Salient Accented, but Not Unaccented Words Reveal Accent Adaptation Effects

17 Pages·2016·1.19 MB·English

View metadata, citation and similar papers at core.ac.uk brought to you by CORE
provided by Frontiers - Publisher Connector
ORIGINALRESEARCH
published:07June2016
doi:10.3389/fpsyg.2016.00864
The Penefit of Salience: Salient
Accented, but Not Unaccented
Words Reveal Accent Adaptation
Effects
Ann-KathrinGrohe*andAndreaWeber
PsycholinguisticsandAppliedLanguageStudies,EnglishDepartment,FacultyofHumanities,UniversityofTübingen,
Germany
In two eye-tracking experiments, the effects of salience in accent training and speech
accentednessonspoken-wordrecognitionwereinvestigated.Saliencewasexpectedto
increaseastimulus’prominenceandthereforepromotelearning.Atraining-testparadigm
was used on native German participants utilizing an artificial German accent. Salience
was elicited by two different criteria: production and listening training as a subjective
criterion and accented (Experiment 1) and canonical test words (Experiment 2) as
an objective criterion. During training in Experiment 1, participants either read single
German words out loud and deliberately devoiced initial voiced stop consonants (e.g.,
Balken—“beam” pronounced as ∗Palken), or they listened to pre-recorded words with
Editedby: thesameaccent.Inasubsequenteye-trackingexperiment,lookstoauditorilypresented
BerndKortmann,
target words with the accent were analyzed. Participants from both training conditions
UniversityofFreiburg,Germany
fixatedaccentedtargetwordsmoreoftenthanacontrolgroupwithouttraining.Training
Reviewedby:
MelissaMichaudBaese-Berk, wasidenticalinExperiment2,butduringtest,canonicalGermanwordsthatoverlapped
UniversityofOregon,USA
in onset with the accented words from training were presented as target words (e.g.,
HelmutSpiekermann,
WestfälischeWilhelms-Universität Palme—“palm tree” overlapped in onset with the training word ∗Palken) rather than
Münster,Germany accented words. This time, no training effect was observed; recognition of canonical
*Correspondence: word forms was not affected by having learned the accent. Therefore, accent learning
Ann-KathrinGrohe
wasonlyvisiblewhentheaccentedtesttokensinExperiment1,whichwerenotincluded
[email protected]
inthetestofExperiment2,possessedsufficientsaliencebasedontheobjectivecriterion
Specialtysection: “accent.”Theseeffectswerenotmodifiedbythesubjectivecriterionofsaliencefromthe
Thisarticlewassubmittedto
trainingmodality.
LanguageSciences,
asectionofthejournal
Keywords:nativeaccents,adaptation,eye-tracking,salience
FrontiersinPsychology
Received:11February2016
Accepted:24May2016 INTRODUCTION
Published:07June2016
Citation: Languages typically consist of a number of regional dialects—that is, native accents. In the
GroheA-KandWeberA(2016)The southwesternGermanstateofBaden-Württemberg,forexample,onedoesnothavetotravelvery
PenefitofSalience:SalientAccented,
fartoencountervariousnativeaccents,asSpiekermanndocumentedin2008(Spiekermann,2008).
butNotUnaccentedWordsReveal
Thisvariationcanposeaproblemfornon-locals.Whennon-localshearanativeaccentforthefirst
AccentAdaptationEffects.
Front.Psychol.7:864. time,theyoftendonotunderstandwhatisbeingsaidaseasilyasdolocals,whoareexperiencedwith
doi:10.3389/fpsyg.2016.00864 theregionalvarieties.Recentstudieshaveindeedshownthatlistenersprocessaccentsintheirnative
FrontiersinPsychology|www.frontiersin.org 1 June2016|Volume7|Article864
GroheandWeber SalienceinAccentAdaptation
language more easily when they are familiar with the accents animacydecisiontask.Decisiontimesinthesecondanimacytask
than when they are unfamiliar with them (e.g., Adank et al., werenotfasterthanintheoriginaltask.
2009). Adaptation by non-locals to native accents is, however, Short-term adaptation was, however, found by Trude and
possible. Adaptation has been found for longer periods of Brown-Schmidt(2012).Participantswerefirsttrainedonanative
exposure to a novel accent (Evans and Iverson, 2007), but it English accent and then tested in an eye-tracking paradigm.
can even be observed after just 4 min of listening to a new During training, participants listened to scripted dialogs with
accent (Trude and Brown-Schmidt, 2012). This is also true an accented male speaker and an unaccented female speaker.
for second language (L2) learners. Producing a new accent for The male speaker raised the /æ/ before /g/ to [eI], i.e., bag was
only 7 min can facilitate subsequent accent understanding for pronounced/beIg/.Duringtest,targetwordswereeitherspoken
L2 learners, even more so than listening to the accent does bythemaleorthefemalespeaker.Whenback,awordunaffected
(Grohe and Weber, in press). The act of speech production by the accent, was the target and bag the competitor, bag was
arguably makes an accent more salient than listening to that ruled out more quickly as a candidate word for trials with the
accent does. Can the advantage of production experience also malespeakerthanitwaswiththefemalespeaker.Whenbake,a
be observed in a listener’s native language (L1)? Next to signal wordwith/eI/initscanonicalform,wasthetargetandbagacted
modality (production and listening), salience can also emerge as competitor, bake was fixated less often when it was spoken
from concrete properties of the speech signal itself. Acoustic bythemale.Thiseffect,however,waslessstronglypronounced,
distinctiveness of a speech signal can enhance its salience (e.g., i.e., competitor inclusion was more difficult than competitor
Cho and Feldman, 2013). The present study used a training- exclusion.
test paradigm in German in which salience was induced by Specific properties of the tested accents could account for
either a subjective or an objective criterion and looked at the themissingeffectsofadaptationinthestudiesdiscussedabove,
role of salience in native accent adaptation. The subjective but,moreimportantly,speaker-specificitycanexplainittoo.In
criterionwasimplementedthroughtwodifferentaccenttrainings contrasttoAdankandMcQueen,whohaddifferentspeakerswith
(production and listening) and the objective criterion through thesameaccentduringexposureandtestanddidnotfindaccent
the featuring of either accented (Experiment 1) or canonical adaptation, Trude and Brown-Schmidt used the same accented
(Experiment 2) target words during test in an eye-tracking speaker in both of their two experimental phases. Short-term
study. adaptation to native-accented speech may therefore be rather
speaker-specific.Thisproblemhasalsobeenaddressedinstudies
Adaptation to Native Accents onforeignaccentadaptation,withmixedresults.Usingatraining
Familiaritywithanativeaccentfacilitatesaccentprocessing.For testparadigm,BradlowandBent(2008)foundthatgeneralization
example,listenerswithextensiveexperiencewiththeNewYork of accent learning (Chinese-accented English) to new voices is
City(NYC)Englishaccentshowgreaterprimingeffectsforwords onlypossibleifthelistenerisexposedtomultiplespeakersduring
withtheNYC-English-typicalfinalr-droppingthanlistenerswith training(forsimilarfindingsseealsoSidarasetal.,2009).Kraljic
limitedexperience(SumnerandSamuel,2009).Similarly,Adank and Samuel (2007), on the other hand, found with L1 listeners
et al. (2009) found that only listeners who were familiar with that the nature of the tested material has an effect on whether
bothSouthernStandardBritishEnglish(SSBE)andGlaswegian perceptuallearningcangeneralizetonewspeakers.Theyfound
English(GE)showedequalperformanceonbothaccenttypesin generalizationeffectsforplosivesbutnotforfricatives.
a sentence verification task. The familiarity advantage probably
results from adaptation processes, as demonstrated by Evans Adaptation with Production
andIverson(2007).Intheirstudy,studentswhowereoriginally Speakerspecificityraisesthequestionofwhetherthereisaway
from Northern England adapted to SSBE over the course of tomaketrainingmoreeffective,i.e.,allowingforgeneralization
theiruniversitystudiesinSouthernEngland,asshownthrough acrossspeakers,andpotentiallyrenderingcompetitorinclusion
productionandcomprehensiontasks.Processingadvantagesfor morerobust.Thismightbepossiblethroughproductiontraining.
participants’ own accents over an unfamiliar accent were also InarecentstudybyGroheandWeber(inpress),theproduction
found for French listeners (Floccia et al., 2006). In a lexical of a foreign accent in participants’ L2 promoted adaptation to
decisiontask,reactiontimestoitemsinlongsentenceswerefaster that accent in a subsequent lexical decision task. Participants
when sentences were presented in the participants’ own accent first either listened to an English short story that featured the
(Northeastern French) than when they were presented in the replacement of all dental fricatives (“th”) with /t/ (e.g., theft
unfamiliar Southern French accent. Furthermore, participants pronounced as ∗ teft), or they read the same story aloud with
did not adapt to the unfamiliar accent during the course of thesamesubstitutions.Acontrolgrouphadnoaccenttraining.
the experiment (see also Floccia et al., 2009). Additionally, Afterwards,allparticipantscompletedalexicaldecisiontaskon
Adank and McQueen (2007) found no short-term adaption in wordswiththeth-substitutions.Theproductiongroupaccepted
a study with regionally-accented Dutch. In their study, Dutch the accented words significantly more quickly than the control
participantswhowerenotfamiliarwiththeFlemishaccenthadto group did. The listening group produced only a weak training
makeanimacydecisionsonsinglewordsspokenbytwodifferent effect.WhenthesameexperimentwasrunwithL1participants,
speakers,onewithaFlemishaccentandonewiththesameaccent notrainingeffectwasobserved.Referringtospeakereffects,L1
as the participants. Then, participants were exposed to another participantsintheproductiongroupproducedthecriticalaccent
speaker with the Flemish accent before having to repeat the marker,buttheywerelisteningtoanL2speakerinthetestphase.
FrontiersinPsychology|www.frontiersin.org 2 June2016|Volume7|Article864
GroheandWeber SalienceinAccentAdaptation
According to Pickering and Garrod (2013), listeners are more phase. In a subsequent word recognition task, there was an
likely to refer to their own previous production experience if it advantageforDutch-accentedwords.
ishighlysimilartothespeakertheyarelisteningto(e.g.,interms Second,factorsbeyondlinguisticorstructuralpropertiesmay
ofsex,L1background,dialect).Lesssimilarityleadslistenersto also affect salience. For the case of dialect accommodation,
drawontheirexperiencewithothers’speech.SinceonlytheL2 Kerswill and Williams (2002) suggest intensity of dialect
participantsinGroheandWeberhadthesameL1backgroundas contact as one of several factors. Considering the findings on
therecordedtestspeakers,speaker-listenersimilaritywassmaller production effects on accent adaptation, we can extend the
forL1participantsthanforL2participants. list of extra-linguistic factors toward cognitive mechanisms by
Facilitatory effects of producing an accent were also found introducing accent learning modality (production vs. listening)
inanaccentimitationstudywithL1speakersofDutch(Adank asanadditionalfactor.Severalstudieshavefoundanadvantage
et al., 2010). Baese-Berk and Samuel (2016), however, found of production over listening for dialect accommodation; this
that imitating a newly learned L2 sound can even inhibit has been named the production effect. It predicts that overt
learning.Intheirstudy,participantshadtoimitatesoundsfrom production facilitates word recollection when compared with
a sound continuum ranging from /s„a/–/ra/, which is arguably studying a word silently (MacLeod et al., 2010) and also when
difficult for speakers to imitate correctly. A potential acoustic compared with listening to others producing a word aloud
discrepancy between the sound prompt that was presented (MacLeod, 2011). It has been suggested that produced words
and the participants’ productions may therefore have inhibited are more easily recalled because they are more distinctive and
learning effects. A recent discrimination study with Danish therefore more salient. Distinctiveness results from listeners
vowels (Kartushina et al., 2015) supports this interpretation. focusingmoreontheirownthanonothers’productions,which
In that study, production accuracy was increased by concrete are,inthesenseoftheembodimenthypothesis(foranoverview:
feedbackonproductions,whichinturnresultedinbettersound Glenberg,2010),moreembodiedthanothers’productions.
discriminationperformanceafterproductionthanafterlistening Salience, as described above, has been further specified in
training. sociolinguistic research. Referring to Schirmunski (1928) and
Trudgill (1986), Auer et al. (1998) differentiate objective and
subjectivecriteriaofsalience.Forexample,articulatorydistance
Salience in Adaptation
is described as an objective criterion and perceptual distance as
We now turn to the concept of salience, which can potentially
its subjective counterpart. The two relate in that articulatory
explainboththeresultsofaccentadaptationandtheadvantage
distance describes the magnitude which a linguistic token
of produced compared to listened-to tokens. Salience has been
deviates acoustically from the canonical realization, whereas
generally defined as “the property of a linguistic item or
perceptual distance describes which way a listener perceives
feature that makes it in some way perceptually and cognitively
this distance. Based on this information, we can conclude that
prominent”(KerswillandWilliams,2002,p.81).Animportant
subjectivecriteriaincreasethesalienceofastimulus,forexample,
question,however,iswhatexactlymakesalinguisticitemsalient.
duetoregularpractice,andtheresultingcognitivepre-activation.
First, sociolinguistic research on salience suggests that an
Objective criteria refer to properties of a stimulus that itself
accent can increase a word’s salience. As suggested by Trudgill
attractsattentionbecauseofitsdistinct,physicalcharacteristics.
(1986), the phonetic difference between two variant forms
Underthisview,theproductioneffectreliesonthepresenceof
affects their salience; the greater the difference, the more a
an objective criterion. A self-produced word can be physically
dialect speaker is aware of it. Phonetic distance can also
more distinct compared to a word read silently or a word that
be considered within the framework of distinctiveness which
is produced by others because these words were only tested in
assumes that isolated, i.e., distinct, words are more salient
mixedlists,i.e.,oneparticipanthadtolistento/silentlyreadand
than others during encoding—provoking additional processing
producewordsinthesamesession.
and, therefore, better memory (McDaniel and Geraci, 2006,
Insummary,priorresearchhasshownthatnativeaccentsare
for a review). Geraci and Manzano (2010), for example, had
moreeasilyprocessediftheyarefamiliartoalistenerthanifthey
participants study a list of semantically related words that also
arenew.Short-termadaptationtonativeaccentsispossible,and
included a few semantically unrelated, i.e., distinct, words. In
productionalonecanpositivelyaffectforeignaccentlearning,at
ensuing tests,more unrelated thanrelatedwords were recalled.
leastinL2learners.Bothrobustaccentadaptationandtherole
Accordingly,Siegel(2010)claimsthatsaliencerequiresalistener
of production in accent adaptation may be related to salience.
to notice a contrast between two linguistic tokens. In terms
Theroleandconcretenatureofsalienceinlearningaccentedvs.
of phonetic variability, words that carry an accent are distinct
canonicalwordsthroughdifferentformsoftraining,however,is
from their unaccented counterparts and bear greater salience.
notyetclear.
Therefore, they can be learned more easily than unaccented
words1. This was tested in a different memory study (Cho and
Present Study
Feldman,2013).L1EnglishparticipantslistenedtoeitherDutch-
Thepresentstudytakesacloserlookatthisissuebyinvestigating
accented or native-accented English words during a training
subjective and objective criteria of salience separately, using
modality and accent as criteria. In an exposure-test paradigm,
1Forexample,intheiraccountonsocialweighinginspeechperception,Sumner
German participants first underwent native accent training
etal.(2013),predictbettermemoryforaccentedwords—butonlyiftheaccentis
sociallyprestigious. before adaptation was tested by a printed word eye-tracking
FrontiersinPsychology|www.frontiersin.org 3 June2016|Volume7|Article864
GroheandWeber SalienceinAccentAdaptation
task. A subjective criterion was established by having two is applied by default as soon as the auditory input potentially
different types of training (production and listening), while the matches the accent, i.e., from initial /p/ presentation onward.
objective criterion featured accented vs. canonical test words. When, in an eye-tracking experiment, the display includes
Accentedtestwordshadtheirinitialvoicedbilabialorvelarstop PALME and BALKEN and ∗Palken is the auditory target, both
devoiced.InExperiment1,accentedwords(∗PalkenforBalken— PALME and BALKEN should be fixated from word onset until
“beam”)werepresentedduringtrainingandtest.InExperiment disambiguation (including /pal/). Only after disambiguation
2, the same accented words were presented during training should BALKEN be fixated more often than PALME. If the
(∗PalkenforBalken),buttargetitemsduringtestwerecanonical accent rule is not learned robustly enough, the candidates that
words that overlapped in onset with the trained accented word requiretherule(BALKEN)haveaweakeractivationthanthose
(Palme—“palm tree”). Old word pairs from the training phase thatdonotrequiretherule(PALME).Consequently,duringthe
as well as new word pairs that had not been included in the overlapping word portion, PALME will still be more strongly
traininglistweretested.Thismanipulationwasincludedtotest activatedthanBALKEN,andBALKENwillonlybepreferredafter
generalizationoflearning,i.e.,whethertheaccentisonlylearned disambiguation.
fortrainedwordsoralsofornewaccentedwords. Successful recognition of a canonical word (Palme), as in
A subjective criterion of salience was tested by comparing Experiment2,doesnotrequireaccentruleapplication.However,
effects of individuals’ accent productions with that of listening successfulaccentlearningshouldresultinincreasedcompetitor
during training. In contrast to MacLeod (2011), the current activation of words with a /b/ in initial position. This increase
studydidnotmanipulatetrainingmodalityinmixedlistswithin in competitor activation might adversely affect canonical target
participants but rather between participants. This permits the activation. The rule should be applied by default as soon as
comparison of the magnitude of salience based on a subjective the auditory input potentially matches the accent, i.e., also
criterion of the production modality with that of the listening when Palme is presented. Having PALME and BALKEN on
modality. Individual participants are exclusively trained with the visual display, both words should be equally fixated during
one modality. If producing an item in fact constitutes a /pal/. Only after disambiguation should PALME be preferred.
subjective criterion for salience compared to listening to an The same predictions as above emerge if the accent rule is
item,productiontrainingwiththatitemwouldresultingreater not learned strongly enough—the candidates that require the
saliencethanlisteningtraining. rule(BALKEN)areactivatedlessstronglythancanonicalwords
An objective criterion of salience, on the other hand, was (PALME).Consequently,PALMEwillbemorestronglyactivated
manipulated by the presence of both accented and canonical thanBALKENevenfromthebeginningofwordprocessing.
testtokens.InExperiment1,thepresentationofaccentedwords The accent in the present study is an artificial accent that
assigned salience to the test tokens due to their great degree of centersononespecificphonologicalaccentmarkerandtherefore
inherentdistinctiveness.Effectsofaccentasanobjectivecriterion must be differentiated from a dialect. Participants and pre-
have been previously shown (Cho and Feldman, 2013), but recorded speakers are not L2 speakers, and all used standard
withamemoryexperimentinwhichgeneralizationeffectswere Germanpronunciationintheexperimentalcontext.“Standard”
not examined. In the learning phase, the accented words were here means that the pre-recorded speakers did not have a
embeddedinalistoffillertargetwordsthatfeaturednoparticular noticeabledialectthatcouldallocatethemtoaspecificregionin
accentmarker.Thismadetheaccentedwordsdistinctfromthe Germany, and the participants’ speech did not include specific
fillers. Contrarily, in the present Experiment 2, the canonical local (e.g., Swabian) accent properties during the experiment.
testwordswereexpectedtobelesssalient.Experiment2tested The tested accent affected German stop consonants and has,
whether the salience inherent in the learned accent can modify to our knowledge, not been documented as an existing native
the processing of words that include the accent target sound accentofGerman.Itreferstothelenis/fortis-contrastinGerman
in their canonical form. A difference in learning effects can be bilabialandvelarplosives(/b,p/and/g,k/).InStandardGerman,
reflected in the activation differences of canonical target words fortisplosivesarealwaysaspiratedinwordinitialposition,while
starting with the manipulated accent’s target sound. Learning lenis plosives are never aspirated (Jessen, 1998; Kleiner and
thatBalkenispronouncedas∗Palkenpotentiallyincreaseslexical Knöbl, 2015). Our accent neutralized this contrast, i.e., lenis
competitionforthecanonicalPalme,which,inturn,slowsdown velar and bilabial stops were aspirated (/g/ pronounced [kh]
recognitionofPalme.ThisisbasedonTrudeandBrown-Schmidt and/b/pronounced[ph]:Gitterpronounced∗Kitter,andBalken
(2012),whofoundthataccentlearningcanimplytheinclusionof pronounced∗Palken).Theaccentedsoundwasalwaysaspirated,
newcompetitors.Inthepresentstudy,Balkencouldbeincluded similartothecanonicallyfortisstops.Forsimplification,werefer
asanewcompetitorforPalmeaftertraining,resultinginfewer toaspirated,fortisplosives(Palme)as“voiceless”andtothelenis
targetlookstoPalme. plosiveswiththeadditionalaspirationintheaccentedversionas
The pattern of target and competitor activation is especially “devoiced”(∗Palken).
importantduringthesegmentaloverlapoftargetandcompetitor We opted for an accent with a target sound that is
words.Referringtotheprinciplesofanabstractmentallexicon, included in the German sound inventory (Kohler, 1999). This
we assume that accent learning is based on learning pre-lexical makes it easy to produce for native German participants and
rules. When hearing ∗Palken in Experiment 1, successful word promises relatively stable acoustic properties of the target
recognitionrequirestheapplicationofaspecificaccentrule(/b/ sounds across participants. The accent under investigation
→/p/).Iftheaccentruleislearnedrobustlyduringtraining,it has to be differentiated from middle-Bavarian dialects where
FrontiersinPsychology|www.frontiersin.org 4 June2016|Volume7|Article864
GroheandWeber SalienceinAccentAdaptation
bilabial, alveolar, or velar plosives are not realized with not suffer from any hearing disorders and had normal or
an aspiration contrast before /r, l, n, m/; they are always corrected-to-normalvision.Twoparticipantswereexcludeddue
voiceless and unaspirated and therefore neutralize with their to unsuccessful calibration, resulting in the collection of data
leniscounterpart,e.g.,Preiselbeeren—“cranberries”pronounced from 72 participants (26 production group, 22 listening group,
∗Breiselbeeren (Moosmüller and Ringen, 2004). Likewise, in and24controlgroup).
AustrianGerman,thefortisplosives/p/and/t/arenotaspirated
(Siebsetal.,1969),e.g.,Pinsel—“brush”pronounced∗Binsel.In Materials
contrast,theaccentpresentedinthisstudyneutralizedallbilabial WordsduringtheTestPhase
plosives to [ph] and all velar plosives to [kh]. Since the accent Wepresented92wordquadrupletsduringtest,eachcontaining
tested in our study describes a voicing shift in the opposing four German nouns. Twenty-eight quadruplets were based on
directionofexistingnativeGermanaccents,wecanassumethat critical word pairs; 64 quadruplets were based on filler word
none of our participants had had experience with the accent. pairs.The28criticalwordpairswereeachcomposedofatarget
Thisensuredtheobservationofonlylaboratory-specifictraining wordwithaninitialvoicedstopandacompetitorwordstarting
effects. with the corresponding voiceless stop. Only target words were
Wepredictthataccenttrainingwillresultinaccentlearning presented auditorily during the experiment. Fourteen had a
effects. The training modality can determine the amount of bilabialonset(e.g.,targetBALKEN“beam”—competitorPALME
saliencebasedononesubjectivecriterion.Thiswouldbeinline “palm tree”), and 14 had a velar onset (e.g., target GITTER
with prior findings where producing rather than listening to a “grid”—competitor KITTEL “tunic”). We opted for plosives,
wordresultedinbettermemory(MacLeod,2011).Saliencethat becauseithasbeenshownthatperceptuallearningofplosivescan
relies on an objective criterion of the target token is expected generalize across speakers (Kraljic and Samuel, 2007), arguably
to affect looking patterns such that the learned accent affects because they contain hardly no talker-specific information in
processingofhighlysalient,accenteddevoicedtokensmorethan comparison to fricatives, for example. This was important
thatofcanonicalvoicelesstokens. because participants in the training groups were trained with a
differentvoicethanwasheardduringtest.Voicedstopsoccurred
EXPERIMENT 1 only in the initial position of target words. The initial stop
consonantwasalwaysfollowedbyavowel3.Apartfromtheinitial
Experiment 1 tested if salience can result from training as consonant, target and competitor overlapped in at least two
subjective criterion. Critical test words had a native accent and segments.Whenthetargetwordswerepresentedauditorily,the
wereassumedtobehighlysalientbasedontheobjectivecriterion initialvoicelessplosivesweredevoiced(Balkenwaspronounced
“accent.”Duringtraining,nativeGermanparticipantseitherread as ∗Palken), resulting in overlapping word onsets of target and
aloudorlistenedtosingleGermanwordsthathadtheirinitial/b, competitorforatleastthreesegments.Auditorywordswiththe
g/devoicedto/p,k/,e.g.,Balkenpronouncedas∗Palken,while nativeaccent(∗Palken)wereneverexistingwordsofGerman(see
thecontrolgrouphadnotraining. TableA1fortarget-competitorpairs).
Inthetestphaseoftheexperiment,participantsaccomplished Mean log-frequencies of target words were 0.61 per million
theprintedwordvariant(McQueenandViebahn,2007;Weber for velar stop words, 0.85 for bilabial stop words, and of
et al., 2007) of the eye-tracking task (Allopenna et al., 1998). competitors0.67forvelarstopwords,and0.88forbilabialstop
Participants saw four printed words in their canonical spelling words according to the CELEX word form dictionary (Baayen
(including a target, a competitor, and two distractors) on a et al., 1995). In order to form quadruplets, each of the 28
computer screen and were auditorily instructed to click on a criticaltarget-competitorpairswaspairedwithtwosemantically
target word while their eye movements were recorded. They unrelated distractor words that matched in frequency with the
listened to devoiced words (∗Palken) and had to click on a target-competitor pair. Distractor words never had a stop in
visual display that included the target word (BALKEN) and a initialpositionbutcouldcontainstopconsonantsinotherword
competitor (PALME). The competitor allows the investigation positions.
of whether activation of the devoiced token can be as strong The 64 filler word pairs also had a target and a competitor.
as activation of voiceless word forms without an accent. The There were 8 targets with initial /k/, 8 with initial /p/, 16 with
proportion of target fixations was measured and compared initial /t/, and 32 targets with no initial stop in onset position
betweentheproduction,listening,andcontrolgroups. (the “no-stop targets”). For the total of 32 targets with initial
/k/, /p/, and /t/, half of the competitor word onsets overlapped
Participants with the target word onset by at least three segments, and
Seventy-four native German speaking female students (19– half were phonologically unrelated. Two phonologically and
30 years, mean age = 23.8, SD = 2.7; 5 left-handed) from
SouthernGermandialect,thevariable“SouthernDialect”wastestedininitialdata
the University of Tübingen participated for a small monetary
analyses,resultinginnosignificanteffect.Participantswerenotselectedbasedon
remuneration.Onlywomenweretestedinordertoaccountfor dialectcompetence,anditwasnotcounter-balancedacrossconditions;therefore
the fact that the recordings were exclusively made by female thisfactorwasnotincludedinthemethodssection.
speakers. German was their only mother tongue2, they did 3InsomevarietiesofGerman,speakerstendtodevoiceinitialvoicelessstopswhen
theyarefollowedbyaliquid(e.g.,grillencanbepronouncedaskrillen).Byalways
2Fiftyparticipantsindicateddialectfamiliarity(42specificallywithaSouthern havingavowelfollowingtheinitialconsonant,potentialpreviousexperiencewith
German dialect, mostly Swabian). As most dialect speakers had exposure to a theaccentwasavoided.
FrontiersinPsychology|www.frontiersin.org 5 June2016|Volume7|Article864
GroheandWeber SalienceinAccentAdaptation
semantically unrelated distractors were added to each target- effects, as observed by Trude and Brown-Schmidt (2012), can
competitor pair. The 32 no-stop targets were paired with generalizetonewspeakersofthesamesex(bothfemale).
competitors that also did not have stops in initial position. Recordings were carried out in a sound proof cabin with
However, half of them overlapped in onset with the target an Olympus LS-11 sound recorder (44.1 kHz; 16-bit). Every
for at least two segments (e.g., target Seife “soap”—competitor targetword wasrecorded inthecontextofthecarriersentence
Seite “side/page”). There were four types of distractors for KlickenSiejetztauf —“Nowclickon.”Thedevoicedtargetwords
the 32 no-stop target-competitor pairs, each containing eight (∗Palken) and the voiceless words (Palme) were all recorded
distractor pairs. The bilabial (b/p), velar (g/k), and alveolar naturally, that is, the speakers were explicitly instructed to
(d/t) distractor pairs followed the same prerequisites as the pronounce the /b/ in Balken the way they would pronounce
corresponding critical target-competitor pairs. As they were the/p/inPalme.Thebestexemplarofthecarriersentencewas
not presented auditorily, stop+consonant onsets (e.g., Brosche chosen for each voice, and the duration of the carrier sentence
“brooch”—Prospekt“brochure/leaflet”)wereallowed.Thefourth wasmatchedbetweenbothvoices.Then,thecarriersentencewas
grouphadtwosemanticallyandphonologicallyunrelatedinitial addedtoeachtargetwordrecording.
soundsthatwereneverstops.
Altogether, the test included 92 critical trials and four Procedure
practicetrials.Halfofthecriticaltargetsandtheircorresponding AnSR-ResearchEyelink1000set-upwasusedfordatacollection
competitorshadbeenincludedintheprecedingtrainingphase, with a sampling rate of 1 kHz, and the experiment was
andhalfwerenewtoparticipants.Likewise,halfofthetargetsnot programmed with Experiment Builder (SR Research Ltd.,
startingwithastop(other-group)werenewtoparticipantsand Canada). Before the experiment started, the dominant eye of
half were familiar from the training. Every participant had her eachparticipantwasdetermined.Then,participantswereseated
ownexperimentallist,eachstartingwiththesamefourpractice in front of a computer screen and placed their chin on a chin
trials.Fillerandcriticaltrialswereequallydistributedacrossthe rest. They were brought to a position in which they could stay
lists,andacriticaltrialwasalwaysfollowedbyatleastonefiller comfortablyforthedurationoftheexperiment(∼30min).The
trial. There were never more than two old and not more than eye-trackerwascalibrated,thenwritteninstructionswereshown
fivenewtrialsinarow.Thevariousfillerconditionswereequally onthescreen.Participantshadasmuchtimeastheyneededto
distributedacrossthelists. readthemandinitiatedtheexperimentwithamouseclick.
Training
WordsduringtheTrainingPhase
The same training list was presented to each participant from
Seventy-two single words from the above described target-
the two training groups, while the control group received no
competitor pairs were used for training. They included half of
training. The training tokens were presented either visually
the devoiced targets(7 targets with bilabialonset, e.g., ∗Palken,
and auditorily (listening group) or visually only (production
and 7 targets with velar onset) and their respective competitor
group). The listening group first saw a fixation cross for 1000
(Palmefortarget∗Palken).Thedevoicedandvoicelessitemswere
ms, then the orthographic transcript of the training word
included twice in the training list, resulting in 28 devoiced and
(black Arial font, size 24) appeared in the center of the
28 voiceless trials. Additionally, 16 filler targets from the no-
screen. It corresponded to German spelling rules (BALKEN).
stoptargetswereincluded,resultingin72trainingtrialsintotal.
The initial letter that corresponded to the devoiced sound
Training trials with the same initial sound did not occur more
was colored red. Five hundred milliseconds later, the training
thantwiceinarow,andeachdevoiceditemwasfollowedbyat
word was played (∗Palken). Participants listened to the single
leastonecanonicalitem.
words (devoiced, voiceless, and fillers) through noise-canceling
headphones(SennheiserHD215II)andatthesametimefixated
RecordingsofTestandTrainingTokens the transcript on the screen. There was a 2000 ms inter-trial
All tokens used for training and test were recorded by two interval. Participants in the listening group were explicitly told
femalenativespeakersofStandardGermanwithoutanoticeable tolistenattentivelytothewordsandtobeawareofthespeaker’s
regional accent (speaker A: 23 years; speaker B: 28 years). The accent while fixating the orthographic version of the words.
speakersdidnotdiffersignificantlyinF0-rangeorspeakingrate, Witteman et al. (2013) have shown that a single word context
and the authors judged their pronunciation to be comparable. issufficientforlistenerstolearnanaccent.Intheircross-modal
Two different speakers were recorded to have different voices primingtaskparticipantswithoutpreviousaccentexposurehad
for both the training and test phases in the listening group. increasedprimingeffectsinthesecondhalfoftheexperimental
This permitted constant conditions across the training groups listcomparedtothefirsthalf.
because the production condition always involved a different The production group did not wear headphones during the
speakerduringthetraining(theparticipant)andthetest(thepre- training.Theysawthesameorthographictranscriptofthewords
recorded talker). Nevertheless, speaker-listener similarity was on the screen and had to read every single word out loud
granted by having participants and pre-recorded speakers with while their productions were recorded. Participants were asked
the same sex and L1 background in the test phase. Acoustic to pronounce the initial red letter “B” as /p/ and the initial red
differencesbetweenthetrainingandtesttokensareheldassmall letter “G” as /k/. Before every single trial, there was a fixation
as possible. Moreover, it can be tested whether speaker specific cross for 1000 ms, and then the word was shown for 3500 ms
FrontiersinPsychology|www.frontiersin.org 6 June2016|Volume7|Article864
GroheandWeber SalienceinAccentAdaptation
(accounting for the timing in the listening condition: 500 ms recordings.Onaverage,only0.7outof28criticaltrialswerenot
before the sound + 1000 ms mean word duration + 2000 ms devoiced as instructed. The proportion of correct clicks on the
pause). The next trial would then start. Between training and targetduringthetestphasewas94.3%(equallydistributedacross
test,thewritteninstructionsfortheeye-trackingtaskwereshown the training groups). However, five participants did not see the
on the screen. The production group had about 5s to put on mouse cursor due to technical problems. We extracted fixation
theirheadphones,andthelisteninggroupwaitedfor5suntilthe reports with the software Data Viewer (SR Research) and then
initiationofthetestphase.Overall,thetrainingtookabout7min further processed the data with the software R (R development
foreachparticipant. coreteam,2015).Thedatafromeachparticipant’sdominanteye
was used to determine the coordinates and timing of fixations.
TestPhase Only fixations that fell within a cell of one of the four interest
The test phase started with four practice trials. A fixation cross areas—target, competitor, and two distractors—were analyzed
preceded each trial for 1000 ms, then four printed words from (exclusion of 3.4% of the data). The interest areas each had a
a word quadruplet were shown on the screen for 500 ms. The cellsizeof472×344pixelswithadistanceof40pixelsbetween
words were printed in black Times New Roman font, size 34 vertical cells and 60 pixels between horizontal cells. Saccades
on a screen with a white background. Screen resolution was (20.8% of the data) were not added to fixation times. We then
1024 × 768 pixels, and the words were distributed across four analyzed the fixations for the four interest areas in 20-ms steps
different positions (256 × 576, 768 × 576, 256 × 192, and inatimewindowfrom0to1000msaftertargetwordonset.The
768 × 192 pixels), see Figure1. Display positions of target dependent variable “target” indicated whether in the respective
and competitor were pseudo-randomized, and the target never 20-msstepaparticipantfixatedthetarget;“competitor”indicated
appeared in the same display position more than three times a competitor fixation, and “distractor” a distractor fixation.
in a row. The mouse cursor (represented by a small circle) was This resulted in three variables with binary values. Target
locatedinthecenterofthescreenatthebeginningofeachtrial. and competitor fixation proportions were calculated with the
Thenthecarriersentence(about1300ms)followedbythetarget empirical logit function. The plotted fixation proportions were
word was played auditorily. Participants clicked on the target inspectedvisuallytodeterminethecriticaltimewindowtowhich
wordwiththemouse.Visually,participantssawthetargetword linearmixedeffectsregressionmodels(Baayenetal.,2008;Bates
in its correct spelling (BALKEN); auditorily, it had the same et al., 2015) were then applied. For each analysis we built an
accentaspresentedduringthetrainingphase(∗Palken).Asmall individual, best fitting model that included a particular choice
fixation circle appeared on the screen after every six trials to of fixed and random factors. Random effect structure included
initiateanautomaticdriftcorrectioninthecalibrationoftheeye- random intercepts for participants and items as well as those
tracker.Theexperimentconcludedwithalanguagebackground randomslopesthatsignificantlyimprovedthemodelfitastested
questionnairebasedontheLEAP-Q(Marianetal.,2007). bylikelihoodratiotests.Significanceoffactorswasindicatedby
t > |2|. Corresponding p-values, as reported in the text below,
Analysis and Results
weredeterminedwithlikelihoodratiotests.Asfixedeffects,we
Duringtraining,theproductiongroupperformedtheinstructed considered training (production vs. listening vs. no training),
accentquitewell.Theexperimenterdecidedbasedonperceptual familiarity (old, i.e., included in the training, vs. new), sound
judgments whether the critical training tokens were devoiced condition (bilabial vs. velar word initial sound), and speaker
as communicated in the instructions. Every instance where (speakerAvs.speakerB).Proportionoftargetfixationswasthe
the devoicing was not clearly perceivable was documented and dependentvariable.
subsequentlyvalidatedbymeansofacousticmeasurementsofthe
DescriptiveAnalysis
Notsurprisingly,thedistractorswereruledoutaspotentialtarget
words very early by the participants (from about 200 ms, see
Figure2), i.e., the fixation proportion of distractors decreased
veryquickly.Aslaunchingaprogrammedeyemovementusually
takes about 200 ms (e.g., Altmann and Kamide, 2004), word
processing is reflected in fixation proportions from this point
intimeon.Competitorswerepreferredovertargetsbyallthree
groupsfromabout280msonuntilabout700ms.Targetfixations
show that the two training groups started to fixate the target
more often than the control group from about 250 ms on. The
advantage of both training groups became more pronounced
andstartedbeingrobustfromabout350mson.Visually,there
wasnodifferencebetweentheproductionandlisteninggroups.
FIGURE1|ExampledisplayofatesttrialinExperiments1and2.In Statistical analyses were run for the time window 250–750
Experiment1,BALKENwasthetargetandPALMEthecompetitor.In ms because it included the whole process of target-competitor
Experiment2,PALMEwasthetarget,andBALKENthecompetitor.RETTER
disambiguation,andhereitbecameevidentthatthetwotraining
andVENTILweredistractorsinbothexperiments.
groupshadastableadvantageoftargetfixationscomparedtothe
FrontiersinPsychology|www.frontiersin.org 7 June2016|Volume7|Article864
GroheandWeber SalienceinAccentAdaptation
FIGURE2|Proportionsoftarget(BALKEN)andcompetitor(PALME)fixationsoftheproduction,thelistening,andthecontrolgroupinExperiment1.
Thebottomlinedescribesthemeannumberofdistractorfixationsofallthreegroups.
controlgroup.AscanbeseeninFigure2,theactualadvantage 8% from listening group data between 250–750 ms and re-ran
lastedmuchlonger—atleastuntil1000ms. the same model with the modified data. Despite the reduction
of the listening group’s target fixation data, training was still
StatisticalAnalysis significant(χ2 =6.2,p<0.05):thelisteninggroupstillfixated
First,amodelwithdatainthetimewindow0–200mswasrun. the target more often than the control group (ßtraining= 0.30,
Thistestedlookingbiasesbeforeprocessingoftheactualtarget SE = 0.15, t = 2.0), and there was no difference between the
wordbegan.Traininggroupwasthefixedeffect,andparticipant productionandlisteninggroups(t =0.2).Thissuggestsrobust
anditemwererandomintercepts.Therewasasignificanteffect differences between the control group and the two training
bytraining(χ2 =7.2,p<0.03);theresultsofthemixedmodel groups.
showthatthelisteninggrouphadmoretargetfixationsthanboth
the control group (ß = 0.39, SE = 0.15, t = 2.6) and the Discussion
training
productiongroup(ß =0.31,SE=0.15,t=2.1),hintingat Wefoundthataccentadaptationwaspossibleafterbothlistening
training
atargetbiasforthisgroup. andproductiontraining.Theproportionoftargetlooksinboth
The second model analyzed data between 250–750 ms. It training groups was higher than in the control group. The
included training group and sound condition as fixed effects listening group, however, fixated targets more often than the
as well as participant and item as random intercepts. Training other groups, even before actual target word processing began,
wassignificant(χ2 =10.7,p<0.005);boththelisteninggroup which might be argued to have affected the listening group
(ß = 0.48, SE = 0.15, t = 3.2) and the production group advantageinthesubsequentcriticaltimewindow.This,however,
training
(ß = 0.33, SE = 0.14, t = 2.3) fixated the target more can be excluded because the pattern of results persisted even
training
often than the control group. There was no difference between whenthefixationdataofthelisteninggroupinthelarger,later
the two training groups (t = 1.0). Furthermore, there was a time window were penalized for its advantage in the initial,
maineffectofsoundcondition(χ2 =7.5,p<0.007),resulting smaller time window. There were no effects of speaker, i.e.,
inmoretargetfixationsforbilabialthanvelaritems(ß = learningoccurredequallywellwithspeakerAandB.Themain
condition
0.35,SE=0.13,t=2.8).Duetothebiasforthelisteninggroup effect for sound condition may be related to specific sound
found from 0–200 ms, the critical time window was further properties but does not further affect the general pattern of
examined.Onaverage,from0–200mstheproportionoftarget results.Moreover,thesamepatternwasobservedforoldtokens
fixations was 8% higher for the listening group than for the fromthetrainingphaseandnewtokens,indicatinglearningofa
control group. To account for this early bias, we subtracted rulethatgeneralizestonewwords.
FrontiersinPsychology|www.frontiersin.org 8 June2016|Volume7|Article864
GroheandWeber SalienceinAccentAdaptation
Our results suggest accent learning for the production and duetocalibrationproblems,resultingin72participants(18–31
listening groups, with no difference between the two training years,mean=23.2,SD=3.2;14left-handed)whosuccessfully
groups. Thus, we found robust effects of accent training when completed the experiment. None of them suffered from any
testingsingleaccentedwords,hintingatagreateffectbytarget hearingdisorders,allhadnormalorcorrected-to-normalvision,
words’ accent as objective criterion of salience. Production and and German was their only mother tongue4. The participants
listening training seemingly do not differ from one another for wererandomlyassignedtooneofthethreeexperimentalgroups
L1intermsofsalience. (24production,24listening,and24controlgroup).
Experiment 1 provides evidence for successful accent
adaptationafterlisteningtoorproducinganaccent.However,the Methods and Material
canonicalcompetitors(Palme)wereactivatedforaverylongtime The training list was based on that of Experiment 1. However,
(untilabout700ms)beforethedevoicedtargetwordwasfixated devoiced(∗Palken)itemswerepresentedtwiceinarow(rather
moreoften.Thistimewindowcoverstheentireinitialportionof than just once), resulting in 100 training trials in total (twice
thewordbeforedisambiguation(averagedisambiguationpoint: the amount of training with the devoiced tokens compared
280 + 200 ms for launching the eye movement = 480 ms; to Experiment 1). Due to the greater amount of training, the
earliestdisambiguationpoint:150+200ms=350ms;longest trainingphasetook1minlonger.
disambiguationpoint:420+200ms=620ms)andevenlonger. During test, the same word quadruplets were presented on
Thissuggeststhat,despitesuccessfulaccentadaptation,canonical thescreen—92criticaltrialsand4practicetrialswiththesame
word forms still remained more easily accessible than accented propertiesasinExperiment1.However,therolesoftargetand
wordforms.Therewaspotentiallynotenoughaccentexposure competitor words were switched. Targets were now voiceless
fortheaccentedformstobeabletofullycompetewithcanonical tokens (Palme) in their canonical form, and competitors were
wordforms.Wesuggestthatifalearnedaccentweretobeable wordsthathaveavoicedonsetintheircanonicalform(Balken).
to have effects on the access of canonical, voiceless words with Auditorily,voicelesswordswerepresented(Palme)thatmatched
the same onset as the accent’s target form, a greater amount of in their onset with the target word on the screen (PALME).
trainingisrequired. Voiced tokens (BALKEN) that had been devoiced during the
Experiment2examineswhetheraccentlearningcanbestrong training(∗Palken),werevisuallypresentedcompetitors.Alltarget
enoughastoaffecttheprocessingofvoiceless,canonicalwords words had already been recorded in the recording session for
with double the amount of accent training. Successful accent Experiment1bythesamefemalespeakers.
learning could imply competition effects from words that were
previously not included as competitors. Thus, accent training Analysis and Results
haspotentiallyeffectsnotonlyforunderstandingaccentedword ThesameprocedureforanalysisasinExperiment1wasapplied.
forms, but accented forms can function as competitors and During training, the production group performed quite well
affect the recognition of canonical word forms. As opposed to in accomplishing the substitutions (mean: 0.8 errors out of 56
Experiment 1 where highly salient accented target words were devoicedwordtrials).Theaccuracyofclicksduringthetestphase
tested, in Experiment 2, we focused on test words that are was99.8%(equallydistributedacrosstraininggroups).Saccades
expected to have a much smaller degree of salience based on (17%ofthedata)andfixationsthatdidnotfallintooneofthe
theobjectivecriterion“accent,”i.e.,standardGermancanonical fourinterestareas(3%)wereremovedpriortoanalysis.
words.Trainingeffectsofdevoicedwords(∗Palken)weretested
on words that canonically start with the accent’s target sound DescriptiveAnalysis
(Palme). In order to increase the likelihood that accented Figure3 illustrates the proportions of target, competitor, and
forms could influence target recognition in their function as distractor fixations of the production, listening, and control
competitors, the training was doubled. If the accent is robustly groups. The distractors were ruled out from the beginning of
learned, we would expect fewer target fixations by the training word processing (200 ms) on, meaning that the proportion of
groupsthanwithoutaccenttraining. fixationsdecreased.Target(PALME)preferencestartedveryearly
(at about 250 ms), and competitors (BALKEN) were quickly
EXPERIMENT 2 ruledoutaspotentialtargetwords.Thecompetitorsstayedata
relatively stable level of activation from 200–400 ms, and then
Again, three participant groups were tested. The training fixationsdecreasednoticeably.Thisrepresentsapproximatelythe
involved the same tokens as in Experiment 1, but the amount intervalwheretargetandcompetitorstilloverlap(meanoverlap:
of training with the devoiced tokens was doubled. During test, 273 ms). Target fixations by training group did not differ from
participants did not hear the devoiced words (∗Palken) this one another from the beginning until the overall mean end of
time, but voiceless, canonical words (Palme), while seeing the word processing (measurements of the voiceless target words
same printed words on the display as the participants from resulted in a mean word duration of 767 ms). Disambiguation
Experiment1. between targets and competitor occurred relatively early, and
there was no clear advantage of one of the training groups in
Participants
4Fifty-threeparticipantsindicateddialectfamiliarity,50ofwhomhadexposureto
Seventy-eight female students from the University of Tübingen
aSouthernGermandialect(mostlySwabian).Thevariable“SouthernDialect”was
participatedformonetaryreimbursement.Sixhadtobeexcluded testedininitialdataanalyses,resultinginnosignificanteffect,asinExperiment1.
FrontiersinPsychology|www.frontiersin.org 9 June2016|Volume7|Article864
GroheandWeber SalienceinAccentAdaptation
FIGURE3|Proportionsoftarget(PALME)andcompetitor(BALKEN)fixationsoftheproduction,listening,andcontrolgroupsinExperiment2.The
bottomlinedescribesthemeannumberofdistractorfixationsofallthreegroups.
targetfixations.Statisticalanalyseswererunforthetimewindow Acoustic Analyses
from 250–750 ms, which included the entire disambiguation Pre-recorded target and training stimuli as well as the tokens
process between targets and competitors and parallels analyses producedbytheproductiongroupduringtrainingwereanalyzed
inExperiment1. acoustically. This tested if the difference between training and
test stimuli was too great for adaptation effects to be observed.
StatisticalAnalysis Particularly in the production group, the acoustic properties of
The baseline model for target fixations (0–200 ms) revealed theaccentedplosiveswerelikelytovaryindividually.Thestops
no significant effect by training (t < 1). Mixed effects models thatmarkthemanipulatedaccentwerefocusedonintheanalysis.
revealed no significant effect of any of the considered fixed Voiceonsettime(VOT)andburstintensity(relativetototalword
effects (all t < |1.3|) in the critical time window (250–750 ms). intensity)weremeasuredforeachtokenthatwaspartofanold
Auditorilypresentedvoicelesswords(Palme)thatstartwiththe criticalwordpair,i.e.,wordpairsthatwereincludedinboththe
same onset as the trained, devoiced words (∗Palken) triggered trainingandthetestphase.Onlyoldwordpairswereincludedin
strong target activation from the beginning of word processing analyses,becausetherewasnoreferencetothetrainingphasefor
on. There was no effect of learning, neither by the production newwords.Eachcriticalvoicelessword(Palme)anditsdevoiced
nor the listening group. In contrast to Experiment 1, the test pairedword(∗Palken)wasconsideredforanalysis.Bothinstances
words did not have the critical accent, but the voiceless paired weretakenasseparatereferencepointsinordertocalculatethe
words with the same sound onset as the devoiced, accented differencesoftherespectiveacousticpropertyvaluebetweenthe
words were tested. As the devoiced training words included a trainingandthetesttoken(Palme).Inthefollowing,wereferto
soundsubstitution,thequestionis,especiallyfortheproduction thePalme-Palmecomparisonasthevoicelesswordpairandthe
group: How much did the acoustic realizations of the devoiced ∗Palken-Palme comparison as the devoiced word pair. During
tokens encountered during training differ from those of the training, one word was presented several times (devoiced: four
voiceless tokens encountered during test? In other words, did times, voiceless: twice). This did not pose any problem for the
the participants’ own productions of the accent differ enough listeninggroupitemsbecausethesamerecordingwaspresented
fromtheproductionsofthetestspeakertopreventgeneralization severaltimes.Fortheproductiongroup,however,singletokens
across speakers? The missing training effect for both groups differed individually. This issue was solved by taking average
reinforces the question of effects of single tokens’ acoustic values.TwoVOT-andtwoburstintensitydifferencevalueswere
properties. Therefore, acoustic properties of both the training assignedtoeachcriticalwordforeachparticipant—onewiththe
materialsandthetestmaterialswereanalyzedinanextstep. valuesfromthevoicelesswordasareferencepoint(Palme)and
FrontiersinPsychology|www.frontiersin.org 10 June2016|Volume7|Article864