-
Language in Society
GENERAL EDITOIIl'ctcr Trurlgill, Chair of Errglish Linguistics,
University of Fribourg
ADVISORY EDI lORSJ.K. Clhanrbcrs, Itrotessor of l-inguistics,
University of'lbrontoRalph F.rsolcl. Prolessor of l-irrguistics,
Georgetown UniversityWilliarn l,abov, Prol'essor of Linguistics,
University of PennsylvaniaLcsley Milroy, Professor o[ Linguistics,
University of Michigan, Ann Arbor
l. Language and Social Psychology, edited by ltowatd Giles and
Robert N. St Clair2. Languagc and Social Nctworks (second edition),
Lesley Milroy3. Thc Ethnograplry oI Cornnrunication (second
edition), Muriel Saville-Tioike4. Discoursc Analysis, Miclnel
Stttbbs5. Tlrc Srtcitrlinguistics of Socicty: Introductiorl to
Sociolingrristics, Volunre 1, Ralph Fax>ld6. 'l ' lre
Sociofinguislics of Language: Irrlroduction to Sociolinguistics,
Volume ll, Ralph FasoLd7. I 'hc Larrgrragc ol 'Chi ldren and
Adolescents: T 'he Acquis i t ion of Conrmunicat ive
Competence,
Strz.ttntrc Ilonutirtc8. Lrrrrguagc, the Scxcs anrl Socicly,
Philip M. Smitlt9. 'l 'f rc Languagc o[ Advert.ising, Tiolrcn
\testergaanl atrl Kinr Sclutfierf 0- Dialccts in Conlact , Petcr
Tiudgi l lI l. Pidgcn and Creole [-inguistics, Peter
Miililhtiusler12. Observing and Atralysing Natural Language: A
Critical Account of Sociolinguistic Method, Izsley MilrcyI 3. Bil
ingual isnt (seconrl etlition), S uzanne Ronnine14.
Sociolinguistics and Second Language Acquisition, Dentis R.
Przstr:rtI5. Pronouns:rnd People: The Linguistic Construction
ofSocial and Personal Identity,
I'etcr Miihlhdusler and Ront ltarrtl(r. Politically Speaking,
Jolm Wil.sottf 7. Thc Language of thc News Me
-
146 Quonti tat iveAnalysis
Stcp 2. Calculate the probabil i t ies o[communication based on
each languagecombination by mult iplying the proport ion at the end
of each row by theproport ion at the head oIeach column, provided
that there is at least one languagein comnron. The result would be
table B.
Table B Tablefor tha calculation of the index of
communicalionfor the example in obiective l6
Group IGroup 2
6
A(7/ r0) AB(r/r0) A C(2/ I0)
Language Attitudes
LANGT]AGE ATTITUDES: WHY AND WHAT?
Much of the earlier part o[ this book has bee n focused on
society as a whole.Societ ies, however, are composed of ind iv
iduals; whatever people do wi thlanguage in a society happens when
somebody ta lks to somebody e lse. In th ischapter and the next ,
we are going to concetr t rate on indiv ic luals and what thcydo wi
th language. Our at tent ion wi l l be drawn to the methods and
theor ies of asocial science we have not encounterecl yet, social
psychology. Onc of t lresubjects social psychologists have found
useful to study is language attitudes.
The study of attitudes in general begins with a decision between
twocompeting theories about the nature of attitudes. Most
language-attitudework is based on a menla l is t v iew of at t i
tude as a state oI readiness; anintervening var iable between a st
imulus af fect ing a person and that person'sresponse (Agheyis i
and Fishman 1970:138, Cooper and Fishman 1974:7) . Aperson's at t i
tude, in th is v iew, prepares her to react to a g iven st imulus
in oneway rather than in another . A typ ical menta l is t def in i
t ion of at t i tu t le is g ivcrrby Will iams (197421): 'Attitucle
is considered as an interrral state aroused byst imulat ion of some
type and which may mediate the organism's subsequentresponse.' This
view poses problems for experimental method, because if anat t i
tude is an in ternal s tate of readiness, rather than an observablc
response,we must depend on the person's reports of what their
attitudes are, or inferattitudes indirectly from behavior patterns.
As we know, self-reported dataare olten of questionable validity,
and inlerences from behavior take theresearcher one step away from
what he has actually observed. A great deal ofeffort in
language-attitude research has gone into devising ingenious
experi-ments designed to reveal at t i tudes wi thout making
subjects over ly consciousof the process.
The other view of attitudes is the behavioru, view. On this
theory, attituclesare to be found simply in the responses people
make to social situations. TIrisviewpoint makes research easier to
undertake, since it requires no self-reports
B(5/ tolAI](2/ to\BC(3/ t0)
t4/ t00
5/100
2/ too
3/ 100
4/ t00
6/tQO
Step 3. Acld t l re proport ions in the table. The resul t is
34/100.Step 4. Sclcct the correct answer f rom the choices
given.
l l l l e ab l c t o r ccogn i ze a co r rec t s t a t cn ren t
o f wha t Kuo ' s i ndex o f commun i ca t i v i t ynlcasurcs -
IU Dc ablc to rccogrr iz ,e what L icbersorr and I lansen's
correlat ional study showedabou t t hc r e l a t i onsh ip be tween
d i vc r s i t y , as measu red by Fo r rnu la A , anddcvcloprr
tcrr t .I le ablc to re cognizc thc general pat terns of communicat
iv i ty indices in Singaporean t l Wcs t Ma lays i a , anc l t l r
e i r s i gn i f i cance f o r t hc s t a tus o f t he na t i ona l
I anguageirr each coulr t ry.Bc ab le t o r ccogn i ze t hc conc
lus i on t ha t an ana l ys i s o f t he i nd i ces o f commun i
-ca t i v i t y f o r I l i nd i -U r< lu anc l Eng l i sh i n I
nd ia l eads t o .Bc ab le t o s t a te wha t Faso ld ' s C va lue
measu res .Bc ablc to rccognizc the s igni f icance of the fact
that C-values are higher at the statea r r< l r l i s t l i c t
l e vc l s i n I nc l i a t han t hey a re a t t he na t i ona l l
eve l .
I Q
70
7 l22
-
148 Language Attitudes
or indircct inferelccs. It is only necessary to observe,
tabulate, and analyzeovert bel rav ior . At t i tudcs of th is sor
t , however. would not be qui te asinteresting as they would be if
they were define d mentalistically, because theycannot be used to
predict other behavior (Agheyis i and Fishman
1970:138).Nevcrtheless, the more straightforward behaviorist
approach, in whichat t i tudcs arc just one k ind of response to a
st imulus, cer ta in ly can not be ru ledout .
Another issue that ar ises in the considerat ion of at t i tudes
is whether or notattitudes have idcntif iable subcomponents.
Generally speaking, social psycho-logis ts whg accept the behavior
is t def in i t ion v iew at t i tudes as s ingle uni ts .Mcntaf is
ts usual ly consider at t i tudes to have subparts , such as cogni
l ive(knowledge ) , a.ffective (feeling), an
-
150 Languoge Attitudes
attitude research, in either original or modihed form, is the
matched-guisetechnique developed by Wallace Lambert and his
associates (Lambert et al.1960; Lambcrt 1967). The pure
matched-guise technique aims at total controlof a l l var iables
except language. To achieve th is , a number of b i l
ingualspeakers fluent in the languages under investigation are
recruited. Thesespcakers arc tape-recorded reading exactly the same
passage, once in onelanguage ancl once in the other. The recorded
passages are arranged on atape-recording in such a way that it
appears that each passage has beenrecorded by a different speaker.
Ifspeakers have been recorded once in Frenchand once in English,
for example, the recording might begin with onespeaker's French
performance. The next voice heard might be anotherspcaker's English
guise. The third voice might be a third speaker speaking inEnglish.
Perhaps the fourth voice wil l be the first speaker in her English
guise.By this time, l isteners wil l presumably have forgotten the
voice quality of thefirst speaker and wil l take it that the fourth
speaker is someone they haven'theard before. The two guises ofthe
several speakers who have been recordedare all interspersed in this
way, so that it sounds to the l istener that eachpassage they hear
is f lrom a totally new speaker. In the end, then, they assumethat
thcy have heard twice as many people as they actually have
heard.
A sample of bil ingual l isteners from the same speech community
is thenaskcd to l isten to the recordings and rate the speakers on
various characteri-s t ics, such as in te l l igence, socia l c
lass, and l ikabi l i ty . I f the same person israted diffcrcntly
in dilferent'guises', it has to be the difference in language
thataccounts for it. Since the same person has provided both
samples, it cannot bevoice quality differences that the l isteners
are reacting to. Content iselirninated as a variable by having
translated versions of the same passage readin each language.
It turns out, o[ course, that there are several degrees of
directness andindircctness in language-attitude research, with
various aspects of thecxperimcnt being concealed lrom the subjects.
The matched-guise technique isclirect in the sense that the l
isteners are explicit ly asked to give their opinions oftlte
speaker' 's characteristics. It is indirect in the sense that l
isteners are askedto react to speakers, not languages, and they are
not aware that they arehcar ing the same person in each guise.
S eman t ic diffe ren t ia I s ca I e s
l-he format for l istener responses that is very often used with
the matched-guise technique involves semantic differential scales
(Osgood, Suci, andTannenbaum 1957). These scales designate opposi
te extremes of a t ra i t a teither end and leave a number of blank
spaces between them. A typicalsemant ic d i l ferent ia l scale i
tem appears in f igure 6. l . I f the speaker on tapesounds
extremely unfr iendly, the l is tener would p lace a mark on the l
ineclosest to the word'unfriendly'. If he sounds extremely
friendly, the mark goes
Language Attitudes l5l
at the other end ofthe scale. Ifthe speaker seemsjust average in
friendliness,the middle space should be used, and so on. Similar
scales are constructed forother characteristics. In this way, each
listener has the opportunity to indicatejust where on a scale of a
particular trait a speaker falls.
f r iendly - unfr iendly
Figure 6. I Typical seven-poinl semoilic diff erential scale
The following procedure is used in scoring
sernantic-differential scales.After the responses have been
collected, numbers are assigned to each of thespaces in the scale.
In a seven-point scale such as the one in figure 6. l, a 7 mightbe
assigned to the space nearest to the word 'fr iendly', a 6 to the
next space,and so on. As the responses are tabulated, a tick mark
is made on the blank ateach space on the scale for each listener
who placed his evaluation of t l,atspeaker at that space. For
example, suppose a speaker in a particular guise isjudged 7 for '
intell igence' by five l isteners, 6 by thirteen, 5 by twenty, 4
byeleven, 3 by two, and 2 by one. The init ial tabulation would
look l ike figure 6.2.
ill -rllf lll| Il+ti ,r\1+11\f ltttlltt {tflllf \\ r
i n te l l igen l - un in te l l igen t7 6 5 4 3 2 ' l
o't!":,i,!;l,,f,i;:::,':f :::;:t/,Il:li;'::,{"'Next the number
of marks at each space is multiplied by the value for thatspace.
The results are then totalled. In our example, the calculation
would be:
( 7 x 5 ) t ( 1 3 x 6 ) + ( 2 0 x 5 ) + ( l l x 4 ) + ( 2 x 3 )
+ ( l x 2 ): 35 + 78 + 100 + 44 + 6 * 2= 265 .
This value is d iv ided by the tota l number of l is teners (
the same as the tota lnumber of t ick marks in figure 6.2); in this
case, 52. The result is 5.10, the valuethat would be reported and
subjected to statistical analysis. This nurnber is themean
evaluation for this speaker on the intell igence scale. It is to
beinterpreted as indicating that, on the average, this speaker in
this guise isjudged to be in te l l igent to the degree of a b i t
more than 5 on a scven-pointscale.2
Other methodsAlthough matched-guise-cum-semantic-differential
research is sornething ofa standard in language-attitude research,
other direct methods are possible.Agheyisi and Fishrnan
(1970:147-50) mention three other techniques: ques-tionnaires,
interviews, and observation.
-
152 I-anguage Attilude.s
Ques t ionnaires. l'ftcse can Iave one of two types of ques
lions open or closed.Open quest ions g ivc the respondent maximum l
reedom to present her v iews'but also allow her to stray from the
subject and are very diff icult to score. Irr anopen-qucst ion
qucst ionnaire, respondents may be asked: 'Descr ibe yourreact ions
to th is spcaker 'af ter they have heard a taped sample. In a c
losed-qucst ion, the respondent is g iven a par t icu lar format to
use in recordingresponses. Apart from the semantic differential,
other closed-question[ormats involve yes-no answers, multiple
choice, or ranking schemes. Closedqucst ions are lnuch easier for
respondents to deal wi th and are easy to score,but they lorce
respondents to answer in the researcher's terms instead of
theirown. Perhaps the ideal compromise is to conduct p i lo t
research wi th openquestir>ns ancl usc tlrese results to
construct a closed-question questionnaire.We will sec late r how
such a proceclure was masterfully executed by FrederickWi l l
iams.
Intervieu,s. Intcrvicws are l ike open-question questionnaires
without thequestionnaire. A fieldworker personally asks attitude
questions and recordsthe rcsponscs in written (or tape-recorded)
form as the subject respondsorally. Tlre burde n of recording open
questions is removed from the subject,making i t easier to c l ic i
t opcn responses. and the in terv iewer can guide theconvcrsati
-
154 Language Attitudes
test va l id i ty (F ishman 1968d; Cooper and Fishman 1974:15) .
In h is 1968research, Fishnran not only asked questions involving
the attitudes of PuertoRicans in the New York City area about their
ethnicity, but also invited themto an evening of Puerto Rican
dances and other cultural activity. It was thenpossible to compare
the responses to the questionnaires with whether or not arespondent
answered the invitation, said she would attend, and actually did.
lfa person answered the questionnaire in a way that indicated pride
in beingPuerto Rican and then attended the evening's activit ies,
her attendance wouldbe an indication tlrat her answers to the
questionnaire were valid.
A simplcr dcsign is reported by Giles and Bourhis (1976). In
theircxperiment, two psychologists were introduced to a class of
high-schoolstudents. One o[ them to ld the students that there was
concern abouterroneous ideas people had about psychology. Using one
of two accent guises,hc asked them to wr i te down what they
thought psychology was about so thatstcps could be taken to correct
any misconceptions that might be discovered.After they had begun
writ ing, this psychologist left and did not return. Theother
member oI the team to ld the students that there was a p lan
beingconsidered to have someone address students in high schools
about psycho-logy as a way oI combating nrisconceptions.
Furthermore, she told them, hercolleague who lrad just left was
being considered for the job. When they hadlinished the open
question, they were given a rating-scale questionnaireconsisting of
closed questions. The same procedure was repeated with
anothernratched group of students, but this time the first
psychologist used anotheraccent guise. Of course, the primary
interest was in language attitudes to thetwo guises, not in the
students' concepts about psychology. A simplebelravioral index was
built into the experiment: the amount the students wrotein answer
to the open quest ions about the f ie ld of psychology and about
theexpcrimenter's qualif ications to lecture students could be
compared in each ofthc two guises. It turned out that the students
wrote more when they wereaddrcssed in one guise than when they were
addressed in the other, and thisresult rcirrforced what had been
found in earlier, more traditional matched-guise research.
Furtherrnore, the amount written in response to the two guiseswas
consistent with the attitudes reported on the closed-question
rating scales.Gi les and Bourhis 's exper iment tends to indicate
that conat ive at t i tudesrcvcaletl by the rnatched-guise
technique can be valid. Nevertheless, Agheyisiand Fishman
(1970:150) refer to ' the fami l iar problem of the low degree
ofconsistency between attitude measures and overt behavior', an
issue that hasbeen acldressed by social psychologists for some time
(for example, Linn 1965;Fis l rbein 1965). I t remains to be
determined just what sor t of at t i tudenreasuring instrunrents
are l ikely to be valid.
A final clift iculty with the ordinary matched-guise technique
is connectedwi th i ts ar t i f ic ia l i ty (Gi les and Boir rh is
1976; Bourhis ancl Gi les 1976). Askinglisteners to judge people by
their voices only, though it does provide ma,..imumcontrol over
other variables. is a bit far removed from real-l i fe contexts.
Since
Language Attitudes 155
the pure matched-guise procedure requires that every taped
sample has thesame content, it could easily be the case that l
isteners, after becorning boredwith the repetit ion, might begin to
pay more than normal attention to vocalvariations. Finally, judges
in a matched-guise experiment are provided withan evaluative set.
They are given a rating sheet and told to make judgementsabout the
people they are going to hear . As a resul t , they are set up to
makeevaluat ive judgements in a way that doesn' t happen in ord
inary in teract ivesettings. The experiment about the two
psychologists and the high-schoolstudents was designed to overcome
these objections to sonle extent. Thespeaker was present at the
time of the experiment, rather than being adisembodied voice on a
tape-recording. 1'here was no repeated content and anevaluative set
did not intrude unti l after the subjects had had the chance
towrite what they wanted to say about psychology.
Bourhis and Giles (1976) devised an evelr more ingenious and
naturalisticmatched-guise experiment, in which the subjects l.rad
no idea that they hadpart ic ipated in Ianguage at t i tude
research. The guises involved in th isexper iment were four
language var iet ies re levant to the socio l inguist ics i tuat
ion in wales. These were: ( l ) the h ighest-s tatus pronunciat ion
pat ternfor Br i t ish Engl ish, ca l led RP ( ' received
pronunciat ion ' ) ; (Z) Engl ish wi th arn i ld South welsh accent
; (3) Engl is l r wi th a broad South welsh accent ; ancl(4) s
tandard Welsh. The'subjects 'were theatergoers, and the exper iment
tookthe form ol-a public-address announcement made during the
intermission at aperformance. As you have probably guessed, there
were four forms of themessage, one for each guise. T'he content was
a request for patrons to obtain aquestionnaire form in the foyer to
fi l l out in order to lrelp the theater planfuture programs. Two
kinds of audience 'participated' in the experimcnt. AnAnglo-welsh
audience consisted of those people who at tencled the theaterdur
ing f ive evenir rgs on which two f i lms in Engl is t r were
presented. I t wasassumed that most of these audiences were Welsh
people who spoke onlyEngl ish. The b i l ingual welsh audience were
those who at tended dur ing fourevenings when a p lay in welsh was
presented. The Anglo-welsh audiencesheard the announcement in the
three k inds of Engl ish, wi th one var ietypresentecl on a given
evening (the RP and broacl-accent English versions werepresented on
two evenings each) . The b i l ingual welsh audicnces heard a l l
fourguises, one per evening. l 'he behavior that was measured was
the number ofquestionnaires submitted by menrbers of each of the
two kintls o[ aucjienceswhen the request was made in the various
language guises.
Since the data are nominal (quest ionnaires e i the r subrni t
tec l or notsubmit ted) , the chi square test was used in the
analys is . Tables 6. I and 6.2 anclfigure 6.3 show the results.
The chi square test f lor the cJata in table 6. I allowsthe re ject
ion o[ the nul l hypothesis that quest ionnaire conrpleters and
non-completers are randomly d is t r ibuted among the au
-
156 Language Attitudes
that is not due to chance, but i t is easy to see what that
'something ' is . TheAnglo-Welsh populat ion is equal ly responsive
to the message in RP andmilclly Welsh-accentecl English, but far
/ess responsive whert the nressage is rnbroadly Welsh-accetrted
E,nglish. The results for the bil ingual Welsh audiencesare quite
diffe rcnt.
Table 6. I Behavioral results to matched-guise experimentfor
Anglo-Wel sh audiences
GuiseQue.stionnaireconpleted bEW mEl l 'RP
Languoge Attitudes 157
difference between responses to mildly Welsh-accented English
and Welsh ispresented in table 6.2. The nul l hypothesis , that
table 6.2 represents randomlydist r ibuted data, can be re jected-
I t is c lear that s igni f icant ly more nrembers ofa b i l ingual
Welsh audience wi l l complete the quest ionnaire when asked to
doso in Welsh tlran when asked in English, even if i t is
Welsh-accented.
Apart f rom the resul ts , the Bourhis and Gi les exper iment
shows how theusual objections to matched-guise research based on
artif iciality can beovercome. Although no speaker was physically
present, the context was atotally normal one; undoubtedly the
subjects had no idea that their languageattitudes had been
investigated. No one audience had to l isten to the samecontent
more than once, and no evaluat ive set was int roduced. The
measurewas oIunmoni tored behavior , and not of scaled evaluat ions
of speakers. Onthe other hand, some control is lost by not having
the same subjects respond toall the guises theater audiences heard
the request in different guises). The samewas true of the
experiment with the high-school students. It is assumed thateach
set of subjects is the equivalent of every other set of its type,
for thepurposes o[ the attitudes being measured. That is, each
evening's Anglo-Welsh audience, for example, is taken to have the
same predilection torespond to the various language guises as any
other evening's Anglo-Welshaudience. There seems to be no reason to
doubt that this is a safe assurnption,in the experiments I have
described. But if artif iciality is dinrinished at the cost
RP bEW mEW WGuises
Figure 6.3 Bar graph of behavioral results to
motched-guiseexperimen! for bilingual llelsh audiences.
Source: adapted from Bourhis and Giles (1976:15, figure 2)
Ycs
No
i l822.5Vo
40677.5To
538.25Vo
5999l.8Vo
6625.07o
19875.0Vo
12-60.9; ;r(0,00 IRP=receivcd pronunciat ionbEW:broadl y
Welsh-accentcd Engl ishm EW=mild ly Wels l r -accented Engl
ishSorrrce: data f rom Bourhis anr l Gi les 1976: l4
I-able 6.2 Behavioral results to matched-guise experimentfor
bilingual Welsh audiences
GuiseQuestionnairetornpleted mEllt
Ycs
No9.2Vo
r0926.5V048
c0oEEoo
o, :aCc
-
o)o
oooaEo
oC
W
x'?= 8.201 2
-
158 Language Attitudes
of using more than one sample of a population, the experimenter
should becareful that all the samples are equivalent and
representative' Otherwise,uncontrolled variablcs can
unintentionally be introduced'
APPLICATIONS
Social slruclure
The study of language at t i tudes is inst ruct ive in i ts own
r ight , but i t is evenmore valuable as a tool in i l luminat ing
the socia l importance of language- Inthis section, we are going to
examine language attitudes as one way o[understanding how language
is used as a symbol of group membership. Wewill also see lrow
attitudes serve as a clue for identifying diglossic
relation-ships.
Group identity. In chapter l, we saw that language served
unifying andseparatist functions for sociocultural groups. You
might think that attitudestudies woulcl show that any language
serving these two functions would behighly evaluated by i ts
speakers. I t turns out that th is is a l i t t le too s
imple.Where a society has l inguistic varieties in diglossic
relationship, the usualattitude is that the High language is a
purer and better language than the Lowlanguage. Of course, the
unifying and separatist functions are most l ikely to befulfi l led
by the Lou, language. If High language varieties are generally
morehighly valued than Low varieties, you might be led to exactly
the oppositeconclusion; a l inguistic symbol o[contrastive
self-identif ication is l ikely to bepoorly evaluated by its
speakers.
Perhaps there are some k inds of quest ions that would e l ic i
t h ighercvaluat ions for Low var iet ies. I f so, then they could
be bui l t in to a matched-guise experiment. If the associated
semantic differentiat scales included bothaffective ancl status
items, it might be predicted that speakers using the Highguise
would be rated higher on the status scales, but the same speaker
wouldbe rated higher on the affective scales in the Low guise. A
speaker using ahigher-status guise might be rated higher in intell
igence, say, or occupationalstatus, but would get higher ratings
for friendliness and trustworthiness in hisin h is lower-status l
inguist ic guise. Researchers who have tested th is hypothe-s is ,
however, have obta ined unexpected resul ts . For example, d 'Angle
jan andTucker, conducting attitude research involving European
French and twovarieties of Canadian French in Quebec, found that
the European Frenchspeaker was rated not only more intell igent and
better educated, but morel ikcable than Canadian French speakers;
the l ikeabi l i ty resul t was unexpected(d 'Angle jan and Tucker
1973:22) . Simi lar ly , the resul ts of at t i tude researchwi th
Mexican Amer icans conducted by Carranza and Ryan (1975:99)
werethat both Anglo and Mexican Americans ratcd English higher than
Spanishon status scales, as expected, but also on solidarity
scales, an unexpected
Language Attitudes 159
outcome for the Mexican Americans. Attitudes of black Americans
towardsstandard English and Vernacular Black English must be
closely analyzed inorder to discover indications of positive
attitudes toward the vernacular(Hoover 1978). El-Dash and Tucker
(1975:46) cite tlrree studies, includingtheir own, in which
superposed language varieties were preferred over mothertongues,
according to the results of attitude research.
On the other hand, attitude research proved a valuable tool in
the analysisof Albanian as a possible language of group identity
among a socioculturalgroup in Greece called the Arvanites
(ar-van-ee-tess). Using a closed-questionquestionnaire rather than
the matched-guise technique, Trudgil l and Tzavaras(197'7) were
able to trace the declining status of Arvanitika (the
Albaniandialect) as a language ofgroup identity. The responses they
obtained revealeda clear pattern according to age, as table 6.3
shows.
Table 6.3 Percentage responses to these questions, by
age-group:'Do you like to speak Arvanitilca?','Do you think
speaking
Arvanitika is a good thing to do?', ' ls it anadvontage to speak
Arvanitika?'
Indifferent NoYes
AgeLike to Good Advan-speak? thing? toge?
L i k e t ospeak?
Advan-toge?
Goodthing!
Like to Good Advan-speak? thing? tage?
5-9t0 - l 4t5-2425-34J5-4950-5960+
t l r16 t'l 12l 7 1 8 3 4l0 J2 4046 64 5667 95 6719 86 97
t 0t'l4 l6753J Ia lL I
l0 t'l'13 3848 5066 4935 364 3 3
1 4 3
89 89 8267 l0 5042 35 16
3 3 l l2 t 8l t 00 0 0
Source: data from Trudgill and Tzavaras ( 1977)
Table 6.4 Percentage ansu,ering'Yes' and'No' to the question:'Is
it necessary to speak Arvonitika to be an Arvonitis?', by
age-group
Age Yes No
l0-14t5-2425-3435-4950-5960+
J J
5816726183
61422428J J
t 7
Source: d,ala from Trudgill and Tzavaras (1977)
The answers to the question 'Is necessary to speak Arvanitika to
be anArvanitis?'seem at f irst glance to be inconsistent with the
general pattern of
-
160 Language Attitudes
table 6.3. The answers to th is quest ion are tabulated in table
6.4. A major i ty inevcry age group thinks that it is no' necessary
to speak Arvanitik-a to be anAruani t is , cXcept the youngesr . As
Trudgi l l and Tzavaras (1977:180_l8 l )cxplain this rcsult, most
members of the older age-groups realize thatAivanitika is dying
out, but hope that their ethnic identity can be preservednone the
less.a As a resul t , they are obl igecl to make room for Arvani
tes whotlon't speak Arvanitika. The youngest speakers, far from
being more hopefulabout ihe future of Arvani t ika, seern to
foresee the demise of both the
language and the ethnic d is t inct iveness. Their at t i tude
seems to be that i t tsn" . " r r " .y to speak Arvani t ika to be
an Arvani t is , there are fewer Arvani t ika,p.ake^ thari therc
otrce were and therefore fewer Arvauites, but this is nopart icu
lar cause for concern. As Trudgi l t and Tzavaras (1977: l8 l ) put
i t :They are prepared to concedc that it is necessary to speak
Arvanitika to be an Arvanitisbecause, although they are aware that
Arvanitika is dying out, they do not regard theloss of t lre
languige or of the ethnic identity as undesirable. lt seems' that
is, that withyoungcr peopie cJnfl icts ofthis type are being
resolved in favour ofa Greek identityand the Greek language
Trudgi l l and Tzavaras 's c l i rect open-quest ion at t i tude
quest ionnaireappcars to give a more accurate picture of the
function of a language as atliniicator of gro,,p identity ttran
more sophisticated matched-guise researclt
-
162 Languoge Attitudes
havc becn i l ' the l is teners were only guessing (p
-
l(t4 Language AttitudesOf those who heard and recognized a shift
to informal Canadian French,67
per ccnt said thcy would nol make such a shift themselves. The
majorityopin ion that ernerged under d i rect quest ioning seems to
be that one ought toenrphasize one's Quebecois ident i ty by res is
t ing any inc l inat ion to shi f t toEuropean Frerrch to
accommodate a European speaker, but not to overdo i tby d iverg ing
in to in formal s ty le. More indi rect ev idence of th is , in the
form ofh igher rat ings on a l fect ive scales for someone who
actual ly behaved that way,d ic l not show up. I f i t is t rue t l
ra t matched-guise research (and th is exper i rnentis a sort oI
matched-guise procedure) taps more covert attitudes than
directquestioning does, it could be that there is a lag between
overt and covertattitucles. l-hat is, on a more surface level, the
growing independencemovement in Que bec has led to an emphasis on
language as a symbol of groupident i ty , but the o ld ideas about
European French as a 'bet ter ' d ia lect , pers is tat deeper
levels-
The sarne researc l . r team conducted a s imi lar exper iment
in Wales, wi th RPl l r i t ish Engl ish replac ing European
French, mi ld ly Welsh-accented Engl ishcorresponding to formal
Canadian French, and broadly Welsh-accentedEnglish in the place of
informal Canadian French. The subjects were pupils ata secondary
school in South Wales. The same format , involv ing a supposedath
lete and two spor ts commentators, one speaking RP and one
speakingE,nglish with a mild Welsh accent, was used. As in Quebec,
the speaker wasrated signil icantly more intell igent when he
shifted to RP during the interviewwith the RP-speaking interviewer.
Unlike the results in the Quebec srudy,thcrc were signil icant main
effects on two of the affective scales. The speakerwas judgetl more
trustworthy and more kind-hearted when he maintained anr i lc l
Welsh accent when ta lk ing to the RP-speaking commentator than
whenhc shiltecl to IlP, and even higher on both scales when he
shifted to a broadWelsh accent . I t appears that Welsh-accented
Engl ish is bet ter-developed as asyrnbol of group ident i ty in
Wales than Canadian French is in Quebec, s incethc expected rcact
ions in a matched-guise exper iment show up to somedcgrec.ro
At t i tuc lc research sheds l ight on the status of language
var iet ies asinclicators of group identity, but not always in the
nrost straightforward way.In par t icu lar , the c lass ic
matched-guise exper iment , in which l is teners areasked tojuclge
speakers one by one on the basis oftape-recorded monologues,of ten
does not lead to predicted resul ts . On the other hand, the Trudgi
l l andTzavaras results and tlre direct-question part of the
Bourhis, Giles andLarnbert research in Quebec appear to g ive resul
ts that are easier to in terpret ,but may tap only sur face react
ions. More sophist icated appl icat ions of thematched-guise
technique, such as the one used by Bourhis and his associates
inQuebec and wales, show promise of probing covert attitudes more
accuratelythan the s impler appl icat ions do.r I
Diglossia. In Ferguson's original description of diglossia,
community
Language Attitudes 165
at t i tudes toward the High and Low var iet ies were an
important component .High varieties have greater prestige and Low
varieties are often disparagecl.At t i tude-study resul ts ought to
fa l l in to predictable pat terns in communir ieswhere d ig loss
ia ex is ts . I t turns out that they do to a very considerable
extent .As examples of this, we wil l look at two attitude studies
conducted in societieswi th d ig loss ia. One was carr ied out in
Egypt (El -Dash and Tucker 1975) anclinvolves Classical and
Colloquial Arabic (one of the four example cases usedby Ferguson).
Although diglossia is not the focus of El-Dash and Tucker'swork,
their results reflect diglossia quite clearly. The second study
involvesSpanish and English in a Mexican American community
(Carranza and Ryan1975). Here we have an example of broad d ig loss
ia, s ince two d is t inctlanguages are involved. Unlike El-Dash
and Tucker, Carranza and Ryanexplicit ly set out to find attitude
patterns typical of diglossia.
El-Dash and Tucker employed a classic matched-guise model, with
thecontent controlled by having speakers discuss the same topic
(the Gizapyramids, an emot ional ly neutra l subject) rather than
having them read astandard passage. Two Egyptian speakers were
selected who could speak bothClassical ancl Col loquia l Arabic,
and Engl ish. Qui te natura l ly , they had anEgyptian accent when
they spoke English.t2 It was not easy to find speakerswho could
assume both Classical and Colloquial Arabic guises. The reasonsthat
El -Dash and Tucker g ive for th is won' t surpr ise anyone who is
fami l iarwi th d ig loss ia:
The dist inct ion between Classical and Col loquial Arabic is
not completely c lear,however, as var ious gradat ions exist
between the Arabic of the Koran and thc speecl . rused by the man
in the street to d iscuss dai ly af fa i rs. Morcovcr, Classical
Arabic . . . isnot a spoken language, but rather a wr i t ten form
used throughout the Arabic-speakingwor ld. This form may be read
oral ly , but is seldom spoken extemporaneously. I t isreported
that very few indiv iduals can actual ly speak Classical Arabic f
luent ly . (El-Dash and Tucke r 1975 :35 )The speakers they used
were self-conscious about attempting to speakClassical Arabic
spontaneously. The very task of tape-recording their spcechhad an
effect on their Colloquial Arabic performance as well; they seemed
touse a slightly 'elevated' style of the colloquial dialect
(El-Dash and Tucker1975:53). In spite of these problems, the
research results give a clear picture ofa sor t of t r ig loss ia s
i tuat ion, wi th Egypt ian Engl ish occupying a posi t ionbetween
the Classical and Colloquial forms of Arabic. Listeners were asked
toevaluate the speakers in all the language guises on four
semantic-differentialscales, for intell igence, leadership,
religiousness, and likabil ity^rr The resultswere subjected to
analysis oI variance. A significant main effect for languagevariety
was found for all four traits.
Unfortunately, the discovery of a language main effect is not
very helpful.To see why this is so, consider table 6.6, a slightly
simplif ied reproduction ofone line of El-Dash and Tucker's
(1975:42) table 2. The significant ,F-ratiosimply tells us that
somewhere among the six language guises, speakers of
-
166 Language Attitudes
some guises are judged significantly more intell igent than
speakers in otherguises. It does not tell us in which guises the
differences are significant.Fortunately, there is another
statistical test, called the Newman-Keulsmultiple-comparison test,
which is designed to identify the source of variationin cases such
as this. El-Dash and Tucker applied this test and the results
aresummarized in table 6.7.
An examination of table 6.7 shows that speakers of Classical
Arabic andEgyptian English are rated as more intell igent and as
having greaterleadership abil ity than speakers of Colloquial
Arabic. This is consistent withthe diglossia pattern; the average
person in the society would think thatHigh-dialect speakers have
these traits to a greater degree than Low-dialectspeakers. Although
Classical Arabic received higher ratings on both traitsthan
Egyptian English did, the difference was not significant in either
case.For religiousness, since religious use is generally a
High-variety function,speakers of Classical Arabic are rated
significantly higher than speakers ofeach o[ the other two guises.
However, the fact that English is not indigenousto Arab culture is
important here. Although English in general shows up as aHigher
language, this is not relevant where religion is concerned.
ColloquialArabic speakers are judged slightly higher for
religiousness than EgyptianEnglish speakers are, but not
significantly so.'a The one affective scale,' l ikeabil ity', fails
to show an advantage for the Low language, ColloquialArabic. On the
contrary, the judges found speakers of Classical Arabics igni l
icant ly rnorc l ikeable than speakers of e i ther of the other two
var iet ies.The expectation that the 'homier' language might rate
higher on an affectivescale is again not fult i l led.
7-able 6.6 Mean ratings for intelligence, according to guise
Language Att i twles l6jTable 6.7 Comparison of the mean
evaluations of four rraits
for speakers in three guises
TroitClassicalArabic
EgyptianEnglish
ColloquialArabic
Intel l igence
Leadership
Religiousness
Likeabil i ty
t0 .2510.25
8.748.74
9.389 .38
9.489.48
9.59
8.56
7 . 1 I
8 . 7 1
8.498^49
1.207.20
7.751.15
8 . 5 18 .5 t
* indicates that the differences between the scores is
significant at p(0.01.f indicates that the difference berween the
scores is significant at p(0.05.Where t l rere is no symbol midway
between the guises ( for example, for inte l l igence,
betwcenClassical Arabic and Egypt ian Engl ish) , the di f fercnce
between the scores is not s igni f icant .Source: data from El-Dash
anil Tucker fl975)
Table 6.8 Contparison of the mean evaluations of threeguises for
suitobility in Jbur situotions
SituationClassicolArabic
EgyptianEnglish
ColloquialArabic
T'ruitSrgni-Jicance
6.366.36
9 . l l9 . 1 I
9 .799.79
9 . l 39 . 1 3
6.69
8 . l 6
8 .84
6.60
8.708.70
8. r9B. l 9
7 .187. tB
Clossical Colloquiol Egyptian AmericanArobic Arabic English
English
BritislrEnglish F-ratio
At home
At school
On radio andtelevis ion
Irormal speeches
Intel l igence 1O.25 8.49 9.59 9 . r 3 8.45 20.3) p(0.01Sozrcc.
t lata f rom El-Dash and Tucker (1975:42, table 2)El-Dash and
Tucker a lso asked a number of quest ions about the sui tabi l i
ty
of the differcnt language varieties for use in several
situations. The answerswere given in a semantic-differential scale
lormat, ranging from high to lowsuitabil ity. The possible scores
ranged, as in the matched-guise part of theexperiment, from 2 to
12. The subjects were asked to judge the suitabil ity o[cach guise,
as they heard it, for use: ( l) at home; (2) at school; (3) at
work; (4)on radio and television; and (5) for formal and re l
igious speeches. There wereno significant differences among the
three guises for the 'at work' situation,but the other four
situations gave exactly the results you would predict from
aknowledge of diglossia. The results for these four situations are
given in table6.8, using the same format as in table 6.7.
* indicates that the di f ference between the scores is s igni f
icant at p(0.001.t indicates thar the difference between the scores
is significant at p(0.05.where there is no symbol midway between
the guises ( for example, at school , between Egypt ianEngl ish and
Col loquial Arabic) , the di f ference between the scores is not s
ieni f icant .Source: d,ala from El-Dash and Tucker ( 1975)
colloquial Arabic is by far the most suitable variety for use at
home, andclassical Arabic is least suitable, ranking significantly
lower (at p(0.05) thanEgyptian English. At school, on the other
hand, classical Arabic is the most
5.39* 5 .39
-
l(rlJ I.anguagc Attitudassui tablc var iety , wi th no d i l
lerence between Egypt ian Engl ish and Col loquia lArabic. I ror
use on radio and te lev is ion and in formal speeches, there is
athrec-way d is t inct ion. Classical Arabic is s igni f icant ly
more sui table thanei t l rer of the other two, and Egypt ian Engl
ish is s igni f icant ly more sui tablethan Col loquia l
Arabic.
The two parts of El-Dash and Tucker's research design combine to
paint arather clear picture of diglossia being reflected in
community languageat t i tuc les. Speakers of Classical Arabic and
Egypt ian Engl ish are seen ashaving greater leadership abil ity
and higher intell igence than speakers ofCol loquia l Arabic, and
thei r way of speaking is perceived as most su i table loruse on
radio and television and in formal and religious speeches.
ClassicalArabic is s igni f icant ly more sui table for use in the
same two s i tuat ions thanEgyptian English is, but there is no
significant difference between the two forin te l l igence and
leadership abi l i ty . Classical Arabic speakers are judged
morereligious than speakers of either of the other two language
varieties and it isconsic lered more sui table for use at school .
For use at home, Col loquia lArabic is by lar the most suitable,
and Classical Arabic is judged the leastsui tablc .
The only result that doesn't seem to come out the way we might
expect is thejudgements on l ikeabi l i ty . Speakers ot the I {
igh d ia lect , Classical Arabic, werejudgccl rnost l ikeable,
whereas we might have expected speakers oICol loquia lArabic to gct
th is cvaluat ion. But E, l -Dash and Tucker 's s tudy is not
unusualin this rcspcct; t lte superiority o[ prestige languages on
affectivc scales is a[a i r ly common resul t in at t i tude
studies.
1-he study of attitudes about Spanish and English in a Mexican
Americancornruuni ty in Chicago (Carranza and Ryan 1975) was
designed expl ic i t ly wi thdiglossia in tnind. Two specially
developed sets oIsemantic-differential scaleswere used: a
stdlus-stress[rg set and a solidarity-stressing set. The designinc
luded the mir ror- image feature in the st imulus tapes. (Recal l
that a mir ror-image design nrcans that each l inguist ic guise in
the exper iment wi l l appear incach of two or rnore situations.)
Carranza and Ryan's application involvedlour paragraphs: ( l ) a
narrat ive about a mother prepar ing breakfast in thekitclren in
Spanish; (2) a paragraph on the same topic matched for style
inIlnglislr; (3) a paragraph about a teacher giving a history
lesson in school inI rngl ish; arrd (4) a matched paragraph in
Spanish. This gave four passages:Engl ish-FIonre, Spanish-Home,
Engl ish-School and Spanish-School . Thestatus-stressing scales
were scales for education, intell igence, successfulness,ancl
wealth. TIre four solidarity-stressing scales were for
friendliness, good-ness, kindness, and trustworthiness. The
stimulus tape contained [ourreadings of each o[ thc four passages
by a different freshman student at theUniversity of Notre Dame.
Each reader was a native speaker of the languageshe was reading. As
a resul t , the tape conta ined s ix teen readings by s ix teendi f
ferent speakers. r5
'lhe subjects were anglo and Mexican American high-school
students at a
Language Attitudes 169
cathol ic h igh school in chicago. The Mexican Amer ican
students were nat ivespeakers of Spanish and studied and used
English in school. 1'he anglostudents were native speakers of
English, but had studied Spanish in highschool.r6 The Mexican
American students, of course, were the only ones whowere members of
the diglossic English-Spanish conrmunity; the anglosubjects were a
sort ofcontrol group. Carranza and Ryan began the researchwith the
following four hypotheses (Carranza and Ryan 1975:89).I Mexican
American ratings wil l be higher for Spanish in the home
domain,
but higher for English in the school domain.2 Anglo ratings wil
l be more favorable for English than Spanislr in botlr
domains.3 Mexican American ratings wil l be higher for Spanish
on the solidarity
scales but higher for English on status scales.4 Anglo ratings
wil l not dif lfer for the two scale types.
The results were analyzed by a four-factor analysis of variance:
group(anglo or Mexican American) x scale type (status scales or
solidarity scales) xcontext (home or school) x language (English or
Spanish). There wcre nogroup effects at all, either as main effects
or interactions, which simply meansthat the anglos were marking
their scales no differently, as a group, t lran theMexican
Americans were. This is a surprising result, but might possibly
beexplained iIwe assume that the anglo speakers were takingthe
point of vicw ofSpanish speakers. This is plausible, if several
facts are considered.
First, the resea.rch was carried out in mixed classes, with
anglo and MexicanAmerican students l istening to the tapes and
marking the scales at the sametime. Second, the anglo students were
all students of Spanish and werelearning something about the
language and culture of Spanish speakers.Perhaps most important,
most of the anglos no doubt realized that thecomparison of Spanish
and English in context was not relevant to them. Therecould be no
question of Spanish being appropriate in /lreir homes. The onlyway
to make sense of the procedure would be to adopt t l.re perspective
oI theirMexican American classmates, which they seem to have done
very success-fully.
On the other hand, lhere were other significant effects. There
was asignificant main effect for language (F= 9. 10, p(0.01), with
English having amean rating on all scales of 4.77 and Spanish of
4.62 (on seven-point scales,with favorable adjectives assigned the
value 7). Overall, then, both theMexican Americans and the anglos,
possibly taking the Mexican Americanperspective, evaluated English
more highly than Spanish, a typical result [orHigh languages in
diglossia. Much more interesting were two of the
significantinteraction effects, for context by language (F= 14.46,
p{0.001) and scale typeby language (lF : 6.80, p(0.05). This means,
for the context by languageinteraction, that English and Spanish
were rated significantly differently,depending on whether they were
heard in the home context or in the school
-
110 Language Attitudes
context. Sirnilarly, the ratings for the two languages were
significantlydiffcrent, depencling on whether a solidarity or a
status Scale was beingnrarkctl.
' I 'hc nrean ratings showing these interactions are displayed
in table 6.9. By
context, wc can see that Spanish is rated higher than English,
on the average'in the honre context, and English is higher in the
school context. This isprecisely the sort of result one would
expect in a diglossic community. AsCarranza ancl l{yan ( 1975:99)
point out: 'The results confirm the significanceof the rnanipulat
ion of the context . I f context were to be ignored, the resul
tswould have inclicatecl only an overall preference for English.'
The results forscale type are not qui te so c lear , in the sense
that Engl ish is rated h igher onboth scalcs. l 'he clifference on
the solidarity scales, however, is very small; i t isrnore substant
ia l on the status scales. The fact that there is a s igni f icanti
r r tcract ion nreans that th is pat tern is meaningfu l . In othe
r words, on balance,tlre subjects seem to recognize the distinction
between the two languages withrcspect to the conccpts o[ s tatus
and sol idar i ty . The d is t inct ion is in thedirection that
would be predicted from an understanding of diglossia.
Table 6.9 Mean ratings for Spanish ond Englishb), context and by
sclla lyps
Context Sc'ale type
IIome School Solidarity Status
Language Attitudes l7l
discussed. Here, we wil l examine how the attitudes of teachers
are measuredand what sort of results emerge.
Of all the social psychologists, educators, and psycholinguists
who havedone research on language attitudes in education, none has
done better workthan Frederick wil l iams. I7 Based on a number of
research projects conductedin the mid- 1970s (Will iams, Whitehead,
and Miller t97 l, 1972 Will iams 1973;Shuy and Will iams 1973; Will
iams 1974; Will iams and associates 1976),Will iams's painstaking
research design and statistical procedures lrave beenrewarded with
particularly clear and replicable results. It is worth our while
toreview his methods and results in some detail.
Williams's research design. Williams's fundamental method is to
havesubjects evaluate recorded speech samples (audiotaped and
videotaped)according to the familiar semantic-differential scale
format. His stimulusrecordings do not use the matched-guise
technique, except in the very loosestsense. There is no attempt to
have the same speaker speak in more than oneguise, but there are
samples of different children of the same age, who arematched for
ethnicity and social status - the two independent variablesWill
iams is most interested in. The recordings involve relatively
freeconversation, and there is no attempt to control content.
Although someresearchers would find the lack of control
distressing, the kind of recordingsWill iams uses avoid the repetit
ion and lack of spontaneity that sometimesplague more tightly
controlled research.
In Will iams's research, very special care went into the
developmcnt of thesemantic differential scales themselves. Whereas
rnany attitude researchdesigns simply use scafe-defining adjectives
that seem reasonable to theresearcher, Will iams (197 :T) went
through a four-step procedure lcading upto the main research
effort. First, a pilot study was conducted in which a smallgroup of
teachers l istened to speech samples and evaluated speakers iu
anopen question format in which they were asked to describe the
speakers intheir own terms. Second, a variety of adjectives was
taken from these open-question responses and put into a set of
prototype scales. This proceclure gavesome assurance that the
scale-defining adjectives were appropriate toteachers, since
teachers had spontaneously used them irr describing howchildren
talk. Third, these prototype scales were used by another group
ofrespondents to evaluate language samples. Fourtb, a statistical
techniquecalledfactor onalysis was used to find out whether the
various scales revealedmore basic response dimensions.
The Iast step requires some explanation of factor analysis.
Factor analysis isa complex procedure, so I won't try to explain
exactly how it works. Instead,we wil l just look at the kind of
results it gives. The basic idea is this: somerimestwo scales
measure much the same thing. Imagine a
semantic-differentialquestionnaire with the following two
scales:
Er tg l i sh
Span i sh
Dif l 'ercrrce
4.604.'73
- 0 . t 3
4.944 .5 t0.43
4.824.T1
0.05
4.724.410.25
A ncgat ive d i f fe rence ind ica tes tha t the d i f fe rence
is in favor o f Span ish ..So i l rcer da ia f rom Car ranza and
Ryan (1975)
l 'he resul ts in Egypt and in a Mexican Amer ican communi ty in
the Uni tedStatcs indicate that d ig loss ia is t ransparent to
conf i rmat ion by matched-guiselanguage-at t i tuc lc
research.
Education
Perhaps thc nrost usefu l appl icat ion for language-at t i tude
research beyond theunclerstanding of socia l s t ructure is in
educat ion. At t i tude studies conductedin cducation have been
oftwo types: ( I ) language attitudes of teachers; and (2)languagc
attitudes of second-lang,uage learners. The second type of study
isusually set up to find out if learners' attitudes toward the
language they arelearn ing af fect thei r progress. We wi l l not
take up th is k ind o[ research here,but sorne references are g
iven in the b ib l iographical notes to th is chapter . Involumc I
I , the i rnpor tance of teacl rer at t i tudes in general educat
ion is
-
172 l-anguage Attitudes
1'he speaker is:I in te l l igent - uninte l l igent2 br ight -
dul l
Probably t l re same speakers who rank h igh on the ' in te l l
igence' scale wouldalso rank h igh on the 'br ightness ' scale, and
v ice versa. That is , you wouldexpcct that a speaker who seemed to
be a 6 on the in te l l igence scale would a lsobe a 6 on the
brightness scale, or in any case a 7 or a 5. Iractor analysis is
aproccdure that te l ls the researcher to what extent an expectat
ion l ike th is iscorrect . I fa quest ionnaire were used that
conta ined the two scales above, andour guess is correct, the two
items would receive a high ' loading' in factorarralysis. (For our
purposes, ' loading' simply refers to a relatively largenumber.
)
The example we used is del iberate ly ar t i f ic ia l to make
the point c lear .Probably no one would use too such
closely-related adjectives, unless theresearcher wanted to use each
to check the validity of the other.rt Further-more, lactor analysis
can reveal relatedness amonB a set of several items, notjust two.
For example, in Wi l l iams's research, the fo l lowing set of
scalesformed a 'cluster', according to the results o[ factor
analysis: unsure-confiderrt, active-passive, hesitant-eager, l
ike-dislike talking. ln other words,teachers had pret ty much the
same qual i ty in mind when they marked a i ry oneoI these
scales.
The c lustcr of scales I have just l is ted was one o[ only two
t l ]a t Wi l l iamsdiscovcrcd. l-abelled confidence-eagerness,
these scales measured teachers'global attitucles basecl on a
child's f luency and enthusiasm. The second clusterwas called
ethnicity-nonstondardness. l-hese scales seemed to tap the
teachers'attitucles toward speech features associated with lower
versus higher socialstatus and wlrite versus non-white ethnicit
ies. Scales associated with thesetlinrerrsions were then
established as the appropriate ones to use.
Two dimensions may seem to be very few for captur ing the ent i
re array o[teacl rer at t i t t rdes; how conf ident can we be that
they are adequate? Wi l l iams(1974:24) points out a number of
indications that the two dimensions are bothvalid and reliable.
Validity is indicated in two ways. First, features found inthe
actual speech samples could be used to 'predict' the results on the
twokinds of scales, in a s tat is t ica l sense. Using a procedure
known as l inearregression analysis, Will iams discovered that the
frequency of variousnonstandard grammatical and phonological
features - such as a d sound forvoiced ,lr or clause fragments -
could predict ratings on the ethnicity-nonstandardness scales to a
large degree. In other words, a chi ld who used alot oI
pronunciations like dem and dose and a lot of clause fragments
would belikely to get high scores for being'ethnic'(in the context
of Will iams's researchthis meant black or Mexican American) and
'nonstandard'. Similarly, thefrequency of 'hesitation phenomena' l
ike uh and uhm was successful inpredicting the results on the
confidence-eagerness scale. The second reason
Language Attitudes li3for accepting the validity of the
two-dimensional model is best explained bywi l l iams h i rnsel f
(1974:24) : 'The two factor model was found in a lmostexactly the
same interpretable dimensions in separate Chicago and rexasstudies,
where scale development in both began with selection of
acljcctivesfrom teachers ' d iscussions. ' In other words, two p i
lo t s tudies in separatecommuni t ies, involv ing teachers who d
id not know each other , gave the sametwo-factor result. Reliabil
ity was established in a technical statistical sense bythe
calculation of 'reliabil i ty coefficients'(wil l iams 1974:24).In
aclclir iol, oneindication of reliabil i ty emerged from those
experiments that involveclrepeated evaluations oIthe same speech
samples by the same teachers. Evenwhen the ratings were done almost
three weeks apart, the results wereessentially identical.
In a less technical sense, the two dimensions have considerable
plausibil i tywhen we try to discover what they mean in real l i
fe. It appears that the teachersare respondingtotheform in which
the children express themselves (ethnicity-nonstandardness), and
the manner in which they deliver their talk (confidence-eagerness).
Furthermore, the youngsters received high and low ratings on
thescales in an easily interpretable pattern, as figure 6.4 shows.
Notice thath igher-status chi ldren, for example, tend to be rated
non-ethnic and standard(and also confident and eager), whereas
lower-status chilclren received theopposite ratings. As far as
ethnicity goes, there is relatively l i tt le differencc oneither
scale between black and white middle-sratus children, although
thewhite youngsters are rated a l itt le less confident and eager
and a l itt le morenon-ethnic and standard than the black
youngsters. The lower-status blackchildren are rated very much more
ethnic and nonstandard than white lower-status children, and also
somewhat less confident and eager.
wil l iams's resulrs. Perhaps the major result of the research
carriecl out bywill iams and his colleagues is the discovery of the
two-factor model thatre l iab ly and val id ly captures at least
two major d imensions of teacherattitudes. Beyond this, there were
other fascinating results. For example,when the ratings of white
versus black children by white teachers and blackteachers were
compared, it was discovered that black teachers rated blackchildren
substantially more non-ethnic and standord than white teachers
did(wil l iams 1973:122-3).ln the ratings of white children, borh
black and whireteachers evaluated high-status white children about
the same for ethnicity-nonstandardness, but black teachers found
low-status white children moreethnic and zozstandard than white
teachers did. In other words, where therewere differences between
the two groups, teachers tended to rate children oftheir own
ethnicit ies more non-ethnic and standard than teachers of the
otherethnicity did. The ratings on the confidence-eagerness
dimension betweenblack and white teachers were very similar.
Another key result found by wil l iams was the role of
stereotypes in theevaluation of speech (wil l iams 1973:l17-23).
wil l iams found it easy to elicit
-
Non-ethnic-standa rd
174 Language Attitudes
a : Low s ta tusO : M idd le s t a tusW: Wh i t eB : B l ack
Nonconl idenl Conf ident-eaget
Ethnic- nonstandardFigure 6.4 Graphic display of rhe two-factor
model
Source: adapted from Williams (1974:25,ligure l)evaluations o[
speech, using the same semantic-differential scales, when ,taspeech
sarnples were presented, but subjects were simply presented with
ethniclabels. For example, a teacher would be told to rate the
speech qualit ies of an'anglo', 'black' or 'Mexican American'
child, based on her experiences - realor anticipated - with such
children. Stereotypes elicited in this way weredifferentiated by
the ethnicity of the child. The stereotype for anglo children
isthat they are far more confident and eager and non-ethnic and
standard thanchildren from either of the other two groups. Black
and Mexican Americanchildrcn were rated close to each other toward
the ethnic-nonstandard end ofthe ethnic i ty-nonstandardness scale.
They were d is t inguished on theconfidence-eagerness scale,
however, with black children stereotypicallyrated substantially
more confident and eager than Mexican Americanchildrcn (but sti l l
not as confident and eager as anglo children). Not only was
itpossible to elicit stereotypes, but there was evidence that the
results werehighly re l iab le (Wi l l iams 1973: I l8-20) .
An interesting question occurred to the researchers: what role
didstereotypical attitudes play in teachers' evaluations of real
speech samples?Would a child be evaluated largely on the basis of
the stereotype, regardless ofwhat she sounded like? Or did
stereotypes come into play only when no speechsamples were
presented? Would teachers respond only to the characteristics ofthe
speech sarnple, regardless of the ethnicity of the child who was
talking? Orwas there perl.raps some sort of interplay between
stereotype and speech-sample characteristics? Will iams and his
associates developed an ingeniousmethod for answer ing th is quest
ion (Wi l l iams, Whi tehead, and Mi l ler l97 l ;Will iams 1973).
Specizri l videotapes were carefully prepared, involving thevisual
image of chitdre{ of all three ethnicit ies talking. In each
instance, thechildren were fi lmed from a side view so that their l
ips could not be read. This
Language Atli ludes 175
was important, because the same audio track was dubbed onto all
of thevideotapes. Furthermore, this audio track containecl a
standard Englishpassage. The result was that the subjects (this
time white student teachers)viewed black, anglo, and Mexican
American children, all seeming to speakstandard English.re Figure
6.5 displays the results.
o :
o -
VideotapeStereotypeWhite chi ldB lack ch i l dMexican American
chi ld
Noncon f i den t Conf ident-eager
M C F _
Figure 6.5 Ratings of the same standard English audio santples
with onglo,black, and Mexican Anterican video images
Source: from Williams (1973:126, figure 7)Notice first that the
stereotype and videotape ratings for the anglo child are
very close. The evaluations of the black and Mexican American
children are inmarkedly different positions, in spite of the fact
that the actual speech is thesame as was heard with the anglo
child's videotape. If speech characteristicswere the only thing the
subjects were responding to, all three circles should bein the same
place, that is, where the circle for the anglo child is. on the
otherhand, if the stereotypes alone were relevant, then the circles
for the black andMexican American children ought to be as close to
their stereotypc ratings asthe anglo child's videotape rating is to
his stereotype rating. what appears tobe happening is that the
stereotypes are a sor t of 'anchor point ' , usecl by theevaluators
when evaluating a speech sample. This anchor point keeps theactual
speech sample, no rnatter how standard it in fact is, from
beingevaluatdd too far from the stereotype. Nevertheless, the
speech clraracteri-s t i cs a re t aken i n to accoun t , w i t h t
he eva lua t i ons o f t he b lack andMexican Amer ican chi ldren
's s tandard Engl ish movi .g away f rom thestereotype 'anchors'
toward the non-ethnic-standard end of the ethnicity-nonstandardness
scale. Since the anglo child spoke tlre way the subjectsexpected,
the videotape and stereotype ratings are almost identical.
Althoughwi l l iams ( 1973:126) caut ions that these resul ts are
nor necessar i ly def in i t ive,because the experiment involved a
relatively small group of subjects and a
IIII t
Non-ethnic - s t a nda rd
Ethnic-non sta ndard
-
176 Language Attitudes
lirnitecl number of stimuli, i f they prove Lo be
representative, the significancecoulcl be important. It appears
that expectations might lead teachers to hear anon-whi te chi16 as
sounding somewhat ethnic and nonstandard, regardless oIhow standard
her speech actually is.20
Wi l l iams and h is col leagues have not taken the next s tep
yet and at temptedto show that the attitudes which their research
reveals actually influence howteachers act toward students. Other
researchers, however, have madeat tempts to invest igate th is
problem, Sel igman, Tucker , and Lambert (1972)ftrund that 'voice'
was just as significant as composition-writ ing and picture-
-
178 Language Attitudes
first for another group. when this order proved to make no dif
ference, the resultswere combined to give only three condit
ions'
l0 Apparently, direct questioning was -not used in the Welsh
experiment'
I t is
pJssibte that Canadian Frenchias advanced as a marker of group
identi ty sincethe mi