The medium-term dynamics of accents on reality television

43 Pages·2017·0.44 MB·English

THEMEDIUM-TERMDYNAMICSOFACCENTSONREALITYTELEVISION
MorganSonderegger MaxBane PeterGraff
McGillUniversityand UniversityofChicago UniversityofVienna
CentreforResearchon
Brain,Language,andMusic
Howflexibleisanindividual’saccentduringadulthood,andhowdoesthisflexibilityrelateto
longer-termchange?Previousworkhasfoundthataccentsareremarkablyflexibleinconversa-
tionalinteraction,butpredominantlystableoveryears,leadingtoverydifferentviewsoftherole
medium-term
ofindividualsincommunity-levelsoundchange.Thisarticleexamines accentdy-
namics(daystomonths)bytakingadvantageofa‘naturalexperiment’:arealitytelevisionshow
wherecontestantsliveinanisolatedhouseforthreemonthsandareconstantlyrecorded,forming
aclosedsystemwhereitispossibletobothdeterminethedynamicsofcontestants’speechfrom
daytodayandreasonaboutthesourcesofanyobservedchanges.Webuildstatisticalmodelsto
examinetimedependenceinfivephoneticvariableswithinindividuals,in14.5hoursofsponta-
neousspeechfromtwelveEnglish-speakingcontestants.Wefindthattimedependenceinpronun-
ciationisubiquitousoverthemediumterm:largedailyfluctuationsinpronunciationarethenorm,
whilelonger-termchangeoverweekstomonthsoccursinaminorityofcases.Thesepatternsmir-
rortheconflictingfindingsofpreviousworkandsuggestapossiblebridgebetweenthetwo.We
arguethattimedependenceinphoneticvariablesisinfluencedbycontrastbetweensounds,aswell
assystematicdifferencesbetweenspeakersinhowmalleabletheiraccentsareovertime;however,
wefindonlylimitedevidenceforconvergenceinindividuals’accents.Ourresultshaveimplica-
tionsfortheoriesoftheroleofindividualsinsoundchange,andsuggestthatmedium-termpro-
nunciationdynamicsareafruitfuldirectionforfuturework.*
Keywords:longitudinalstudies,accentdynamics,phoneticvariation,individualdifferences,lan-
guagechange
Introduction.
1. Two of the most striking aspects of language are stability and
change.Bothaspectsareespeciallyclearforhowsoundsarepronounced,thefocusof
thisarticle.Atthelevelofindividuals,weintuitivelyknowthataspeaker’saccentcan
shiftasafunctionofherinterlocutor,andthatsomepeople’saccentsseemtoshiftper-
accentchange
manently( ),forexample,aftermovingtoanewdialectregion.How-
ever, we also intuitively know that many people’s accents are stable over time: this
defaultassumptionunderliesourabilitytoidentifywheresomeoneisfrombytheirac-
cent.Atthelevelofspeechcommunities,manyaspectsofthephoneticandphonologi-
sound change
cal system are stable at a given point in time, yet cases of are always
occurring(e.g.Labov1994).Thisarticleaddressesthreebroadquestionsaboutaccent
dynamicsandsoundchange:howplasticaretheaccentsofindividualspeakersduring
adulthood,whatarethesourcesofthedynamicsofaspeaker’saccent,andwhatisthe
relationshipofaccentplasticityinindividualstocommunity-levelsoundchange?
*We thank Molly Babel,Lauren Hall-Lew,Tyler Kendall,Jane Stuart-Smith,NatashaWarner,and an
anonymousrefereeforhelpfulcommentsondraftsofthisarticle,andthemanyotherpeoplewhohavepro-
videdusefulfeedbackonthisproject,includingPeterAuer,RinaFoygelBarber,PaulBrazeau,AlexFine,
JamesKirby,MattGoldrick,JohnGoldsmith,DanJurafsky,RobinQueen,JasonRiggle,TylerSchnoebelen,
CharlotteVaughn,andAlanYu;aswellasaudiencesatCLS2010,LSA2011,CUNY2011,NWAV2012,
LabPhon2014,ICPhS2015,andatUSC,IndianaU.,U.Michigan,McGill,U.Edinburgh,U.Ottawa,U.
Minnesota,U.Toronto,Cornell,NorthwesternU.,andMIT.WethankN.Rothfels,M.Nelson,T.Knowles,
M.Schwarz,M.Labelle,L.Bassford,R.Hwang,H.Hsieh,andM.Coumelforresearchassistance.Wegrate-
fullyacknowledgepermissionfromChannel4/EndemoltoaccessfootagefromBigBrother9UK.Prelimi-
naryversionsoftheVOTresultsweregiveninBaneetal.2010andSonderegger2015,andinSonderegger
2012forallvariables.ThisworkwassupportedbygrantsfromtheSocialSciencesandResearchCouncilof
Canada(#430-2014-00018)andtheFondsdeRechercheduQuébecSociétéetCulture(#183356).
598
PrintedwiththepermissionofMorganSonderegger,MaxBane,&PeterGraff.©2017.
Themedium-termdynamicsofaccentsonrealitytelevision
599
short-term
Previousworkhasaddressedthesequestionsontwotimescales: shifts
inpronunciationduringconversationandlaboratoryexperiments(seconds–hours),and
long-term
accentchangeoverthelifespan(years).
Short-term.
1.1. Onelong-standingbodyofworkemphasizestheflexibilityofthe
sound systems of individuals, which are constantly updated over time as a result of
interaction(Hockett1965,Paul1880;seeGarrett&Johnson2013).Forexample,con-
emergentist
temporary theories of phonetics and phonology (e.g. Johnson 1997,
Pierrehumbert2001),whosecoretenetsgobacktotheNeogrammarians,assumethata
speaker’scognitiverepresentationofsoundsincludesallexampleseverheard.Thisrep-
resentationisupdatedwitheachnewinteractionandusedinspeechproduction.Thus,
emergentist models predict that pronunciation norms can and will shift over time, in-
cludingduringadulthood,asaresultofinteractions.
Aprimary source of evidence for accent plasticity in adults comes from short-term
shiftsinhowonespeaks(overseconds–hours),inconversationsandinlaboratoryset-
tings,underexposuretootherpeople’sspeech.Theseshifts(termedimitation,(phonetic)
convergence,accommodation,etc.)havebeendocumentedusingavarietyofparadigms
(Pardo2013).Asassessedbyperceptualmeasures,subjectsshifttheiroverallpronunci-
ations toward a target voice in word-shadowing tasks (Fowler et al. 2003, Goldinger
1998)andtowardtheirinterlocutorinconversationalinteraction(e.g.Pardo2006).Spe-
cificacousticvariablestypicallyalsoundergoshort-termshiftsinbothinteractiveand
noninteractivesettings.Forexample,AmericanEnglishspeakersincreasedtheVOTof
word-initial voiceless stops toward a model talker in an imitation task (Nielsen 2011)
andshiftedtheirproductionsoffivevowels(inF1/F2space)towardamodeltalkerina
shadowingtask(Babel2011).Pardo(2009)alsofoundthatAmericanEnglishspeakers
shiftedvowelformantsinconversationalinteraction,butinmorecomplexways.Inthese
andotherstudiesofshiftsinphoneticvariables,theexistenceofsomeshift,asaresultof
exposuretoanotherperson’sspeech,isquiterobustacrossspeakers.
Short-termshiftsinpronunciationaretosomeextent‘automatic’consequencesofin-
teraction(Delvaux&Soquet2007,Goldinger1998).Theyarealsoheavilymodulated
bythreetypesoffactors,whichresultinthesignificantvariabilitybetweenspeakersin
thesizeanddirectionalityofshiftsobservedinmostshort-termstudies:
Socialfactors
ii(i) ,suchasattitudetowardtheinterlocutorandgender(Babel
Communication accom-
2010, Bourhis & Giles 1977, Namy et al. 2002).
modationtheory
(Gilesetal.1991,etseq.)proposesthattheseshiftsresult
accommodation
from individuals managing social distance using an strat-
convergence divergence
egy,suchas or (e.g.toexpresssolidarityordis-
approval).
Linguistic factors
i(ii) . For example, Nielsen (2011) cites contrast mainte-
nancetoexplainwhyEnglishspeakerswillshiftVOTofvoicelessstopsto-
wardlengthenedbutnotshortenedVOT(cf.Babel2011,Mitterer&Ernestus
2008).
Individual differences
(iii) , correlated with factors such as personality traits
andgender(Namyetal.2002,Yuetal.2013).
Insum,plasticityinpronunciation,modulatedbymultiplefactors,isthenormover
styleshifting:
seconds–hoursforadults.Thisplasticityisinlinewiththeubiquityof
shiftsinaspeaker’slinguisticusageasafunctionoftheaddressee,topic,andsoforth
(e.g. Bell 1984, Eckert 2000, Rickford & McNair-Knox 1994), possibly many times
overthecourseofaday(e.g.Coupland1980,Hindle1980).
600 LANGUAGE,VOLUME93,NUMBER3(2017)
Changebyaccommodation
.Theexistenceofsubstantialaccentplasticityininter-
actionsgoesnaturallywiththeviewthatchangeduringadulthoodisanimportantdriver
ofsoundchangeincommunities.VersionsofthisviewgobacktotheNeogrammarians
changeby
(Paul1880),assummarizedbyAuerandHinskens(2005),whocallitthe
accommodation model
(CBA) .TheCBAmodelexplainstherelationshipbetweenin-
dividual-levelandcommunity-levelchangeasasequenceofthreesteps:(1)peopleac-
commodatetoeachotherduringinteractions;(2)eventually,agivenindividual’snorms
changeastheresultofaccumulationoftheseinteractions;(3)theinnovationspreadsin
the wider community (as individuals undergo step 2). Different versions of the CBA
modeladdressdialectchange/contactsettingsalone,orallsoundchange(e.g.Bloom-
field 1933, Garrett & Johnson 2013, Labov 1990, Trudgill 1986, 2004), and differ in
whether short-term shifts (and thus steps 2–3) are fundamentally automatic (every in-
teractionmakesthesoundsystemofinterlocutorsmoresimilar)orfundamentallysocial
(as proposed by communication accommodation theory). Regardless of their mecha-
nism, short-term shifts are generally assumed to be ‘the driving force of language
change’(Auer&Hinskens2005:356).Thefactthatmostshort-termliteratureconsiders
adultssuggeststhatsteps1–2referbydefaulttoadults.Thus,thelinkbeingmadebe-
tween accent plasticity in adulthood and the nature of sound change is this: because
long-termchange
adults’pronunciationsaresoplasticininteractions(step1), during
soundchange
adulthood(step2)playsanimportantroleincommunity-level (step3).
Long-term.
1.2. Onatimescaleofyears,thetraditionalviewisthatanindividual’s
linguisticsystemislargelyfixedinadulthood(e.g.Chambers2003:197).Theassump-
apparent-time
tionofaccentstabilityoveryearsunderliesthewidespreaduseofthe
construct
tostudysoundchangeinprogressusingasynchronicsample(Cukor-Avila
&Bailey2013).
Previous work has examined the extent of postadolescent accent plasticity in two
kindsofsettingswhereaccentchangetowardchangingcommunitynormsseemsintu-
itively likely: individuals who remain in the same speech community, where some
panelstudies
changeisinprogress( ;Sankoff2005,2013),andindividualswhohave
dialect change studies
moved between dialect regions ( ; Auer et al. 2005, Siegel
2010).1 For example, Sankoff and Blondeau (2007) examine change in rhotic realiza-
tion([r]→[ ],acommunity-levelsoundchangeinprogress)inMontrealFrenchusing
ʀ
apanelofthirty-twospeakers,recordedelevenyearsapart,andfindrelativestabilityin
mostindividuals(72%),withtheremainderincreasingtheir[ ]use.Whilemostlarge-
ʀ
scaledialectchangestudiesdonotbreakdownresultsbyindividualspeakers,thosethat
do suggest that significant accent change in adulthood is uncommon. For example,
Foreman (2003) examines six American speakers who settled in Australia for whom
longitudinal data is available over ten to twenty-seven years; five of the six speakers
show little or no significant change across six phonological variables. However, case
studieshavehighlightedthefactthatdramaticaccentchangeispossible.Forexample,
QueenElizabethII’srealizationofEnglishvowelshaschangedsignificantlyoverfifty
yearsofradioaddresses,oftenparallelingcommunity-levelchanges(Harringtonetal.
2000, et seq.); andYiddish folk singer Sarah Gorby shows a mixture of stability and
changeacrossphonologicalvariablesoverfiftyyears,withchangegenerallytowardthe
standard(awayfromhernativedialect)(Prince1987).Moregenerally,long-termstud-
iesfindthatbothvariablesandspeakers(foragivenvariable)differsubstantiallyinthe
degreeofplasticityoveryears(Siegel2010:51),foravarietyofreasons.Thoughfurther
1Wedonotconsideragegrading,whereavariable’susechangesduringadulthoodinapredictableway.
Themedium-termdynamicsofaccentsonrealitytelevision
601
workisneededgiventhisheterogeneity,thepicturethatemergestodatefromlong-term
studies across a range of languages and variables (e.g. Bowie 2005, Brink & Lund
1975,Nahkola&Saanilahti2004,Sancier&Fowler1997,Stanford2008)isthat‘the
defaultforadultsisapparentlystability’,whileaminorityshowsignificantpronuncia-
tionchange,usuallyinthedirectionofcommunity-levelchangesinprogressorambient
norms(Sankoff2013:274).
Generationalchange
.Thefindingthatadultsshowlimitedandheterogeneouslong-
termplasticityinpronunciationgoesnaturallywiththeviewthatchangeduringadulthood
isnotaprimarydriverofcommunity-levelsoundchange.Aninfluentialproponentof
thisviewisLabov(1994,2007),whoarguesthatpronunciationchangesinternaltospeech
generationalchange transmis-
communitiesoccurprimarilyvia :intergenerational
sion incrementation
ofnormsfromadultstochildren,and bychildreninthedirection
ofchangesinprogressduringchildhoodandadolescence.Bycontrast,thetypicalmech-
anismofchangesrootedindialectcontactisdiffusioninthecourseofinteractionbetween
adultswhousedifferentpronunciations.Diffusionislessimportantthantransmissionas
asourceoflinguisticdiversitybecauseitisdrivenbyadults,whoshowlessandmorespo-
radic‘capacitytochangetheirlinguisticsystems’comparedtochildrenandadolescents
(Labov2007:349).WhileLabov’saccounthasbeensubjecttodebate(e.g.Babeletal.
2013),itrepresentsacommonview:changeinadulthoodisnotanimportantfactorinthe
majorsourceofsoundchange(internalchange),whileexternalchangeisalessimportant
sourcepreciselybecausechangeinadulthoodislimitedandsporadic.
Mediumterm
1.3. .Empiricalevidencefromtheshort-termandlong-termliteratures
broadlysuggeststhataccentsarebothplasticduringinteractionsandlargelystableover
thelifespan,andassociatedtheoreticalviewpointsreachconflictingconclusionsaboutac-
centplasticityinindividualsanditsrelationshiptocommunity-levelsoundchange.This
conflictcomesfromextrapolatingfromshort-termandlong-termresultstoaccentplas-
mediumterm
ticityinindividualsoverthe (days–months),aboutwhichlittleisknown.
Long-termstudiessuggestthataccentstabilityisthenorminadulthoodoveryears,
butlittleisknownaboutaccentdynamicsinadultsonshortertimescales(cf.Barden&
Großkopf 1998, Evans & Iverson 2007, Pardo et al. 2012, where sampling points are
separatedbymonths).Areasonablenullhypothesisformedium-termdynamicswould
sta-
bethatanindividual’sspeechdoesnotchangeoverdays–months;wecallthisthe
tionarityhypothesis
.Wecandistinguishtwoversionsofthestationarityhypothesis,
thinking about the trajectory of an individual’s ‘baseline’ use of a phonetic variable
everyday,overaperiodofmonths(schematizedinFigure1),aftercontrollingforother
factors(style,linguisticcontext,etc.).Inthestrongversion,therecouldbenotimede-
pendenceatall(typeA),eitherfromdaytodayoroverlongerperiods.Aweakerver-
by-dayvariability
sionwouldbethatindividualsvaryfromdaytoday( ),butnotover
longerperiods(typeC).Thereisalmostnoworkexaminingwhetheranindividual’sac-
centfluctuatesfromdaytoday(cf.Heald2012,Pisoni1980,discussedbelow),though
NahkolaandSaanilahti(2004)suggestthispossibility.2
Short-termstudiesshowthatspeakersregularlyadjusttheirpronunciationduetothe
speechofothers.Theproposedlinkbetweenshort-termshiftsininteractionandlonger-
term change is at the heart of the CBAmodel and motivates many short-term studies
(e.g. Babel 2011, Delvaux & Soquet 2007, Nielsen 2011, Pardo 2006).An important
persistencehy-
initialassumptionbeingmadeintheseliteratures,whichwecallthe
2Incontrast,Cukor-Avila(2015)andRickfordandPrice(2013)haveexamineddailyfluctuationsinindi-
morphosyntactic
viduals’useof variables,focusingoneffectsofstyleshiftingandmethodology.
602 LANGUAGE,VOLUME93,NUMBER3(2017)
Figure
1.Schematicofpossibletypesoftimedependenceinaphoneticvariablewithinindividualspeakers:
nochange(A),timetrend(B),by-dayvariability(C),by-dayvariabilityandtimetrend(D).Solidline
indicatesthevariable’svalueoneachday;dashedlinesindicateitsmeanvalueoverseveraldays.
pothesis
,isthatshort-termshiftspersistbeyondindividualinteractions;theymaythen
accumulateandeventuallyleadtochangeinanindividual’spronunciationnormsover
themediumterm.Toourknowledge,however,thereisnoworkexaminingthepersis-
tence of short-term shifts in use of a phonetic or phonological variable for more than
onehour.3Thepersistencehypothesispredictsthatifanindividualengagesinthesame
type of interactions sufficiently often, there should be steady change in his ‘baseline’
use of a phonetic variable over days–months—beyond any by-day variability—which
time trend
we term a .Aphonetic variable could show a time trend without by-day
variability(Fig.1:typeB),orbothatimetrendandby-dayvariability(Fig.1:typeD).
By-dayvariabilityandtimetrendsformtwoindependentdimensionsofmedium-term
timedependence:anindividuals’pronunciationcouldshoweither,both,orneither.
Medium-termaccentdynamicsinaclosedsystem
1.4. .Thedisconnectbetween
short-termandlong-termplasticityandthepaucityofempiricaldataontimescalesinbe-
medium-term
tweenmotivatesthecurrentstudy,whichinvestigates timedependencein
speechproductioninindividualsbytakingadvantageofa‘naturalexperiment’:thereal-
itytelevisionshowBigBrotherUK,whosestructureisuniquelysuitedtoinvestigating
howandwhyanindividual’saccentchangesoverthemediumterm.Theshowcontains
speechfromthesameindividuals,recordedonanear-dailybasisoverthreemonths,mak-
ingitpossibletoexaminewhetheraspectsofcontestants’speechshowby-dayvariabil-
ity,timetrends,orboth.Contestantsinteractwitheachotherconstantly,withoutaccess
totheoutsideworld;thehouseisthusalinguistically‘closedsystem’(Baneetal.2010),
asmallcommunitywherepersistenceofshort-termshiftsinpronunciationmightbeex-
pectedtobeespeciallylikely,andwhereitisinprinciplepossibletotestwhetherchange
inanindividual’sspeechcanberelatedtosocialinteraction.
We first describe a corpus of spontaneous English speech from one season of this
show, in which we examine five phonetic variables for twelve contestants over up to
threemonths(§§2–3).Foreachvariable,webuildstatisticalmodelstocharacterizeits
timedependencewithineachspeaker,aftercontrollingforotherfactors(§4).Wethen
usetheresultsofthesemodelstoaddresstworesearchquestionsaboutaccentdynamics
overthemediumterm.
First, what qualitative kinds of time dependence do phonetic variables show within
individualspeakersoverthreemonths(§5)?Weconsiderthetypesoftimedependence
showninFig.1,andwefindthatmedium-termtimedependenceofonekindoranother
isubiquitousacrossspeakersandvariables,andthattimedependenceisdueprimarily
toby-dayvariabilityandsecondarilytotimetrends.Second,towhatextentcanweac-
3Goldinger(2000)showspersistenceoveroneweekofanoverallpercept,butacousticvariablesarenot
examined.
Themedium-termdynamicsofaccentsonrealitytelevision
603
count for the observed patterns of variability over time (§6)? We consider the same
typesofpotentialsourcesthatmediateshort-termpronunciationshifts—socialfactors,
linguisticfactors,andindividualdifferences—andfindsomeevidenceforaroleofeach
inshapingmedium-termaccentdynamics.
Thesefindingsallowus(§7)toevaluatethestationarityandpersistencehypotheses,
to situate medium-term dynamics between the disparate patterns individuals show on
short-termandlong-termtimescales,andtoevaluatethemechanismofaccentdynam-
icsinindividualsandthepossiblerelationshiptocommunity-levelchange.
Thebigbrothercorpus
2. .
Show description
2.1. .The Big Brother corpus consists of speech from the ninth
seasonoftherealitytelevisionshowBigBrotherUK(Channel4/Endemol),whichaired
from June 5 to September 5, 2008 (ninety-three days).4 Twenty-one contestants (or
housemates
) took part in the show: sixteen entered the Big Brother house on day 1,
andfivemoreenteredlaterintheseason.Housematesweregraduallyevictedduringthe
seasonbyacombinationofnominationbyotherhousematesandvotingbytheviewing
public.Thelasthousemateremainingwonacashprize.
While on the show, housemates had essentially no linguistic input from the outside
world:theydidnotleavethehouse,interactwithpeoplenotinvolvedwiththeshow,or
(withrareexceptions)haveaccesstomedia.Theyspentmostoftheirtimeinteracting
witheachotherinsomeform,forexample,inunstructuredconversationsorwhilepar-
diary
ticipating in tasks set by Big Brother. The one important exception was the
room
, an isolated room where housemates could go to speak with ‘Big Brother’. Big
Brotherwasinfactdifferentpeople(maleandfemale)atdifferenttimes;he/shecould
seehousemates,butcouldnotbeseenbythem,andcommunicatedonlythroughaudio.
Housemateswenttothediaryroomtotalkabouttheirfeelings,eventsinthehouse,and
soforth,andcouldeithergotothediaryroomvoluntarilyorbecalledbyBigBrother.
Housemateswererecordedatalmostalltimes,includingbywearablemicrophones.
Duringtheseason,thepubliccouldseevideoofhousematesviaalivefeedorvariouspro-
ducedshows,includingdaily‘Highlights’showsandweekly‘Diaryroomuncut’shows,
whichconsistedofcontinuoussegmentsfromthehouse,presentedwithoutcommentary.
Table 1 gives basic demographic information for the twenty-one housemates, who
come from diverse dialect regions. Sixteen housemates are native speakers of British
dialects. Five housemates reside in the UK but are not native speakers of British di-
alects. Sara and Darnell are native English speakers fromAustralia and the USA, re-
spectively;SylviaandMohamedwereborninSierraLeoneandSomalia,respectively,
and speak near-native English with light accents. Kathreya is a native Thai speaker
whoseEnglishisheavilyaccentedandfrequentlyungrammatical.
Corpusdescription.
2.2. Thecorpusconsistsofallsegmentsfromthe‘Highlights’
and‘Diaryroomuncut’showswhereasinglehousematewasinthediaryroom,which
diary room clips clips
we call (or ).Audio from all clips was broadcast quality.The
749 clips contain roughly 14.5 hours of housemate speech. The corpus is unbalanced
across housemates (Table 1), because housemates were on the show for different
amountsoftime:thelesstimeahousematespentinthehouse,thelessfrequentlytheir
speechwassampled(clipsoccuronceper0.8–5.4days,fordifferentspeakers).Thus,to
addressourresearchquestionsabouttimedependenceinindividualhousemates’useof
4 Information about the show and housemates comes fromWikipedia (2012a,b).This corpus is an ex-
pandedversion(roughlytwiceaslarge)ofthatusedinSonderegger2012.
604 LANGUAGE,VOLUME93,NUMBER3(2017)
speaker gender age dialectregion dayson #ofclips housemate
show speech
Dale M 21 N.England 65 43 0:43
Darnell M 26 USA 93 73 1:24
Kathreya F 31 L2Thai 90 56 1:08
Lisa F 40 N.England 86 53 1:02
Luke M 21 N.England 58 49 1:09
Michael M 34 S.Scotland 93 81 1:42
Mohamed M 25 London 90 51 0:53
Rachel F 24 S.Wales 93 69 1:03
Rebecca F 21 W.Midlands 51 34 0:32
Rex M 24 London 93 71 1:25
Sara F 27 Australia 64 39 0:44
Stuart M 25 N.England 57 32 0:38
Alexandra F 23 London 14 17 0:25
Belinda F 44 S.England 15 7 0:08
Dennis M 24 S.Scotland 23 7 0:09
Jennifer F 22 N.England 30 13 0:20
Mario M 43 N.England 37 15 0:19
Maysoon F 29 London 27 5 0:03
Nicole F 19 S.England 22 19 0:24
Stephanie F 19 N.England 9 9 0:08
Sylvia F 22 London 23 6 0:09
total
749 14:28
Table
1.Demographicinformation,lengthofstayintheBigBrotherhouse,andamountofdataintheBig
Brothercorpusforeachofthetwenty-onehousematesfromBigBrother9UK.Thehorizontalline
separatesthetwelve‘corehousemates’(top)andtheotherhousemates.
each phonetic variable—in particular, distinguishing between by-day variability and
time trends—we can only consider housemates who spend a relatively long period in
thehouse.Fiftydays(ofninety-three)waschosenasanarbitrarycutoff,leavingtwelve
core housemates
housemates (of twenty-one), who we refer to as the (seeTable 1).
The core housemates account for 85.6% of speech in the corpus and are the focus of
mostanalysesbelow.
The corpus is limited to diary room clips of single housemates in order to best ad-
dress our research questions, given that it was not feasible to transcribe more than a
fractionoftheproducedepisodes(>100hours).Limitingthecorpustoonetypeofin-
teractioninarelativelyconstantsettingallowsustominimizedifferencesinspeaking
stylebetweendifferentdays(i.e.clips)andthusbetterassessthestationarityhypothe-
sis.5TheonlyinteractionintheclipsiswithBigBrother,whoseroleisusuallylimited
tobriefquestionsoranswers.Theregisterofthespeechisgenerallycasualandconver-
sational, with characteristics of a sociolinguistic interview, but with the clear self-
awareness and performativity expected given that the interaction may eventually be
televised.Thecorpusallowsustoexaminehoweachhousemate’s‘baseline’linguistic
usagevariesovertime,abstractingawayfromshort-termshiftsthatmayoccurduring
conversationwithotherhousemates.
Eachclipwassegmentedintospeakerturnsandorthographicallytranscribedbyre-
searchassistantsandthefirstauthor.Theorthographictranscriptionandaudioofeach
clip were then force-aligned using a version of the HTK-based aligner from FAVE
(Rosenfelderetal.2011,Youngetal.2006),customizedfortheBigBrotherCorpus.
5Notethatitwasnotpossibletoassesstimeofday(e.g.morningvs.evening)giveninformationinthecor-
pus;see§7.1fordiscussion.
Themedium-termdynamicsofaccentsonrealitytelevision
605
Data.
3. Weexaminethedynamicsoffiveaspectsofpronunciationforspeakersin
phonetic variables
the Big Brother house ( ): voice onset time (VOT), coronal stop
goose strut trap
deletion(CSD)rate,andthequalityofthreevowels( , ,and ,usingthe
‘lexicalset’notationofWells1982foravowel’srealizationinagivendialect).
These variables provide complementary evidence for our research questions about
theexistenceandsourcesofmedium-termtimedependenceinpronunciation.First,all
variables differ greatly across varieties of English, meaning the possibility exists for
contestants’ pronunciations to influence each other. VOT has been found to be very
flexibleovershort-termtimescales(e.g.Nielsen2011,Sancier&Fowler1997),making
it a logical place to look for medium-term plasticity. Examining CSD lets us study
whethermedium-termtimechangeobtainsformorecategoricalaspectsofpronuncia-
tion, as well as continuous phonetic parameters. The three vowels examined differ
along two dimensions thatare possible sources of medium-termtimedependenceand
mightaffectthelikelihoodofconvergencebetweenhousemates.Previousworkhasar-
goose strut trap social
gued that, compared to , the realizations of and have high
salience
acrosstheUK,whichhasbeenarguedtoconstrainwhichvariablescanshift
overtheshortandlongterm(e.g.Babel2010,Trudgill1986)andmightaffectavari-
Goose soundchange
able’smedium-termplasticityaswell. isundergoing incommu-
nities across the UK (e.g. Docherty 2010, Haddican et al. 2013, Hawkins & Midgley
strut trap
2005),while and arerelativelystable.Speakersmightshowmoreflexibil-
ityintherealizationofavowelundergoingsoundchange,iftheirexposuretoagreater
numberofvariants(outsidethehouse)givesthemabroaderrangeof‘self-exemplars’
todrawoninconvergingtowardotherspeakers.
Wefirstintroduceeachphoneticvariable,thendescribeannotationanddata-cleaning
steps,andsummarizetheresultingdataset.
Voice onset time
3.1. . VOT, the time difference between the onset of a stop’s re-
lease burst and the onset of voicing in a following segment, is an important phonetic
cueforthecontrastbetweenEnglish‘voiced’and‘voiceless’stops(e.g.Docherty1992,
Lisker&Abramson1967).VOTmaybepositive(burst+aspirationduration)ornega-
tive (duration of voicing preceding burst). (Phonologically) voiceless stops are pro-
duced with variable degrees of positiveVOT, while (phonologically) voiced stops are
produced with either shorter positive VOT or negative VOT. Variability in VOT has
beenstudiedmostlyinlabspeech(seeAuzouetal.2000,Docherty1992),butsomere-
centworkexaminesspontaneousEnglishspeech(Chodroffetal.2015,Yao2009),in-
cluding Stuart-Smith and colleagues’ (2015) study of variability in positive VOT in
spontaneousGlasgowvernacular—whoseanalyticalchoicesweoftenfollow,andwho
discuss the issues involved in measuring VOT in spontaneous speech. Like Stuart-
Smithandcolleagues,wemeasuredpositiveVOT(summedburstandaspirationdura-
word-initial
tion) for every stop token produced with a burst or aspiration, without
taking account of negative VOT or voicing during the closure. Thus, ‘voiced stops’
phonemically
throughoutthisarticlealwaysmeans voicedstops,asopposedtostops
producedwithnegativeVOT.
Annotation and data set
. We measured VOT semi-automatically for all word-
initialstopsinthecorpus(n=16,784voiced,13,777voiceless),usingaproceduresim-
ilartothatofStuart-Smithetal.2015.
automaticmeasurement
First,an ofVOTforeachtokenwasobtainedbyapply-
ingAutoVOT(Keshet et al. 2014, Sonderegger & Keshet 2012).AutoVOTrequires a
classifierthathasbeentrainedonmanuallylabeledVOTdata,aswellasawindowof
606 LANGUAGE,VOLUME93,NUMBER3(2017)
timeintheaudiofileforeachtoken(specifiedinaPraatTextGrid;Boersma&Weenink
2011)inwhichtosearchforthebeginningoftheVOTinterval.Applyingtheclassifier
toeachtokeninafileyieldspredictedVOTintervals,whichareoutputtedonanewtier
oftheTextGrid.TopredictVOTforvoicelessandvoicedstops,weusedtheclassifiers
forvoicelessandvoicedEnglishstopsdistributedwithAutoVOT.Thewindowforeach
token was taken to begin and end 25 ms before and after the force-aligned segment
boundaries.Theparametersforminimum/maximumpredictedVOTweresetto15/250
msforthevoicelessclassifierandto5/150msforthevoicedclassifier,andotheralgo-
rithmparameterswerekeptatdefaultvalues.
manualcorrection
Second,inthe phase,twoorthreephoneticallytrainedannota-
tors(twoforvoiced,threeforvoicelessstops)reviewedthepredictedVOTintervalsin
Praat.Foreachinterval,theannotator:(a)checkedwhethernoburstwaspresent(e.g.
stopisrealizedasafricative),inwhichcasethetokenwasmarkedforexclusion;(b)de-
terminedwhethertheVOTintervalboundarieswerewhereshewouldhaveplacedthem
ifannotatingVOTmanually;and(c)manuallycorrectedtheboundaries,ifthiswasnot
thecase.Step(a)resultedin3,474excludedtokens.Manualannotation(stepsb–c)was
performed in Praat.The left boundary was placed wherever the first ‘large’amplitude
increaseinhigh-frequencyfricationoccurred,establishedusingtheamplitudetrackand
spectrogram; if there was a gradual rise in frication, the boundary was placed at the
midpoint.Whenmultipleclearlyseparateburstsoccurred,thelastonewasusedasthe
leftboundary.Therightboundarywasdeterminedprimarilyusingthewaveform:ifpe-
riodicitywasnotpresentbeforetheburst,atthezerocrossingclosesttotheonsetofpe-
riodicity;ifperiodicitywaspresentthroughouttheburst,atthepointwhereamplitude
begantoriseandgeneralwaveformshapechangedabruptly.
Anumberoffurtherexclusionsweremade.All1,511tokensfromKathreyawereex-
cluded, since her data may show extensive transfer fromThai (which has a three-way
VOTcontrastforstops).Atotalof280tokenswithmissingvaluesforvariablesusedin
themodelsbelow(listedinTable2)wereexcluded.EightytokenswithVOTsoutsideof
1–80msforvoicedstops,or8–175msforvoicelessstops,wereexcludedashavingex-
treme VOTs, with the cutoffs determined by visual inspection of the distribution of
VOT(separatelyforvoicedandvoicelesstokens).Becausespeechratehasalargeef-
fectonVOTrelativetoothervariables,weexcludedsixty-twotokenswithaspeechrate
greater than ten syllables/second as having extreme speech rates (presumably due to
forced-alignmenterrors),withthecutoffdeterminedbyvisualinspectionofthedistri-
butionofalltokens.
ThefinalVOTdatasetconsistsof12,908voicedtokens(from788words,678clips)
and12,246voicelesstokens(from964words,668clips),acrosstwentyspeakers.
Coronalstopdeletion
3.2. .Coronalstopdeletion(a.k.a.t/d-deletion)isavariable
processinEnglishinwhichword-finalcoronalstops(/t/and/d/)aresometimesdeleted
inword-finalconsonantclusters(e.g.bestas[b s]vs.[b st]).6CSDhasbeenexamined
ɛ ɛ
indozensofstudies(usually)ofspontaneousspeechacrossmanyvarietiesofEnglish
overthepastfiftyyears(reviewedinHazen2011,Schreier2005,Tagliamonte&Temple
2005).FewerstudieshaveexaminedCSDinBritishEnglishvarieties(e.g.Tagliamonte
&Temple2005,Temple2009);particularlynotableisTagliamonteandTemple’sstudy
ofYorkEnglish,whoseanalyticalchoicesweoftenfollow.Mostpreviousworktreats
6CSDisonetypeoffinalconsonantclusterreduction(CCR)inEnglishandthemostcommonlystudied.
SeeHazen2011ontherelationshipbetweenCSDandCCR.
Themedium-termdynamicsofaccentsonrealitytelevision
607
CSDasabinaryvariable,although‘deletion’isclearlyacategoricalapproximationfor
gradientrealizationofcoronals(e.g.Temple2009).Weuseacategoricalbinarynotion
of‘deletion’inordertomakeuseoftheresultsofpreviouswork—includingTanneretal.
2017,wheretheCSDdatasetdescribedbelowisanalyzedfromadifferentperspective,
unrelatedtoourresearchquestionsabouttimedependence.
Annotationanddataset
.Annotationwascarriedoutbyfourphoneticallytrained
annotators, for all speakers except Kathreya (twenty speakers), whose data was ex-
cludedduetoanear-categoricaldeletionrate(presumablybecauseword-finalclusters
ending in t/d are phonotactically illegal in Thai). Each token (n = 14,259) for these
twentyspeakerswhoseunderlyingformendedinat/d-finalconsonantclusterwasman-
ually annotated for the surface realization of the preceding consonant and the phono-
logical context surrounding the final coronal, the length of any following pause (or
other nonspeech), and the CELEX wordform ID (which was used to determine mor-
phologicalclass),aswellastherealizationofthefinalcoronal.7
Coronalstoprealizationwasfirstannotatedatafine-grainedlevel,usingspectraland
auditorycriteria.Annotatorschosefromeightcategoriesdescribingpossiblesurfacere-
alizationsoft/d(usingtheacousticcuesofburst,glottalization,suddenclosure;realiza-
tionasaglottalstop,anunreleasedcoronal,etc.).Incaseswheretheunderlyingt/dwas
followedbyasurfacet/d(e.g.wantto),thetargett/dwastakentoberealizedonlyin
cases where there was evidence for realization of a sequence of two distinct coronal
stops.Ifthesecondt/dwasclearlypresentfollowingaclosure(usuallyduetothepres-
enceofaburst)andnootherevidenceofaprevioust/drealizationwaspresent(suchas
a glottal stop), the t/d was taken to be the realization of the following word, and the
word-finalt/dwasannotatedasunrealized.
Thefine-grainedannotationwasthencollapsedintoabinaryvariableofpresentver-
susdeleted,followingpreviouswork(e.g.Tagliamonte&Temple2005),wherethepres-
enceofanyunambiguousphoneticreflexoftheunderlyingt/dindicatespresence.We
alsofollowedTagliamonteandTemple(2005)indiscardingalltokensendingin/rt/or
/rd/inrhoticvarieties,giventhatmanyhousematesarespeakersofnonrhoticvarieties.
Wefurtherexcluded141tokenswithoutaprecedingconsonantintheirsurfacereal-
ization, twenty-two tokens where speech rate could not be reliably determined, one
tokenwithannotationerrors,andfifty-ninetokensatturnends.Thislaststepwasnec-
independent
essaryinordertoincludepausedurationandfollowingcontextas factors
affectingCSDrate,followingTanneretal.2017,wherethemotivationforthischoice
(ratherthancodingpauseasanalternativefollowingcontext)isdiscussed.
The final data set consists of 12,788 tokens (from 570 words, 667 clips), across
twentyspeakers.
Vowel formants.
3.3. Accents of English differ primarily in their vowels (Wells
1982).WithinagivenvarietyofEnglish,vowelqualityisdeterminedprimarilybythe
goose strut
firsttwoformants(F1,F2).WeconsideronlyvariationinF1andF2,for , ,
trap
and .
TherealizationofthesevowelsvariessystematicallyacrossEnglishdialectregions;
thisvariationwillbeimportantforassessingwhetherfactorssuchassocialsalienceaf-
fect medium-term time dependence.We describe the expected pronunciations of each
7Wedidnotmakeexclusionssuchasfrequent(e.g.and)orcontracted(e.g.don’t)forms,asinmuchpre-
viousCSDwork,insteadaccountingforlexicaldifferencesusingby-wordrandom-effectterms.