak

clear

as if he/she could be afriend of mine

like the way I would liketo speak

formal

incorrect

polite

sloppy

7

7

7

7

7

7

7

7

1

7

1

REFERENCES

B R A T T , M . F . (1974). Evaluative reations to speech varieties in 5 year old children. M. Litt. thesis.University of Edinburgh.

CHEYNE, W. (1970). Stereotyped reactions to speakers with Scottish and English regional accents.British Journal of Social and Clinical Psychology, 9, 77-79.

DiL, A.S. (ed.) (1972). Language, Psychology and Culture: Essays by Wallace E. Lambert (Stanford).E D W A R D S , J . (1979a). Social class differences and the identification of sex in children's speech.

Journal of Child Language, 6,121-127.E D W A R D S , J . (1979b). Judgements and confidence in reactions to disadvantaged speech. In H. Giles

and R. St. Clair (eds.). Language and Social Psychology (Oxford).G A T B O N T O N - S E G A L O W I T Z , E. (1975). Systematic variations in second language learning. Ph.D.

dissertation, McGill University.G I L E S , H . (1970). Evaluative reactions to accents. Educational Review, 22, 211-227.G I L E S , H. (1971a). Ethnocentrism and the evaluation of accented speech. British Journal of Social

and ainical Psychology, 10,187-88.G I L E S . H. (1971b). Patterns of evaluation in reaction to RP, South Welsh and Somerset accented

speech. British Journal of Social and Clinical Psychology, 10, 280-281.LABOV, W. (1966). The Social Stratification of English in New York City (Washington, D.C.).LAMBERT, W.E., HODGSON, R.C., GARDNER, R.C. and FiLLENBAUM, S. (1960). Evaluational

reactions to spoken language. In Dil (1972), pp. 80-97.LAMBERT, W.E. (1967). A social psychology of bilingualism. In Dil (1972), pp. 212-236.LAMBERT, W.E. (1969). Ethnic identification and personality adjustments of Canadian adolescents

of mixed English-French parentage. In Dil (1972), pp. 266-90.

23 2 Linguistic A ttitudes in Scotland

L A M B E R T , W.E. (1971). The effects of speech style and other attributes on teachers' attitudes to-wards pupils. In Dil (1972), pp. 338-351.

LE P A G E , R.B. (1978). Projection, focussing, diffusion. Society for Caribbean Linguistics: OccasionalPaper, 9.

M A C A U L A Y , R .K.S . and T R E V E L Y A N , G. (1973). Language, Education and Employment inGlasgow. A report to the Social Science Research Council.

MiLROY, J. and MiLROY, L. (1977). Speech and context in an urban setting. Belfast WorkingPapers in Language and Linguistics, Vol. 2, No. 1, 1-85.

MiLROY, L. and M C C L E N A G H A N , P. (1977). Stereotyped reactions to four educated accents inUlster. Belfast Working Papers in Language and Linguistics, Vol. 2, No. 4, 1-10.

MOSER, C. and K ALTON, G. (1958). Survey Methods in Social Investigation (London).OSGOOD, C, SUCI, G.J. and TANNENBAUM, P.H. (1957). The Measurement of Meaning (Urbana,

IU.).R O M A I N E , S. (1975). Linguistic variability in the speech of some Edinburgh schoolchildren. M. Litt.

thesis. University of Edinburgh.R O M A I N E , S. and R E I D , E .C. (1976). Glottal sloppiness?: A sociolinguistic view of urban speech in

Scotland. Teaching English. The Journal of Teachers of English in Scotland, 9, 12-18.ROMAINE, S. (1978). A Sociolinguistic Investigation of Edinburgh Speech. Interim report to the

Social Science Research Council.R O M A I N E , S. (1979). The social reality of phonetic descriptions. Northern Ireland Speech and

Language Forum Journal, 5, 21-36.T A Y L O R , D.M. and C L E M E N T , R. (1974). Normative reactions to styles of Quebec French. Anthro-

pological Linguistics, 16, 202-218.T R U D G I L L , P. (1974). The Social Differentiation of English in Norwich (Cambridge).U L D A L L , E .T . (1960). Attitudinal meanings conveyed by intonation contours. Language and Speech,

3, 223-234.