Phonetic detail is used to predict a word’s morphological ...

Phoneticdetailisusedtopredictaword’smorphologicalcomposition123

Meghan Clayardsab, M. Gareth Gaskellc and Sarah Hawkinsd* 4 5 6* Corresponding author: [email protected] 7a. Department of Linguistics, McGill University, Canada 8b. School of Communication Sciences and Disorders, McGill University, Canada 9 10c. Department of Psychology, University of York, UK 11 12d. Centre for Music and Science, Faculty of Music, University of Cambridge, UK 13 14 15 16Abstract17Aneye-trackingexperimenttestedthehypothesisthatlistenersusewithin-word18finephoneticdetailthatsystematicallyreflectsmorphologicalstructure,whenthe19phonemesareidentical(disindiscolour(trueprefix)vs.discover(pseudoprefix))20andwhentheydiffer(re-covervs.recover).Spokensentencepairs,identicalupto21atleastthecriticalword(e.g.I’dbesurprisediftheboysdiscolour/discoverit),were22cross-splicedattheprefix-stemboundarytoproducestimuliinwhichthecritical23syllable’sacousticseithermatchedormismatchedthesentencecontinuation.On24each trial listeners heard one sentence, and selected one of two photographs25depicting the pair. Matched and mismatched stimuli were heard in separate26sessions,at leastaweekapart.Matchedstimuli led tomore looks to the target27photograph overall and time-course analysis suggested this was true at the28earliest moments. We also observed stronger effects for earlier trials and the29effectstendedtoweakenoverthecourseoftheexperiment.Theseresultssuggest30thatnormalspeechperceptioninvolvescontinuouslymonitoringphoneticdetail,31and,whenitissystematicallyassociatedwithmeaning,usingittofacilitaterapid32understanding.33

Keywords 34perception, spoken language understanding, morphological structure, acoustics 35 36

Phonetic detail predicts morphological status

page2of44

1 Introduction37

There is increasing evidence that morphological differences in phonemically-38identicalwordsorpartsofwordsarereflectedintheirphoneticrealisation(e.g.39Hay,2003,benHedia&Plag,2017;PlagandbenHedia,2018;Rose,2017;Smith,40BakerandHawkins,2012;Sproat&Fujimura,1993;Seyfarthetal.2018;seealso41Strycharczuk,2019forareview).Assuchevidenceaccumulates,questionsremain42unanswered as to whether the observed patterns in production influence43listeners’ perceptual behavior. This paper addresses perception of a44morphologicalcontrast involvingprefixes(e.g.disindiscolour,re- inre-joined)45and matched non-prefix word-initial syllables that are either phonemically46identicaltotheprefix(e.g.dis-indiscover)orcontrastinthevowelphoneme(e.g.47re- in rejoiced).Both typesofmorphological contrastmanifest as twodifferent48acoustic-phoneticpatternsaffectingtheentireword-initialsyllableandbeyond,in49prosodically-controlledfluentlabspeech(Smith,Baker,&Hawkins,2012)andin50conversationalspeechinseveralregionalvarieties(Hay,Hawkins,Stuart-Smith,51Smith and Fromont, in prep). The question asked iswhether listeners use the52resultantacoustic-phoneticdistinctioninrealtimetofacilitatewordrecognition53in connected speech. If they do, thiswould indicate that they use the internal54acousticstructureof the first syllableofaprefixedorpseudo-prefixedword to55identify its stem, andby implication the lexical item’smorphological structure.56Whenthephonemesdonotdiffer,thiswouldamounttoidentificationofbound57morphemesfromveryfinedifferencesinthesyllable’sinternalacousticstructure,58andtheirusetopredictlexicalidentity.59

1.1 Phoneticsofmorphologicallycomplexwords60

By far the largest body of relevant experimental work examines temporal61relationships between acoustic segments or articulatory gestures in suffixed62words.Comparedwithphonemically-matchedorsimilarmonomorphemicwords,63suffixedwordshaveconsistentlydifferentphoneticcharacteristicsintheirsuffix,64intheirstemandatthemorphemeboundary.Althoughanumberofstudieshave65demonstratedsuchproperties,severalhaveconfoundednumberofmorphemes66with number of syllables and/or foot length (e.g. Kemps, Wurm, Ernestus,67Schreuder, & Baayen, 2005; Lehiste, 1972). The reported differences between68bimorphemicandmonomorphemicwordscannonethelessbeconsideredrobust69inthatseveralstudiesthatcircumventedtheseproblemsshowedthesametypeof70results(Sugahara&Turk,2009;Cho,2001;Seyfarthetal.,2018).Thesetemporal71relationshipshavealsobeenshowntoaffectthearticulatorygesturesofEnglish72/l/ at morpheme boundaries resulting in gradient spectral differences in /l/-73darkness (Sproat & Fujimura, 1993; Lee-Kim, Davidson, & Hwang, 2013;74Strycharczuk&Scobbie,2016,2017;Turton,2017;Mackenzieetal.,2018)75

Prefixeshave received less attention than suffixes, butOh andRedford (2012)76showdurationaldifferencesinnasal-nasalsequencesdependentonwhetherthe77sequenceincludesamorphologicalboundaryasinun-namedorawordboundary78as in fun name. Smith, Baker and Hawkins (2012) and (Hay et al., in prep)79document complex, systematic acoustic effects of prefix status for the initial80syllablesofwordpairssuchasdiscolourvs.discoverandmistypesvs.mistakes,in81whichthefirstmemberofeachpairbeginswithatrue(productive)morpheme82


page3of44

whereas the second member does not, despite having the same phoneme83sequence.Consequently,theinitialsyllablesofdiscoverandmistakesaretermed84pseudoprefixes. Prefixes of this type areparticularly interesting formodels of85speech perception because, if their distinctive phonetic detail 1 is processed86differentlyfromthatofpseudoprefixes,thiswouldsuggestthattheirproperties87are directly associated with their status as bound morphemes. In contrast,88perceptualstudiesexaminingthesametypeof issue forsuffixes typicallyshow89listeners’sensitivitytodifferencesinthestem,ratherthaninthesuffixitself,as90discussedbelow.91

Thedistinctionbetweentrueandpseudoprefixesisnotcompletelyclearcut,due92toanumberofinteractinginfluencesofquitedifferenttypes,alsodiscussedbelow.93However,itispossibletocontrolforsuchinfluences.Smith,BakerandHawkins94(2012)usedcriteriathatprovidedtightcontroloverboththetypeofwordandits95phonetic and semantic context. They followedWurm's (1997) strict semantic96criteria to select stimuluswords: inwords such asdiscolour andmistypes, the97initial syllables,dis-andmis-, are true prefixes because colour and typesmean98roughlytheoppositewhendis-ormis-areadded.Incontrast,thewordsdiscover99andmistakes are monomorphemic because cover and takes do not mean the100oppositeofdiscoverandmistakes.Furthermore,eachoftheprefixedwordschosen101hada lowerfrequencythanthefrequencyof itsstem,therebyconformingwith102Hay’s (2003) criterion (developed for suffixes) for a relatively strong and103unambiguousmorphemeboundary.104

Smith,BakerandHawkins(2012)elicitedsuchtrueandpseudoprefixpairsfrom105speakers of Standard Southern British English (SSBE) in fast, casually-spoken106scripteddialoguesinwhichtheprosodicandsegmentalstructureofthecritical107utterancesweretightlycontrolled.Acoustic-phoneticmeasuressupportedearlier108impressionisticclaims(e.g.Hawkins,2010;Ogdenetal.,2000;Whitley,citedby109Simpson,2005)thatthefirstsyllablesofthetrue-prefixedwordsconveyaheavier110beat in context due to small differences in the acoustic properties of their111componentsegments.As illustrated, forexample, inSmith,BakerandHawkins'112(2012)Figure1,whichshowsspectrogramsandphonologicaltreesformistimes113(trueprefix)andmistakes(pseudoprefix),oneveryreliableacousticdifferenceis114thedurationofaperiodicityfor[s]relativetothedurationofperiodicityof[ɪ]:the115[s]takesupamuchlargerproportionofthesyllableinpseudoprefixes.Another116isthatthesecondformantfrequencyof[ɪ]ishigherandclosertoF3inthetrue117prefix,suggestinglesscentralisation.Athirdisthatwhenavoicelessstopisinthe118onset of the second syllable of the word, its voice onset time (VOT) is long119following the true prefix, but short following the pseudo prefix. In sum, the120morphologicalstatusisreflectedinseveralphoneticcharacteristicsthataffectall121segments in the first part of the word. These differences create systematic122differencesintheoverallpatternofrelationshipsbetweentheacousticsegments123withinthefirstsyllable,termedhereitsinternalacousticstructure,aswellasat124thesyllablejunctureandthereafter.Theinternalstructureofthefirstsyllableis125suchthat,thoughbothtrueandpseudoprefixsyllablesaremetricallyweak,true126prefixes are more phonetically prominent—indeed, many dictionaries accord127some true prefixes a secondary stress, whereas pseudo prefixes are never128

1SeeAppendixAforwhatismeantby‘phoneticdetail’.


page4of44

accordedone.Thus,inanygivenspeechregister,thetrueprefixconveysaheavier129rhythmicbeatthanthepseudoprefix.130

Whilethesedifferencesarereasonablydistinctforanygivenspeechregister,there131canbe‘gradient’effectsincaseswheretheprefixstatusofawordisinflux(which132occursforavarietyofreasons(cf.Hayetal.,2005)andbetweendifferentspeech133registers and modes of data collection (Smith, 2012; Hay, 2018; Zuraw &134Peperkamp, 2015). Furthermore, while the majority of prefixes behave as135described above, a few do not (Plag, 2014). Words whose prefix status was136ambiguouswereexcludedfromthepresentstudy.137

Insummary,systematicphoneticmarkersof the internalcompositionofwords138areembeddedinthespeechsignalandsoarepotentiallyavailabletothelistener.139Thesekindsofcuesdifferfromotheracoustic-phoneticeffectssuchasthosedue140toassimilationofplaceofarticulationacrosswordboundaries,inthattheyoccur141word-internallyinarangeofspeechregisters,includingcareful,clearspeech,and142so are integral to the identity ofwords. The goal of this paper is todetermine143whetherlistenersareinfactsensitivetosuchsubtledistinctivepatternsanduse144themtobuildexpectationsaboutmorphemicstructureandhencewordidentity145astheyinterpretutterancesinrealtime.146

1.2 Perceptionofacousticcuestowordstructure147

Lexical identification experiments using gating tasks and cross-modal priming148(Davisetal.,2002),andeye-trackingwiththevisualworldparadigm(Salverdaet149al., 2003), show that listeners’ early perceptual responses are sensitive to150acoustic-phoneticdetailthatsignalswordboundaries.Theycontrastsyllablesthat151areeitherfollowedbyawordboundaryorarepartofalongerwordasincapand152captainorhamandhamster.Whenhearingsyllablessuchascap,listenerswere153morebiasedtowardsamonosyllabicinterpretationwhencaphadbeenspokenas154a monosyllabic word rather than as part of a polysyllabic word. While these155studiesconfirmtheimportanceofphoneticdetailtolexicalidentification,theydo156not examine influences due to morphological structure within words, and, as157summarized at the end of this section, they confound a number of linguistic158variableswhichdesignsusingthetruevspseudoprefixdistinctioncancontrol.159160Relatively littlework has examined perception ofmorphological structure (i.e.161word-internal junctures). Blazej & Cohen-Goldberg (2014) tested whether the162effectofnumberofsyllablesfoundforthehamversushamsterstudiesextended163towordswhicharealsomulti-morphemicbyvirtueofcontainingsuffixese.g.clue164andclueless.Theyfoundthesamepatternasstudiesthatexamineshorterwords165embedded in longer monomorphemic words: listeners anticipated the longer166wordafterhearingashorterfirstsyllable.ApairofsimilarstudiesbyKempsand167colleagues(Kempsetal.,2005a;2005b)usinglexicaldecisionandamorpheme168decisiontask(singularorplural)foundcompatibleresultsforembeddedstemsin169pluralsinDutchandforcomparatives(e.g.stronger)andagentnouns(e.g.worker)170in Dutch and English. However because the stimuli in these studies compared171monosyllabic,mono-morphemicwordswithpolysyllabic,poly-morphemicwords,172it is impossibletotellwhethertheir listenersweresimplyanticipatingalonger173word (an ability previously demonstrated for mono-morphemic words), or174


page5of44

whethertheywereanticipatingthepoly-morphemicstructureofthelongerword,175orboth.176177As described in the next section, the present study of true and pseudo prefix178perceptioncircumventsmostof theseproblems.Furthermore, there is intrinsic179theoretical interest in distinguishing boundaries within words from those180betweenwords,andphoneticsupportformakingthatdistinctionfromworkon181prefixesun-andin-(Oh&Redford,2012).1821831.3 Thepresentstudy184

The present paper uses eye-tracking in a visual world paradigm to test our185hypothesis that listeners can use the internal acoustic structure of the initial186syllableofaprefixedorpseudoprefixedwordtopredictmorphologicalstructure187that itself predicts word identity. The focus on prefixes allows questions of188perceptual sensitivity to the morphological status of a syllable to be assessed189without the confounds of morphological complexity, polysyllabicity, and word190lengthwhichcharacterizethestudiescitedinSection1.2.Wecompareunitsthat191arecomparableexceptinmorphologicalcomplexity:allwordsarepolysyllabic;all192boundariesofinterestareword-internal;andnofirstsyllablehasanindependent193lexical meaning—pseudo prefixes, comparable to cap in captain, convey no194meaningindependentoftherestoftheword,whileprefixesarenotindependent195lexicalitems,sothoughtheyconveyameaning,itisonlyproperlyinterpretablein196thecontextofthemeaningoftherestoftheword.Furthermore,unlikethecritical197syllablesinthecapandcaptainstudies,ourcriticalsyllablesaremetricallyweak198(lackprimarystress);weaksyllablesareoftenthoughttoplayasubordinaterole199inlexicalidentification(e.g.Cutler&Butterfield,1992).200201Our study also differs from studies that have shown effects of phonetic (sub-202phonemicorwithincategory)detailonlexicalaccess(e.g.McMurray,Tanenhaus,203Aslin&Spivey,2003;Dahan,Magnuson,Tanenhaus,&Hogan,2001)inthatour204phonetic detail contributesprimarily to rhythmic andnot segmental structure.205Furthermore,unlikepreviousstudies that testedphoneticdetailusingminimal206pairs (e.g.McMurrayet al., 2003), thepresent study contrastsword sequences207mostofwhosesecondandlatersyllablesarenotminimalpairs,solistenersinour208experimentdonotneedtousethefinedetailofunstressedmis-ordis-atall in209order to distinguish thewords or the sentencemeanings. Thus ours is a very210stringenttestoftheperceptualsalienceofphoneticdetail:thecuesareinweak211syllables and they are followed very swiftly by much clearer disambiguating212evidence.Ifwefindevidencesuggestingthatthesecuesareuseddespitetheirnot213beingessentialtothetask,thenwehaveverystrongevidenceofthepervasiverole214ofphoneticdetail,andrhythmicdetailinparticularinspokenwordrecognition.215Furthermore,suchfindingswouldstrengthentheevidencethatlistenersextract216clues tomany levels of linguistic structure from the finephonetic detail in the217signal.218219Weusedfourtypesofprefixes:mis-anddis-asalreadydiscussed,andtheprefixes220re-andex-asinre-peel/repealandex-trampoliner/extravagance.Thesyllablesre-221and ex- follow the same patterns as dis- and mis-, i.e. the true prefixes are222


page6of44

rhythmically stronger, but in this case syllabic reduction in the pseudo prefix223happenstocrossaphonemecategoryboundary,whereasthesametypeofsyllable224reductiondoesnotproduceacategorychangeinthedis-mis-set.Consequently,225forre-andex-,thevowelphonemesinthecriticalsyllablesdiffer:re-peel/ri:ʹpi:l/226but repeal /rɪʹpi:l/ or /rəʹpi:l/; ex-trampoliner /ɛksʹtrampəli:nə/ but227/ɪksʹtravəɡəns/or /əksʹtravəɡəns/in SSBE, the regional varietyusedhere (see228Smithetal.,2012formoreexplanation).Thuswhileallprefixesdifferrhythmically229fromthepseudoprefixes, thesyllables inre-andex-differsegmentallyaswell.230This distinction is represented in our design as the independent variable231PhonemeChange, with re-ex- changing vowel phoneme, and dis-mis- not. As232segmentaldifferencesareuncontroversiallypartofthelexicalrepresentation,we233canpredictthatthesegmentalcuestomorphologicalstructurewillbepickedup234inspokenwordrecognition.Themis-anddis-stimuliareamorestringenttestof235the hypothesis that non-segmental phonetic detail is important for identifying236morphological structure because they share the same first four phonemic237segments.However,becausediscriminationofprefixeshasnotbeentestedusing238eye-trackingbefore,andtheirrelativelyabstractmeaningsnecessitatetheuseof239relativelycomplexvisualstimuli,there-andex-stimuliprovideacheckthatour240methods are sensitive enough. Then the question is whether nonphonemic241morphologicaldivergencewillalsobeexploited.Ifso,howstrongisitcompared242withthephonemiceffectanddoesithavethesametimecourse?243244In sum, our experiment was designed to test whether listeners exploit subtle245acoustic cues to morphological structure. In particular, do they identify246morphologicalstructure,eventotheextentofpredictingthattheyarehearinga247morphologically-complexwordbeforetheyhaveheardthestem,ingoodlistening248conditionswhenthetaskdoesnotdemandit?2492502 Methods251

2.1 Design252

We used cross-spliced spoken sentences to manipulate whether the acoustic253informationincriticalsyllables(i.e.inthetrueorpseudoprefix)wasconsistent254with themorphological structure of the rest of theword.We presented these255stimuli to participants and asked them to choose between two pictures, one256representingasituationwhosedescriptionincludedthetrueprefixedwordand257the other a situationwhose description included thematchedpseudoprefixed258word(seeFigure1forillustration).259

260


page7of44

261Figure 1. Construction of match and mismatch stimuli illustrating one pair of262sentences. SubscriptT indicates a portion from the original utterance that263containeda trueprefixedword (e.g.displaces,discolour). SubscriptP indicatesa264portionfromtheutteranceoriginallycontainingthepseudoprefixedword(e.g.265displays,discover).Subscriptnumbersrefertodifferentsentencerecordings.For266eachsentencethetargetandcompetitorimagesaregiven.267

Thewordsinapairofsuchsentenceswereidenticaluptothecriticalsyllable.If268perceptualbehaviourisinfluencedbytheacousticinformationavailabletosignal269truevs.pseudoprefixeswhileitisbeingheard,thenwewouldexpectlistenersto270be delayed in correctly identifying the target word (and therefore the target271image) when that acoustic information does not match the morphological272structureof the targetword/image.Ourmainanalysis therefore compared the273looks to target images (defined as the image consistent with the post-splice274continuation of the spoken word and sentence) on trials in which the critical275syllablecontainedacoustic information thateithermatchedormismatched the276targetimage(factorMatch).Forexample,inFigure1,theimagecorrespondingto277thetargetforAswandisplaceswaterwhenitlandsistheimageofaswanlanding278on water, regardless of the prefix status of the cross-spliced critical syllable.279Similarly,thecompetitoristheimageconsistentwiththesentence’spair,inthis280casethetwoswans(seeFigure1).281

Wealsoconsideredwhethertheeffectofthecriticalsyllable(Match)dependedon282a number of factors. The factors of the design are summarized in Table 1. As283discussedabove,wetestedifMatchdependedonwhethertheprefixmanipulation284entailed a phoneme change as in re- or ex-, or not as inmis- or dis- (factor285PhonemeChange).286


page8of44

Table1:Summaryoffactorsandtheirlevelsinthedesign.287

Factor Levels NestingMatch match:acousticinformationincriticalsyllable

matchesthemorphologicalstructureofthetargetword/imagemismatch:acousticinformationincriticalsyllabledoesn’tmatchthemorphologicalstructureofthetargetword/image

withinsubjectsanditems

InteractionswithMatch PhonemeChange dis-/mis-

re-/ex-withinsubjectsbetweenitems

Group M1:matchonsession1/mismatchsession2M2:mismatchonsession1/matchonsession2

betweensubjectswithinitems

TrialNumber continuousvariablefrom1to99 withinsubjectsanditems2

PrefixStatus true:criticalsyllablesplicedfromatrueprefixedwordpseudo:criticalsyllablesplicedfromapseudoprefixedword

withinsubjectsanditems

288

Creating matched and mismatched stimuli is crucial to our experimental289manipulation. However, it also disrupts the natural systematic association290betweentheacousticinformationinanygivencriticalsyllableanditsfunctionas291a true or pseudo prefix in the word. The fine phonetic detail of interest thus292becomesuninformativewithinthecontextof theexperiment.Becausewewere293worried about the effects of this disruption on the listeners,we presented the294matchedandmismatchedsentencesinseparatesessionsonseparatedays.Inour295analyseswetestedifoureffectofMatchdependedonwhetherparticipantsheard296all the matches on Day 1 (M1) or on Day 2 (M2) (factor Group, see297Counterbalancingsectionformoredetails).Furthermore,wetestedwhetherthe298effect of Match changed over the time course of the experiment (factor299TrialNumber—the order that each trial occurred in the experiment for each300participant).301

Finally,exploratoryanalysesalsoconsiderediftheeffectofMatchdependedon302whether theacousticsof thecritical syllablewere fromaword thathada true303prefixorapseudoprefix(PrefixStatus).Forexample,sentencesinFigure1witha304subscriptT1forthecriticalsyllabledishaveaPrefixStatusoftrueandthosewitha305subscriptP1 forthecriticalsyllablehaveaPrefixStatusofpseudo(anexampleis306alsoillustratedinFigure3).307

2.2 Participants308

Participantswere34nativeEnglishspeakersattheUniversityofYork(meanage30921years,range18-32,24women),withnormalorcorrected-to-normalvisionand310nohistoryofspeechorhearingproblems.Eachparticipated intwosessionson311differentdays at least oneweek apart. Each session took approximately45-50312

2By-itemrandomslopeswerenotfitduetosparsityofitemdataforanygivenTrialNum.


page9of44

minutes. Two additional participants were discarded due to errors in data313collection.314

2.3 AuditoryStimuli315

All stimuli can be found in the Open Science Framework repository316(https://osf.io/dsyxu/DOI 10.17605/OSF.IO/DSYXU).Sentenceswereconstructed317for32pairsoftargetwordsdifferingintruevs.pseudoprefixstatusoftheirfirst318syllable(e.g.displaces/displays).AsexplainedintheIntroduction,thesecomprised319two types (factor PhonemeChange). In the dis-mis- type (e.g., dis-:320discolour/discover[N=7];mis-:mistypes/mistakes[N=4]),atleastthefirstfour321phonemesofeachtrue-pseudopairwereidentical.There-ex-typefollowedthe322sameprinciple of having identical phonemes into at least the beginning of the323secondsyllable,exceptthat forthesewordsthefirstsyllable’svowelphoneme324differedwithprefixstatus(e.g.,re-:[ri:]re-strings/[rə]restricts[N=16];ex-:[ɛks]325ex-trampoliner/ [əks] extravagance [N=5]).Primary lexical stresswason the326secondsyllableofeachcriticalword.Eachwordwasplacedinasentencewhich327wasidenticaltothatofitspairbeforethetargetword,andinsomecasesafterit,328and could be illustrated by a picture (e.g. It was difficult because Sam329distrusted/distractedhim).330

Sentencepairsthatdifferedafterthetargetwordhadthesameintonationaland331foot structure, and hence number and stress-pattern of syllables, though not332necessarilythesamewordboundarieswithinafoot.Inonedis-andoneex-case333identical foot structure was achieved by adding an extra syllable because one334targetwordhadonelesssyllablethanitspair:Aswandisplaceswaterwhenitlands335and A swan displays its plumage to its mate; It’s a perfect example of ex-336trampoliners’senseofbalanceandIt’saperfectexampleofextravaganceinpublic337spending.Theseadditionsweremadeimmediatelyafterthetargetwordsoasto338matchthefootstructurecreatedbythelongertargetword,theparticularwords339beingchosentobesimilartothelongermemberofthepairinconnectedspeech340e.g.displaysitsvs.displaces.Allsyllablecountswereasstandardlypronouncedin341SSBE,aswellasintheparticularstimuli(e.g.discourteoushadthreesyllables,not342four:/ˌdɪsʹkɜtjəs/.AppendixBshowsthecompletelist.343

Because our focuswas onmatching the phonetic structure of the stemswhile344usingword pairs that could occur in sentences that were identical before the345criticalwordandhadidenticalprosodicstructurethroughout,wecouldnotmatch346targetwordsonfrequency.However,thisshouldnotbiastheresults,sinceover347and above the fact that word frequency is not a primary determinant of the348morphologicaldistinction itself (Smithet al., 2012,Hayet al., inprep), thekey349comparisonsweretobebetweencross-splicedstimuli inwhichthelexical item350wasthesame,theonlydifferencebeingintheacousticsignalinitsfirstsyllable,as351describedbelow.352

These32pairsofsentenceswererecordedin6randomordersbyamaleSSBE353speaker. Quality was controlled as follows. To minimize reading effects, the354speakerhadfamiliarizedhimselfwiththesentencesandpicturesforsomedays355beforetherecording,andwasencouragedtolookatthepictureratherthanthe356textwhilerecording.Contrastivestressonthecriticalwordswasavoidedinthat357only one picture was displayed at a time, and at least two other sentences358


page10of44

separated recording of the two members of each sentence pair. Trained359phoneticianschecked thestimulibothat the timeof recordingandafterwards.360Errors(utterancesthatcontaineddisfluenciesorthatsoundedunnatural,unclear,361orinappropriatefortheintendedmeaning),includingborderlinecases,werere-362recorded. Two recordings of each sentence were chosen for cross-splicing to363createa‘match’anda‘mismatch’versionofeachmemberofthepair.Sentences364were initially chosen for naturalness and the best impressionisticmatch of f0,365rhythmandloudness.Followingthis,thechosenpairsofstimuliwereinspected366acoustically to ensure that they not only sounded acceptable in their original367contexts,butthat,relativetoeachother, the internalacousticstructureofeach368critical syllable conformed to expectations derived from Smith, Baker and369Hawkins(2012),primarilyusingdurationalcriteria.Thiswasnecessaryinorder370that any observed behavioural differences could be interpreted in terms of371differencesininternalacousticstructure.SeeSection2.7foracousticanalysesof372thestimuli.373

Stimulusconstructioninvolvedtwotypesofcross-splicing,accordingtowhether374theresultanttokenwasa‘match’or‘mismatch’stimulus.Sentenceswerecutat375theendofthecritical(target)syllable(justbeforetheburstofthenextstopifthere376was one, as inmistimes) and cross-spliced either with the end of an identical377sentencefor‘match’stimuliorwiththeendofitspairfor‘mismatch’stimuli.Thus378eachstimuluswasconstructed fromtwoseparaterecordedtokens,spliced just379after thecriticalprefix/pseudo-prefix syllable: separate recordingsof the same380sentencefor‘match’stimuliandrecordingsofdifferentsentencesfor‘mismatch’381stimuli.382

Thus, as Figure 1 shows, four versions were created from each sentence pair383corresponding to the four combinations of true and pseudo prefixes and384continuations: the matches true-true and pseudo-pseudo, and the mismatches385true-pseudo and pseudo-true, for a total of 128 test stimuli (32 pairs x 4386conditions). These procedures meant that no perceptually significant acoustic387informationaboutthesecondsyllableofthewordwaspresentinthefirstsyllable:388secondsyllablesallhadidenticalorverysimilarvowelqualities(seeAppendixB),389and formis- and dis- syllables followed by a stop, Baker (2008) showed that390listenerscouldnotpredictthefollowingvowelunlesstheyheardtheburstand391followingVOT.392

Anadditional67fillersentencepairswereconstructed.Ofthese,30pairshadbeen393recordedbythesamespeakerandusedinapreviousexperiment;inthepresent394studytheycomprisedanindependentexperimentrunatthesametime.These30395consisted of pairs of sentences identical except for oneword, differing only in396whetheritcontainedan/r/or/l/(e.g.ramsvs. lambs),hereafterr-lsentences.397Matches andmismatcheswere created as described above, except that critical398wordsweresplicedintothesentences,ratherthanabuttingthefirstpartofone399sentencewiththesecondpartofanother(seeHeinrich,Flory,&Hawkins,2010400fordetails,andthelistofwordsandsentences).Oftheother37fillerpairs(listed401inAppendixC),allbutfourweredesignedtomimictheprefixonesinthatthey402contained awordwith a true or pseudomorpheme either before or after the403target/disambiguatingwords.ForexampleYoupurifywater/whiskybydistillingit.404Theremainingfourfillersfollowedthesamesemanticandprosodicprinciplesas405


page11of44

the others; three of them contained a word beginning re- paired in the other406sentencewith a non-re- word. These 37 fillerswere recorded twice; themost407naturalofeachwaschosen,andnotspliced.Sixadditionalfillertrialswerecreated408inthesamewayandusedinpracticeblocks.409

2.4 VisualStimuli410

A photograph was chosen to represent each sentence from images publicly411available on the web, and photographs we took ourselves. Care was taken to412ensure that pairs of images were similar in complexity and colourfulness, as413judged by six people, the four authors and two research assistants. See the414examples in Figure 1. The largest dimension of each imagewas 600 pixels. As415notedintheResults(Section3),baselinemeasuresoflookingpreferencetakenat416themomenttheauditorystimuluswaspresentedandattheonsetofthecritical417wordrevealednosystematicpreferences for thepicturesdepicting trueversus418pseudoprefixes.419

2.5 Procedure420

Participantswereseatedinfrontofadesktop-mountedremoteEyelink2000(SR421Research) tomonitor eye-movementswhile theyperformed the task.Auditory422stimuliwerepresentedoverheadphonesatacomfortablelisteninglevel.Visual423stimuliweredisplayedona16”x12”monitor.Eachsessionbeganwithsetupand424calibration of the eye-tracker followed by two practice blocks of trials, whose425structurewasidenticaltotherestoftheexperiment.426

Testingtookplaceontwodays(Section2.6). Eachday,participantsheardone427trialforeachofthe99pairs(32prefixpairs,30r-lpairsand37fillerpairs).Stimuli428weregroupedinto33presentationsets,eachcontainingthreepairsofsentences,429generallyoneprefixpair,oner-lpairandonefillerpair.Participantswerefirst430presented with a block of six familiarization trials in which each of the six431sentencesfromapresentationsetappeared,alongwithitsaccompanyingpicture,432oneatatime,inrandomorder.Thesentencewasnotspoken,butinsteadprinted433atthetopofthescreen.Thepicturewascenteredinthemiddleofthescreen,as434shown in the left portion of Figure 2. Participantswere instructed to read the435sentence silently and familiarize themselveswith thepicture, then clickon the436picturetocontinue(thetrialdidnotendbeforeatleast2.5secondsofviewing).437


page12of44

438

Figure2:Exampleofa familiarizationblockandcorrespondingtestblock.Left:439Familiarizationblock,6writtensentencesfromeachof3itempairs.Right:Test440block,3spokensentences,onefromeachofthesame3itempairs.Colouronline.441

After the familiarization block came a corresponding block of three test trials442(shownintherighthalfofFigure2),onefromeachpairinthe6-itempresentation443setshowninthelefthalfofFigure2.Figure3showsthestructureofanexample444testtrial.Eachtesttrialbeganwithadriftcorrectionfortheeye-tracker.Onepair445ofpictureswasthenpresented,onecenteredinthelefthalfofthescreen,theother446intherighthalf.Notextwasdisplayed.Thesideonwhichthetrueprefiximage447wasdisplayedwasrandomizedacrosstrials.Aftertwosecondsofpreviewtime,448one of the sentences was played. Participants were instructed to click on the449picturethatmatchedthesentenceasquicklyandaccuratelyaspossible.Oncethe450participantresponded,thepicturesstayedonscreenforanadditional0.5seconds;451theywerethenreplacedbrieflybyablankscreen,afterwhichthenexttrialbegan.452

453


page13of44

454

Figure 3: Structure of a single test trial. As this example trial shows, the455PrefixStatus of the critical syllable can mismatch the Target sentence. In this456example, the acoustics of the critical syllable are from a true prefixed word457(displaces)butthetargetwordandsentencecontinuationarethecorresponding458pseudoprefixedword(displays(its)).Colouronline.459

460

2.6 Counterbalancing461

Asdiscussedabove,wewereconcernedthatifthematchedandmismatchedtrials462wereallpresentedinthesamesession,thefinephoneticdetailofinterestwould463thusbeuninformativewithinthecontextoftheexperiment,andlistenerscouldbe464expectedtoquickly learnto ignore itas theyhaveforsimilarlysubtlephonetic465information(e.g.Hawkins&Nguyen,2001,Experiments2,3aand3b).Thechosen466blocked and counterbalanced design was intended to allow us to assess two467things:howthecriticalphoneticinformationisusedinrealtimewhenheardwith468itsnormal systematicdistribution reflectingmorphological status; and towhat469extentatypicaldistributionsinfluencerecognitionbehaviourintheshorterterm.470Wethususedablockeddesigninwhichallmatchedstimuli(r-landprefix)were471presentedononeday,andallmismatchedstimuliwerepresentedonanother,the472twosessionsbeingseparatedbyatleastaweek.Theorderofmatchandmismatch473wascounterbalanced(factorGroup):18participantsheardallmatchesonDay1474(M1)andallmismatchesonDay2,whiletheother16heardallmismatchesonDay4751,andonlymatchesonDay2(M2).Twoadditionalparticipantswererecruitedfor476group M2 but it was later discovered that they had to be excluded due to477experimentererrorindatacollection.478

Becausetheparticipantswouldbeseeingeachpairofimages(andhearingoneof479themdescribed)onDay1andagainonDay2,itwasimportantthattheynotbe480abletopredictwhichimagewouldbedescribedonthesecondday.Forthisreason,481asecondcounterbalancingfactorwasadded.Thestimuluspairsweredividedinto482twosetssuchthatforeachparticipant,ononehalfoftrialsthesameimagewas483describedonDays1and2,andontheotherhalfoftrials,theoppositeimagewas484


page14of44

describedonDays1and2.Thus,itwasimpossiblefortheparticipantstopredict485whichpicturewouldbedescribedonanytrial.Practiceblocksalsoillustratedthis486pattern.Inallcases(exceptthe37unsplicedfillers)thestimuluseachparticipant487heardwasdifferentonthetwodays,eithermatchingormismatchingdepending488onthedayandthegroup.489

Insummary,forcriticaltesttrials,eachparticipantheardoneofthefourstimuli490describingeachpaironeachday;matchesandmismatcheswerenevermixedin491onesession(orderwascounterbalancedbetweenGroupsM1andM2);andthe492presentationofspokensentenceswithinDay1andDay2wassuchthatpredicting493whichofthetwoimageswouldbedescribedintheexperimentaltrialsshouldhave494been at chance. There were four groups created by counterbalancing these495conditions. Four additional conditions were created with a different random496groupingofstimuluspairs.Trials fromthedis-mis-andre-ex-setsofsentences497wereevenlydistributedthroughoutalltheseconditions.498

2.7 Acousticsofcriticalsyllables499

Weextensivelyanalysedtheacousticpropertiesofourstimulitoensurethatthey500conformed to expectations from the previous phonetic literature and also to501ensurethattheydidnotcontainunwantedbiases.Afullreportoftheseanalyses502is available in the supplemental materials503(https://osf.io/dsyxu/DOI 10.17605/OSF.IO/DSYXU). We found that the critical504syllablesvariedaccordingtoprefixstatusaswewouldexpect.Wealsofoundno505evidenceforsystematicacousticdifferencesbeforethosesyllablesthatmightbias506theinterpretationofthecriticalsyllables.Herewebrieflyreportontheduration507ofthecriticalsyllablesthemselves.508

Figure4showsthedistributionofsyllabledurationsforeachoftheprefixes.True509prefixes were on average 54 ms longer than pseudo prefixes, with dis-mis-510syllables 55 ms longer on average than re-ex- syllables. However, as Figure 4511shows,theseoverallobservationsmaskdifferenceswithinthesyllabletypesthat512areimportantforinterpretingtheeyetrackingresults.Whilemis-anddis-mean513durationsandstandarddeviationsaresimilarenoughthatthetwosubtypescan514beregardedasaroughlyhomogeneousgroup,thisisnotthecaseforthere-ex-set:515therearelargedifferencesbetweenre-andex-syllabledurations,withabsolute516andrelativevaluesforex-patterningmorelikethosefordis-andmis-thanthose517forre-.Theselargedifferenceswithinthere-ex-setareduetothephoneticmakeup518ofthesyllablesandtheconsequentdegreetowhicheachcanbereduced.Whereas519re- canbeseverely reduced,inherentdurational constraintsonEnglish/k/and520particularly/s/meanthatex- ismuch lessopentoreduction.Furthermore,ex-521syllablesarethelongest,andre-syllablestheshortestofall foursyllabletypes.522Becauseofthisgreatheterogeneity,itwasdecidedthatre-andex-shouldnotbe523treated as a single group; and because there were so few ex- tokens it was524necessarytoexcludetheex-stimulifromtheanalysis.525

526


page15of44

527

Figure4:Histogramsofcriticalsyllabledurationsforexperimentalstimuli.Colour528online.529

2.8 Eye-trackingdata:analysisprinciples530

Responses to filler stimuli were not analyzed. Trials in which the participant531clickedontheincorrectimage(didnotmatchthecontinuationofthesentence)532wereremoved(atotalof151trialsor6%ofthedata).Eye-movementsfromall533remainingtrialswerethentime-alignedtothestartofthecriticalword.534

Because our visual stimuli and sentences were relatively complex, we first535establishedthatlistenershadnooverallpreferencefortheimagescorresponding536to one set of words or the other (true or pseudo prefixed words). The mean537proportionoflookstothetwotypesofimagewasalmostidenticalattheonsetof538thecriticalword(plustheexpected200msoculomotordelay;Matin,Shao,&Boff,5391993):pseudo=0.46,true=0.47;pairedsamplet-testsp=0.63bysubjects,p=5400.65by items.Therewas likewisenodifferenceat sentenceonset (p=0.98by541subjects,p=0.94byitems).542

For all analyses, proportion of fixations to the target image (as defined by the543continuationof the sentence)werecomputedovera specific timewindowand544these proportions were transformed to log odds for analysis with linear545regression. Linear regression with log-odds-transformed proportional data is546comparable to logistic regression on data in which each observation is either547targetornot,butallowsforaggregationofdataoveragiventimewindow.548

Allstatisticalanalysesweredoneusingmixedmodellinearregressionusingthe549lmer()functionfromthelme4package(Bates,Maechler&Bolker,2015)inR(R550Development Core Team). Significance was assessed using the Satterthwaite551approximation of degrees of freedom as implemented in the lmertest package552(Kuznetsova,Brockhoff&Christensen,2017)inR.Allfactors(binarycategorical553variables)werecenteredbyusingcontrastcoding(0.5vs.-0.5)andcontinuous554variables were centered and scaled. Centering the variables avoids any co-555linearitybetweentheeffectsandtheirinteractions.FactorswereMatch(match=5560.5,mismatch=-0.5),PhonemeChange(dis-mis-=0.5,re-=-0.5),Group(M1=0.5,557M2=-0.5),andTrialNumber(continuous,scaled).558

Randominterceptsforsubjectsanditems(eachitemwasasentencepair)were559includedinallmodels.Randomslopeswereincludedwhereverthedesignandthe560dataallowed(seebelowfordetails).561

562


page16of44

3 Results563

Thegoalofthefirstanalysiswastodeterminewhetherlistenersweremorelikely564tolookatthetargetimagewhentheacousticsofthecriticalsyllablematchedthe565morphemicstructureofthetargetword,thanwhentheacousticsmismatched.566

3.1 200-800mswindow567

3.1.1 Mainanalysis568

Ourfirstanalysisaggregatedlookswithinawindowfrom200to800msafterthe569onset of the critical word. This window was chosen because it is generally570assumedthatittakesroughly200mstoplanandlaunchaneyemovement(Matin,571Shao,&Boff,1993).Thusthewindowbeginswhenwewouldexpecttoseelooks572influencedby theonsetof thecritical targetword.Thewindowcontinuesuntil573looks to the target start to asymptote (atwhichpointweexpect anyeffects to574disappear).Thusanydelayinidentifyingthetargetshouldbeseeninthiswindow.575

576

Figure5.Fixationproportionstothetargetandcompetitorimagealignedtothe577wordonsetforthematching(blacklines)andmismatching(redlines)conditions.578Solidcurves:lookstotargetimage.Dottedcurves:lookstocompetitorimage.The579dashedverticallineattime=0isthealignmentpoint,thebeginningoftheword.580Thedottedverticallineindicatesaveragesplicepointacrossallstimuli.‘Target’is581definedasthepartofthesentencefollowingthesplicepoint.582

Figure5showstargetandcompetitorfixationsovertimefortrialsinwhichthe583criticalsyllableeithermatchedormismatchedthecontinuationof thesentence584(i.e.thetarget),alignedatthewordonset.Overthecourseofthetrial,participants585lookedmoreatthetargetandlessatthecompetitorandthisdifferenceisbigger586formatchingtrialsaspredicted.Wetestedthedifferenceinlookstothetargetby587examiningtheeffectofMatchinamodelthatalsoincludedPhonemeChangeand588itsinteractionwithMatchtotestwhethertheeffectofMatchwasdifferentforthe589


page17of44

dis-mis- and the re- stimuli. This model included random slopes for Match by590subjectanditemaswellasrandomslopesforPhonemeChangeanditsinteraction591with Match by subject (i.e. the maximal model). The output of this model is592summarizedinError!Referencesourcenotfound.Table2.Therewasahigher593proportion of looks to matched targets than to mismatched targets. Neither594PhonemeChangenortheinteractionofMatchwithPhonemeChangesignificantly595affectedresponseshowever.Seesupplementalmaterialsforitemandparticipant596variabilityineffectofMatch.597

Table2:Modelsummaryfor200-800mswindow.598

599

Estimate Std.Error t p

Match 0.57 0.21 2.73 0.01

PhonemeChange 0.15 0.30 0.48 0.63

Match:PhonemeChange -0.01 0.41 0.03 0.97

600

3.1.2 TrialNumberandGroup(200-800mswindow)601

WealsoconsideredamodelthatincludedTrialNumber3andGroupaswellasall602the two and three-way interactions with Match and PhonemeChange. These603models were considered because, as discussed in the Method (Section Error!604Referencesourcenot found.,Design),wehypothesized thatbeingexposed to605bothmatchingandmismatchingstimulimightweakentherelationshipbetween606theacousticsandthemorphologicalstatus.Thismightleadtoadecreaseinthe607Matcheffectoverthecourseoftheexperiment,oradecreaseintheMatcheffect608justforlistenersexposedtomismatchesonthefirstday(GroupM2).Furthermore,609it may be that only Group M2 would change behavior over the course of the610experiment(aninteractionbetweenGroupandTrialNumber).Themodelagain611found a robust effect of Match and no interaction between Match and612PhonemeChange.Groupwasnotsignificantanddidnot interactwithanyother613effect.TrialNumber,whichonlyapproachedsignificanceasamaineffect(β=-0.17,614SE=0.09,t=1.91,p=0.06), interactedsignificantlywithPhonemeChange(β=-6150.43,SE=0.18,t=2.36,p=0.02)butnotwithMatch(β=-2.6,SE=0.18,t=1.43,p616=0.15).Thispatternindicatesthatlookstothetargetdecreasedoverthecourse617oftheexperiment, inparticularfordis-mis- trials.Therewasalsoatrendinthe618datathatindicatedtheeffectofMatchlessenedoverthecourseoftheexperiment619forthosedis-mis-trials.AnalysesincludingGroupandTrialNumbercanbefound620inthesupplementalmaterials.621

3.1.3 PrefixStatus(200-800mswindow)622

3Themodels reportedheredidnot include any randomslopes forTrialNumber as this led toconvergenceproblems,likelyduetothesparcityofthedataonanindividualoritemlevel.


page18of44

Afinalsetofexploratoryanalysesexaminedwhethertherewereanyasymmetries623in the effects ofMatchdue to PrefixStatus rather thanPhonemeChange. Visual624inspection of the data suggested asymmetries (as shown in the supplemental625materials)withabiggereffectofMatchwhenthecriticalsyllablewastakenfrom626trueprefixes,especiallyforgroupM1andespeciallyfortrialsinthefirsthalfof627theexperiment.AmodelincludingPrefixStatus,TrialNumberandGroupfoundan628interactionbetweenMatch,PrefixStatusandTrialNumber(β=0.93,SE=0.36,t=6292.56,p = 0.01)which indicated that the effect ofMatchwas greaterwhen the630criticalsyllablewastakenfromatrueprefixthanwhenitwastakenfromapseudo631prefix,especiallyinthebeginningoftheexperiment.InteractionswithGroupwere632notsignificant thoughnumerically theeffectofMatchwasgreatest for thetrue633prefixesforgroupM1.634

635

3.2 Timecourseofeffects636

3.2.1 Mainanalysis637

Thepreviousanalysesestablishedthatlistenersspentlesstimefixatingthetarget638image when the acoustic information in the critical syllable mismatched the639morphologicalstructureofthetarget.AsnotedintheIntroduction,wewerealso640interested in how quickly the acoustic information influenced their looking641behaviour, i.e. whether listeners used the acoustic information to drive eye-642movementspredictively,beforetheyheardanydisambiguatinginformation.The643alternativeexplanationofthemismatcheffectobservedabovewouldbethatthe644informationisnoted,butisnotbyitselfsufficienttoguideexpectationsandhence645influence behaviour. Rather it would presumably have a sort of cumulative646influenceonperceptualdecisions thatdependedonoverall acoustic coherence647betweenthefirstandlatersyllablesintheword,andassuchwouldpresumably648influencelaterandnotearlierlooks.Sucharesultwouldbeinteresting,butwould649pointtoarolethatisdifferentfromourhypothesisofastrongpredictiveinfluence650thatisthemainmotivationforourwork.651

To address this issue, we examined the evolution of the effects examined in652previous sections over the course of the trial. Following Clayards, Niebuhr &653Gaskell (2015)andKingston,Levy,Rysling&Staum(2016) webinned theeye654movements into100msbinsandperformedtheregressionmodeloneachbin.655Weincludedthesamefixedandrandomeffectsstructureasthemainmodelabove656(Match,PhonemeChange,andtheirinteraction)aswellasTrialNumberandGroup657and their interactions with the other fixed effects 4 . The estimates for Match,658PhonemeChangeandtheirinteraction(aswellasTrialNumberdiscussedbelow)659areplottedinFigure3intermsoftheestimatesandpvaluesofthefittedmodels.660

4ModelsrunwithoutTrialandGrouphadthesamepatternofresultsforMatch,PhonemeChangeandtheirinteraction,seesupplementalmaterials.


page19of44

661

662663

Figure 6. Results of mixed effects regressions over time for Match,664PhonemeChange, and their interaction. Shading is two standard errors of the665coefficientestimatesascalculatedbytheregressionmodels.Oneachpanel’syaxis,666pvaluesareshownattheleftandbetacoefficientsattheright.Blackcurves:beta667coefficients;redcurves:pvalues.Thedottedredhorizontallineindicatesp=0.05.668ThethickgreysolidhorizontallineindicatesCoefficient=0.Colouronline.669

Figure 6 (left panel) shows that the beta values for Match increase and then670decreaseasthetrialprogresses,asymptotingaround450msaftercriticalword671onset.Atthesecondtimebin,between200and300msfromtheonsetoftheword,672thetwo-standarderrorbarsjustmisstouchingzeroandthepvalueis0.04.After673thatpointtheeffectofMatchisclearlybelowp=0.05untilthelasttimebin,when674itreturnsto>0.05.Thisindicatesthattheacousticpatternofthecriticalsyllable675affects looks to the target fromvery earlyon in the syllable;Matches facilitate676correct prediction of the sentence continuation (the target). Assuming the677standard200mslagbetweenplanningandexecutinganeye-movement,200-300678msafterwordonsetistheearliestpossiblewindowforwhichwemightexpectto679see any effects. The results of that model indicate that at least for many680participantsanditems,thereisaneffectofMatchatthisearliesttimepoint.As681Figure4shows,formostoftheitems,thecriticalsyllableislongerthanthelength682ofthiswindow(100ms)sothedisambiguatinginformationatthesplicepointhas683notyetarrived.Thisstrengthensourclaimthat theacoustic information in the684criticalsyllableisbeingusedtoanticipatethetargetwordandlookstothetarget685aredelayedwhenitmismatches.686

Although themodel on a single largewindow (Section 3.1) found no effect of687PhonemeChange,thistime-courseanalysisshowsthatPhonemeChange(shownin688

Match PhonemeChange Match:PhonemeChange

200 400 600 800 200 400 600 800 200 400 600 800

0.00

0.25

0.50

0.75

1.00

−1

0

1

Time since word onset (ms)

P va

lues

Fixed Effect Coefficients

Parameter Beta Estimates P values


page20of44

Figure6,middlepanel)issignificantinthesecondhalfofthe200msto800ms689window(fromabout500mssincewordonsetandafter theendof the critical690syllable).Thecoefficientestimatesindicatethatthiswasduetomorelookstothe691targetfordis-mis-itemsthanforre-itemslaterinthesentence.Asbeforeinthe692model on a single large window, there is no interaction between Match and693PhonemeChange(Figure6,rightpanel).694

3.2.2 GroupandTrialNumber(Timecourse)695

Asbefore,wealsoincludedGroup(whethertheparticipantsheardallmatchesor696allmismatchesonday1)andTrialNumber(thetrialorderintheexperiment)and697their interactions. As before we found that neither Group nor any of its698interactionshadasignificanteffectatanytimepoint.Themodelonasinglelarge699window had found that looks to the target decreased over the course of the700experiment (effect of TrialNumber), especially for the dis-mis- stimuli701(TrialNumberbyPhonemeChangeinteraction).Themodelsfitevery100msfound702that the effects ofTrialNumber aswell as its interactionwithPhonemeChange703werelimitedtothefirstfewhundredmillisecondsafterwordonset(p<0.05for704thefirstthreetimebins,Figure7).Thisseemstoindicatethatoverthecourseof705the experiment, participants stoppedmaking early looks to the target, i.e. they706stoppedanticipating the target.Aspointedout in the Introduction, thismaybe707because (within the context of the experiment) it was not essential to pay708attentiontotheearlypartoftheword—thecontinuationofthesentencetypically709disambiguatedthetwoimages.Participantsmayhavelearnedthis(implicitlyor710explicitly)astheexperimentprogressed.Therewasalsoatrendfortheearlyeffect711ofMatchtogetsmalleroverthecourseoftheexperiment(MatchbyTrialNumber712Figure7).Noothertwoorthree-wayinteractionshadanytimepointswithap-713value<0.05(seesupplementalmaterialsforfulldetails).714


page21of44

715

Figure7:ResultsofmixedeffectsregressionsovertimeforTrialNumberandits716interactionswithMatchandPhonemeChange.Shadingistwostandarderrorsof717thecoefficientestimatesascalculatedbytheregressionmodels.Oneachpanel’sy718axis,pvaluesareshownattheleftandbetacoefficientsattheright.Blackcurves:719beta coefficients; red curves: p values. Thedotted redhorizontal line indicates720p=0.05.ThethickgreysolidhorizontallineindicatesCoefficient=0.Colouronline.721

3.2.3 PrefixStatus(Timecourse)722

A final analysis considered models that included PrefixStatus instead of723PhonemeChangeanditsinteractionswithMatchandGroupaswellasMatchand724TrialNumber.Figure8(leftpanel)showsthattherewasanon-significanttrend725foraninteractionbetweenPrefixStatusandMatchinthesecondhalfofthe200-726800mswindow(after500msfromwordonset)thatindicatedthattheeffectof727Matchmayhavebeengreaterfortrueprefixes,consistentwiththenumerictrend728from the single large-window model. There were also significant interactions729betweenPrefixStatus,MatchandTrialNumberduringthissamelaterpartofthe730200-800mswindow,asshownintherightpanelofFigure8.Thisindicatesthat731theMatchxPrefixStatusinteractioninfluencedeye-movementsatthebeginning732oftheexperimentbutbecameweakeroverthecourseoftheexperiment,whichis733probablywhyitwasnotstatisticallysignificantwhenaggregatedoverthewhole734time-courseoftheexperiment.735

736

TrialNumber Match:TrialNumber PhonChange:TrialNumber

200 400 600 800 200 400 600 800 200 400 600 800

0.00

0.25

0.50

0.75

1.00

−1

0

1

Time since word onset (ms)

P va

lues

Fixed Effect Coefficients

Parameter Beta Estimates P values


page22of44

737

Figure8:ResultsofmixedeffectsregressionsovertimefortheinteractionsMatch738xPrefixStatusandMatchxPrefixStatusxTrialNumber.Shadingistwostandard739errorsofthecoefficientestimatesascalculatedbytheregressionmodels.Oneach740panel’syaxis,pvaluesareshownattheleftandbetacoefficientsattheright.Black741curves: beta coefficients; red curves: p values. The dotted red horizontal line742indicates p=0.05. The thick grey solid horizontal line indicates Coefficient = 0.743Colouronline.744

745

3.3 Summaryofresults746

Overall,listenersspentmoretimelookingatthetargetimage(theoneconsistent747with the sentence continuation) when the acoustic properties of the critical748syllablematchedthoseexpectedforthetargetwordintheassociatedimage(main749effectofMatchinallmodels),thussupportingthemainhypothesis.Thiswastrue750both when the phoneme changed (re- stimuli) and when only the acoustical751patternwithin the syllables changed (dis-mis- stimuli) (i.e. no interactionwith752PhonemeChangeinthemodelonasinglewindowfrom200msto800ms).753

Wealsoexaminedpatternsoverthecourseofthesentence(successive100ms754windowsbetween200to800msfromwordonset)andoverthecourseoftrials755intheexperiment(fromthefirsttothelasttrialforeachparticipant).Whenwe756examinedthetime-courseofthesentence,wefoundthatlistenerslookedtothe757correctcriticalsyllablefromtheearliestmomentsofitsbeingheard.Thisuseof758theacoustic-phoneticdetailofthecriticalsyllableisreflectedintheeffectofMatch759being significant from the 200-300 ms bin. It indicates that the acoustic760informationwasimmediatelytakenupandusedpredictivelybylisteners.Wealso761foundthatafewhundredmillisecondslater, listenershadmostlyconvergedon762thetargetfordis-mis-stimulibutwerelookinglessconsistentlyatthetargetfor763re-stimuli(i.e.effectofPhonemeChangestartingatthe500-600msbin).Thismay764


page23of44

bebecausemanyofthere-stimuliweregloballyambiguous,orpossiblybecause765the greater acoustic complexity of dis-mis- syllables compared with re- ones766(includingthepresenceofabruptacousticboundarieswithindis-mis-syllables)767makes them more auditorily distinctive. Examining the time-course of the768experiment,wefoundthatastheexperimentprogressed,theearly lookstothe769target decreased, especially for the dis-mis- stimuli (i.e. PhonemeChange by770TrialNumberinteractionuptothe300-400msbin).Thisindicatesthatlisteners771began to respond differently to (especially) dis-mis- critical syllables as they772becamefamiliarwiththestructureofthestimuli.Finally,wefoundthatafterabout773500msfromthewordonset,effectsofMatchweregreatestwhenlistenersheard774a true prefix, though, consistent with the way responses changed as the775experiment progressed, this benefit for matched prefixes was only at the776beginningoftheexperiment.777

778

4 Discussion779

4.1 Themainfindings780

Atageneral level,weaskedwhetherlistenersaresensitivetophoneticdetail–781both segmental and rhythmic information – that systematically reflects782morphologicalstructurewhile involvingnochanges innumberofsyllables.We783furtheraskedwhetherlistenersarelikelytousethisphoneticdetailpredictively784(in order to help distinguishwords that contain true prefixes from those that785containpseudoprefixes)inordinarylisteningconditions—thatis,inaneasyon-786linetaskundergoodlisteningconditionsinwhichtheaimisresponseaccuracy787butnotspeed.788

The threemain questions specific to our experiment were whether there is a789mismatcheffectoverall,whetheritisindependentofthephonemicstatusofthe790acousticdifference(i.e.withdis-mis-aswellasre-),andinparticularwhetherthe791acousticinformationwithinthecriticalsyllableinfluencesperceptualdecisionsin792real time,rather thanonlybeing influential incombinationwiththerestof the793word.Allthreequestionsareansweredaffirmatively.Listenersspentmoretime794fixatingthetargetimagewhenthecriticalsyllablematchedthecontinuationofthe795sentence, whether or not there was a phoneme change, and in real time.796Furthermore, while prefix status did not affect the answers to the threemain797questions, the results suggest that true prefixes may convey more reliable798information about their status thanpseudoprefixes in some circumstances, as799discussedbelow.800

We also made some additional observations. Foremost amongst these is the801evidence for rapid learning during the task. As expected, listeners used the802internalacousticstructureof theprefixedornon-prefixedsyllablepredictively.803However, as the experiment progressed this prediction effect weakened,804presumablyaslistenerslearnedthattheycouldwaitforthesentencecontinuation805toprovidedisambiguatinginformation.Thattheydidnotwaitforthecontinuation806duringearlytrialsmeansthatthecriticalacousticinformationislikelytobeused807predictively innormal listeningconditions.This findinghaspracticalaswellas808theoretical interest: the fact that listeners’ behavior changed early in the809


page24of44

experimentsuggeststhatfutureworkonthistypeofdistinctionshouldconsider810trial number as a predictor variable. Furthermore, prefix status (whether the811initial syllable came froma true or pseudoprefixedword) seems to affect eye812movements.The200-800mssinglewindowanalysisshowedastrongerbenefitof813Match when the critical syllable was a true prefix, and the interaction with814TrialNumber confirmed that this benefit was again especially obvious in the815beginningstagesoftheexperiment.AsnotedintheIntroduction,althoughboth816prefixesandpseudoprefixesareweaksyllablesinthattheydonotcarryprimary817lexicalstressandarenotnormallyaccentedinutterances,prefixesareassociated818with a degree of stress, or rhythmic prominence, that pseudo prefixes in819comparablewordstructureslack.Thepresentresultssuggestthattherhythmic820emphasisthatcomeswithatrueprefixmaybemoreperceptuallycompellingthan821theabsenceofsuchafocus.Thisfurtherencouragesexplorationofthehypothesis822thatrhythmicpropertiesofthesignalarefundamentaltospeechprocessinginreal823time.Thenexttwosectionsdiscussthenatureofrhythmandmetre,andoutline824itsrelevancetoageneralmodelofperceptualprocessing.825

826

4.2 Rhythmandhierarchicalmetrical structure as organising principles827forspeechperception828

It has been argued that the phonetic detail manipulated in our study is best829understoodas reflectingdifferences in rhythmicormetrical structurebetween830prefixesandtheirequivalentphonesinmono-morphemicwords(e.g.Smith,2012;831Hawkins, 2001; 2003; 2010). This argument, and our current results, support832other suggestions in the literature that rhythmic or metrical structure is an833importantpartoftherepresentationusedtorecognizespeech(cf.Salverdaetal.,8342003;Brownetal.,2015;Breenetal.,2014).Thissectionfirstoutlinestheuseful835distinctionbetween rhythmandmetrical structure, and their interrelationship.836Thenitexploreshowthemetrical-rhythmicstructureofspeechmightserveasa837fundamental organising principle for speech perception, melding multi-modal838propertiesofthephysicalsignalwithlinguisticandsocialknowledgetoachieve839communicativesuccess.8408414.2.1 Metreandrhythm842

Musical analysesdistinguish rhythm frommetre. London (2012)expresses the843distinction as follows. Rhythm represents a series of physical events having844particularrelationshipswithoneanother.Inmusic,thesearelargelydurational845(the inter-onset intervals of notes). Metre, in contrast, is a perceptual846phenomenon:anemergentorganizationinvolvingadegreeofperiodicitythatis847constructedby thebrain in response to stimuli that areperceivedas rhythmic848(Fujioka et al., 2012). In hearing metrically, the brain sets up a beat that849hierarchicallystructurestherhythm,focussesattentiononthemetricalbeats,and850allowspredictionofthetimeofoccurrenceoffutureevents(seee.g.Calderoneet851al.,2014;Lakatosetal.,2005;Lakatosetal.,2008).852853Thisdistinctionbetweenrhythmandmetrecanbehelpfulforspeechanalysistoo.854Speechrhythmcanoftenberepresentedsimplyintermsofrelativedurationsof855


page25of44

similarunitsintheutterance.Otherparameters—f0,amplitude,andsometimes856timbre—maycontributetobothrhythmicandmetricalaspectsofspeech.Indoing857so, theycanoverridedurational influencesonperceivedrhythmicandmetrical858structure(e.g.Dilley,Mattys&Vinke,2010,Experiment3a).Thusmetre,whichis859inherentlyhierarchical,canberepresentedforspeechasthemappingofauditory860patterns onto linguistic units, from segments (allophones, phonemes or their861psychological equivalents) through syllables to metrical feet and intonational862phrases.863864Metrical,orbeat-based,structure,enablesestablishmentofametricalhierarchy865where faster rhythmic events canhappenwithin slower ones. Faster rhythmic866events in speech presumably include syllables (or syllable-like units such as867Japanesemorae,hereafternotdistinguished fromsyllables). In linguistic terms868suitable for languages likeEnglish, abeat-basedhierarchyof syllabicweight is869called stress, the main beats being accented syllables (sometimes called870prominence, or primary stress), while less important syllables take secondary871stressorareunstressed.Commontobothspeechandmusicisthatperceptionof872arhythmicgroupcanchangedependingonthelistener’sconstrualofthewider873metrical(e.g.forspeech,sentential)structureitoccursin.Forspeech,preceding874meaning and/or rate of speech influence perception (e.g. Pickett and Pollack,8751963;Ernestus,BaayenandSchreuder,2002;Ernestus,2014;Ernestus,Hanique876andVerboom,2015;Dilley,Mattys&Vinke,2010;Heffner,Dilley,McAuleyand877Pitt,2013;Morrill,HeffnerandDilley,2014),whileexplicitinstructionstoheara878beattraininternaryorelseinbinarytimecanalsoinfluencelisteners’metrical879response, asmeasuredbyEEG (Nozaradan, Peretz,Missal andMouraux, 2011;880Nozaradan,Peretz,andMouraux,2012).881882Theexperienceofrhythmincomplexauditorysignalssuchasspeechandmost883music is learned (Mattys et al., 1999; Hannon & Trehub, 2005).Without such884(usuallyimplicit)learningfromexposuretotherelevantsignals,whichamounts885toacculturation,complexsoundsequencessoundunstructured,evenchaotic.For886speech,suchlearningispartoflinguisticknowledge.Withsuchknowledge,beat-887basedlisteningfacilitatespredictionofupcomingevents,apropertyincreasingly888seenasessentialtosuccessfulcommunicationbetweenindividuals(Philips-Silver889&Trainor,2005;Cirelli,Wan&Trainor,2016).890891Theprinciplesdescribedherehavebeenusedtoexploretherelationshipbetween892speechrhythmandthevariousfrequenciesofcorticalneuraloscillations inthe893brainwhichentraintoexternalstimuli.Oneofthemorecompletemodels,Giraud894and Poeppel (2012), privileges rhythm in speech intelligibility. Amongst other895things,itidentifieslowgamma(25-30Hz,33-40ms)andtheta(4-8Hz,125-250896ms) frequencies as entraining to feature/phoneme-sized and syllable-sized897durations respectively. Delta frequencies (1-3 or 4 Hz, 250-1000 ms) are898implicated in prosodic processing. However, it seems mistaken to necessarily899associateshorterdurationswithphonologicalunitsandlongeroneswithprosody.900Englishhasmany instancesof longphonologicalunitsandshortprosodicones901(Hawkins,2014:1-3).902


page26of44

Astrongerargument,relevanttothepresentstudy,comesfromMai,Minnettand903Wang’s (2016)EEGstudyofMandarinChinese.Theymanipulatedsentencesof904meaningfulvs.nonsensedisyllabicwords,andbackwardsvs.normalspeech.Like905Giraud and Poeppel (2012), they concluded that phonological and syntactic-906semantic processing engage different neural networks, but identified quite907different frequencies: semantic/syntactic processing with fast gamma908frequencies,andphonologicalprocessingwithslowerthetaanddeltafrequencies,909aswellasbeta(13-30Hz,33-77ms).Thesepatternsreflectthatthesyllable,not910the phoneme, is the important contrastive unit in Mandarin phonology:911consonantsstronglydeterminevowelquality,f0operatesovertheentiresyllable912to changewordmeaning, aswell as in longer prosodies, and syllable stress is913relativelyinvariant.914

In sum, perceived speech rhythm involves representation of sequences of915linguisticunitsofthesizeofsyllablesorlonger.Metreishierarchicalandinvolves916representationoftheentirespectrotemporalsignalofaphraseormore.Thatis,917metricalstructure isconstructedor imposedbythe listeningbrain,ratherthan918being a property inherentwithin the physical signal. So recognition of speech919rhythm andmetremay entail recognition of entire complex auditory patterns,920withinwhichsubunitscanbediscerned.Expectedattributesnotclearlypresentin921the physical signal butwhose presence is implied by and compatiblewith the922overallpatterncanbeadducedbyneuralpatterncompletion,aprocessattested923forbothvision(e.g.Meng,RemusandTong,2005;Murray,Kersten,Olshausen,924SchraterandWoods,2002)andspeech (Shahin,BishopandMiller,2009; fora925review,seeHawkins,2014). Assuch,rhythmandmetreofferthepossibilityof926structuring the speech signal such that all its contrasting abstract units are927representableinasystematicandeconomicalwaythatnecessarilyincludes‘top-928down’knowledgeofthelanguageitself.929

4.2.2 Relevancetoperceptionofaffixedwords930

Consistent with the arguments above, every utterance can be described by a931metrical(prosodic)hierarchythatpartlygovernsthephoneticdetailofsegments932and syllables within its domain. Autosegmental-Metrical (AM) phonology (e.g.933Pierrehumbert & Beckman, 1988; Post, D’Imperio & Gussenhoven, 2007; Cho,9342016),andFirthianProsodicAnalysis(FPA,e.g.Ogdenetal.,2000)exemplifytwo935theoreticalframeworksbasedonthisapproach.InCho’s(2016,p136)wordsfor936AM, “Prosodic structure provides a “frame” for articulation based on which937abstract phonological representationswhose phonetic detail is rather coarsely938specified by the phonology of the language are fleshed out with fine-grained939phonetic content in both segmental and suprasegmental dimensions…this940assumption entails that the prosodic structure of an utterance is phonetically941“encoded”intothespeechsignalandthelistenerinturndecodesthestructural942information from the signal and exploits it in speech comprehension.” FPA943embodiessimilarprinciples,andinadditioneverymetrical/prosodichierarchical944structure describing an utterance is linked to its corresponding syntactic945hierarchicalstructure(Ogdenetal.,2000).Thesemetricalprincipleshavebeen946extended to perception, particularly of prefixed vs non-prefixed words, with947discussion of links to grammar,morphological structure, and lexical items and948theirassociativenetworks(Hawkins,2010;HawkinsandSmith,2001;Hawkins,949


page27of44

2003).Theresultsofthepresentexperimentbroadlysupportclaimsinthepapers950cited,butsomeupdatingiswarranted.951

ThepresentstudysupportsfourcentraltenetsofHawkins’theoreticalposition,952madeinthepaperscitedabove.Thefirsttenet(whichbeforethepresentstudy953wasahypothesis)isthatthefinephoneticdetailthatdistinguishesprefixesfrom954nonprefixsyllablesisusedbylistenersinrealtime—i.e.theirbehaviorchangesas955theyhearit.Thisphoneticdetailrelatestomeaning,anddoesnotdependonthere956beingaphonemiccontrasttorelateittomeaning.Byextension,allphoneticdetail957seems likely to be related to meaning, with no intervening ‘levels’ of formal958linguisticstructurelikelytobeobligatoryintheprocessofrelatingsoundtothe959talker’s meaning. This does not deny the psychological reality of intervening960structural‘levels’;itmerelysaysthatagivenlevelofanalysisdoesnotalwayshave961to be accessed for meaning to be understood from the spoken signal.962Neuropsychological support for this claim comes from Krieger-Redwood et al.963(2013),whousedTMStoshowthatprocessesrequiringsemanticcategorisation964(i.e. understanding word meaning) are independent of judgments requiring965phonological classification (i.e. phonemic content); semantic judgments can966operate when phonological processes are unavailable. This has important967consequences: it means that the perceptual system is probably more closely968attuned to general, modality-free properties of pattern recognition than is969normally assumed by perceptual theories based on the mutually exclusive970categoriesandanalyticlevelsoftheoreticallinguistics.971

The second tenet is that every short pattern of sound (segment or several972segments)canonlybedescribedandhenceperceptuallyinterpretedintermsof973its context. The same sound in a different contextmay be interpreted entirely974differently. Our experimental manipulation and analysis addressed this claim975implicitly, especially by virtue of tracking decision changes (e.g. to the critical976syllable)overtime.Alistenercanmapsounddirectlytomeaning,butonlyforthe977contextinwhichitisheard.Contextisbroadlydefined.Itincludestheimmediate978localcontext—that is, theprosodic/metrical structureof theutterance that the979sound is part of—possibly one or more preceding phrases, and the listener’s980understanding of the entire communicative situation. Understanding speech is981thusinherentlysituation-specific(seealsoHawkins,2011;Hawkins,2014).982

Thethirdtenetthatourresultssupportisthatfinephoneticdetailwillbeusedto983accessmeaningwhenitisrelevanttothesituationathand.Thereductioninearly984looks to the target for dis- andmis- stimuli indicates that listeners used the985information in the critical syllable at the start of the experiment but quickly986adaptedtothefactthatthetaskdidnotdemandit—theycouldwaituntillaterin987thesentenceforacousticallyclearerdisambiguation.Thisrapidadaptationtotask988requirementsissupportedbytheliteratureonperceptuallearningandadaptation989tonewaccentsetc.(e.g.Maye,Aslin&Tanenhaus,2008;Bradlow&Bent,2008;990Barden&Hawkins,2013;NguyenandHawkins,2001).991

Thefourthtenetisthat“lackofclearevidenceforaparticularcategory,aswith992thereducedfirstsyllableinmistakes,canbeinformative”(Hawkins,2010,p486).993Thatis,itwasexpectedthattheabsenceofaperceptualbeatonunprefixedforms994of the critical syllables mis-, dis- and re- would help listeners predict995


page28of44

monomorphemicword identity. Thiswas broadly the case. However, as noted996above,intheearlypartoftheexperimentandafterabout500msfromwordonset,997true prefixes helped word identification more than nonprefixes did. Since the998prefixcontainsaheavierbeatthanthenon-prefix,weinterpretthislastresultas999indicating that a perceptual beat is important in driving perceptual decisions,1000whereasitsabsencemaybelessinfluential.Thisisconsistentwithneuroscientific1001studiesshowingthatabeat-basedhierarchyisfundamentaltoselectiveattention1002(Lakatosetal.,2007;Lakatosetal.,2008;Lakatosetal.,,2009;Arnal&Giraud,10032012). An obvious inference is that the prefix is emphasized because it bears1004quasi-independent information about meaning that the phonemically-identical1005nonprefixdoesnot.1006

Ifprefixescarrysecondarystresstodrawattentiontothemandtheirmeaning,1007whyaresuffixesnotsimilarlystressed?Onepossibilitytakesusagaintotherole1008ofpredictioninunderstandingspeech.Understoodwithinthenaturalcontextof1009theirmetricalstructure,prefixesheraldthebeginningofanewlexicalitemthat1010willbeapolysyllabicwordwithamainlexicalstresslaterthantheprefix,andthe1011prefixwillchangethemeaningofthestem.Therearenumeroussubtlechangesto1012segmentaldurationsinpolysyllabicwordsthatdependontheword’sphonological1013and acoustic structure (see Hay et al. (in prep.) for information pertinent to1014prefixes). Suffixes make the word phonologically longer, by adding either a1015syllableorelseasegmentthatmakesthefinalcodamorecomplex.Wordstems1016arelikewisesubtlymodifiedbytheadditionofoneormoresuffixes,andthereare1017differences between words with different types of suffix, compared with1018monomorphemicwords. For example, Plag,Homann, andKunter (2015/2017)1019discusscomplexitiesofregularitiesfoundforthevariousmorphemesrepresented1020byEnglish/s/and/z/(plural,genitive,3rdpersonsingular,etc).1021

Ifthestressonaprefixheraldsapolymorphemicword,itseemsreasonablethat1022thechangesthatasuffixedwordstemundergocouldraisetheprobabilityofan1023upcomingsuffix.PerceptualexperimentsdescribedinSection1.2showthatthey1024do for both English andDutch (Blazej & Cohen-Goldberg, 2014; Kemps et al.,10252005a;Kempsetal.,2005b).However, asnoted there, given that thesestudies1026contrasted monosyllabic, mono-morphemic words with polysyllabic,1027polymorphemicwords,itisimpossibletotellwhethertheirlistenersweresimply1028anticipatinga longerword, or anticipating thepolymorphemic structureof the1029longerword,orboth.1030

Whatisclear,however,isthatthesuffixitselfdoesnot“need”toreceivesecondary1031stressbecauseitspresenceispredictablefromtheinternalacousticstructureof1032itsstem,whichwouldtypicallytakestrongeststressintheword.Togetherwitha1033constraintagainststressingword-finalsyllablesofpolysyllabicwordsinEnglish1034andDutch, and, presumably, usually some grammatical priming, thismight be1035sufficienttoreduceperceptualuncertainty.1036

Insum,thisstudysupportspreviousclaimsandhypothesesthatlistenersusefine1037phonetic detail in real time to efficiently access meaning, but only in its1038appropriatecontextandifthetaskmakesitrelevantand‘cost-effective’todoso.1039


page29of44

The previous section concluded by implying that rhythm and metre have the1040potential toprovidetheunderlying ‘glue’ofspeechcommunicationby focusing1041attention onto critical events in the speech stream. Those critical events are1042associatedwithbeatsthatallowametricalstructuretobecreatedbythelistening1043brain. The metrical structure facilitates prediction and allows meaning to be1044efficientlyaccessed.Thissectionconcludesbybrieflyaddingtwopoints to that1045claim.Thefirstisthatmuchofthespeechsignalisofcoursecrucialtointelligibility1046butdoesnotreceiveametricalbeat.Howmight that ‘non-beat’ informationbe1047processed?Hawkins (2010)suggested thatphoneticdetail in theentirespeech1048signal is continuouslymonitored, for a varietyof reasons including inorder to1049learnaboutcommunicativelysignificantnewpatterns.Thispositionissupported1050byneuroscientificevidencerelatedtothatcitedabovedemonstratingcreationof1051beat-based metrical structure. For example, Schroeder and Lakatos (2009)1052propose that when a stimulus lacks rhythm, lower-frequency neuroelectric1053oscillations entrained to metrical structure are suppressed and replaced by1054continuousmonitoring(vigilance)thatuseshigher-frequencyoscillations.These1055systems can operate simultaneously, differing in balance depending on the1056rhythmicityof thestimulus.Thisclaimisalsoconsistentwithexperimentsthat1057showthatlexicalactivationvariescontinuouslyinawaywhichreflectsvariation1058in the acoustic signal as it unfolds over time (e.g. Allopenna, Magnuson, &1059Tanenhaus, 1998; Gow & McMurray, 2007; McMurray et al., 2003; Warren &1060Marslen-Wilson,1987).1061

1062The second concluding point is that speech normally takes place as part of1063meaningfulcommunicationbetweenpeople.Anyhypothesisconcerningtherole1064ofrhythmandmetreinunderstandingspeechneedstoencompasstheinteractive1065andmultimodalpropertiestypicalofmosthumancommunication.Thepowerof1066multimodal sensory information in facilitating speech intelligibilityand spoken1067communicationiswellknown.Thestrongcorrelationbetweentheauditorysignal1068and visual input from the gestural code and facial expression is equally well1069known.Thereisampleevidencethatrhythmandmetricalstructureplayacrucial1070role in both multimodal integration of a message from a single talker (e.g.1071Schroederetal.,2008)andintheentrainmentthatoccursduringcommunication1072between talkers (e.g.Hassonet al., 2012).Detaileddiscussiongoesbeyond the1073scopeofthispaper,butageneralreviewrelevanttorhythminbothspokenand1074musicalinteractioncanbefoundinHawkins,CrossandOgden(2013).1075

4.3 Strengths,limitationsandextensionsofthestudy1076

Weachievedouraimofdemonstratingthatarhythmicdistinctionoffinephonetic1077detail,withnophonemiccontrast involved,canbeusedtoaccessmeaningand1078predictlexicalidentityinrealtime.Weusedthemorphologicalprefixdistinction1079becauseitsuitedouraimswell,sincethereisrelativelygoodconsensusaboutthe1080meaning of a prefix. Other contrasts could have been used, but few lend1081themselvesaswelltoourprimaryquestion.1082

Unlikepreviousstudies,wecontrolledforconfoundingfactorslikethenumberof1083syllables in themono-morphemicvs.prefixedwords,andprosodicstructureof1084pairsofstimuli.Weputmucheffortintoensuringthatourstimulidescribedand1085


page30of44

illustratedplausiblevisualscenes.Bytrackingourresultsoverthecourseofthe1086experiment we were also able to show that the experiment itself changed1087participants’behaviour.Thispointstoanimportantmethodologicalconsideration1088forfuturestudies.Adaptationtoexperimentalconditionscanoccurwithjustafew1089trials. Researchers should consider this possibility in their analyses before1090concluding that a manipulation did not affect participants’ behaviour, since a1091weakeningoftheeffectoverthecourseoftheexperimentcanobscureimportant1092resultsrelevanttounderstandingspeechinnormal,everydaysituations.Itshould1093also be noted that we found our effect weakened over the course of the1094experiment,eventhoughthestimuliincludedahighproportionoffillertrials,a1095subset of which were designed to counteract effects of cross-spliced critical1096syllables.1097

We expect these results to generalize to most if not all English prefixes. The1098phoneticdetailwillbespecifictotheprefix,andsomewordsandcontextsmay1099differ from the usual pattern. Word-specific influences could include1100decomposability,andtherelativefrequencyoftheprefixedandunprefixedword.1101But theprincipleofmorevs. less stress (or lessvs.moresyllable reduction) is1102expectedtobetrueforallprefixesandpseudoprefixes.Weexpectsomedialect1103differencesintheexactacousticdetails:workinprogresssupportsthisprediction1104whileconfirmingthatthegeneralpatternsholdacrossdialectsofEnglish(Hayet1105al.,inprep).Intermsofwidertheoreticalimplications,thesametypeofreasoning1106canpresumablybegeneralizedtoperceptionofanyotheraudiblecontrastbased1107onnon-phonemicphoneticdetail,aslongastheauditorycontrasthasasystematic1108relationshipwithdistinctionsofmeaning,broadlydefined.1109

Aquestionforfutureresearchiswhetherthemultipleacousticcuesweandothers1110haveidentifiedascontributingtotheprefix-nonprefixdistinctionworktogether1111in concert, or whether any dominate perceptual responses. Perhaps the most1112valuablequestionwecanaskpertainstotheroleofdurationinthesyllablesdis-1113andmis.Syllabledurationisoftenshowntobeanover-ridingperceptualcue(e.g.1114Salverda,Dahan&McQueen(2003)Experiment2).Butfordistinguishingprefix1115andnon-prefixformsofdis-andmis-,iswhatmattersthedurationofthewhole1116syllable,regardlessofitsinternalacousticstructure(e.g.thes:ɪratio),orisitthe1117internal acoustic structure regardless (within limits) of the overall syllable1118duration?We hypothesize that for SSBE, it is likely to be the internal acoustic1119structure,ascarriedbythes:ɪratio,sincethatratio isdistinctive inproduction1120(Smith,BakerandHawkins2012).Wesuggestthisbecausethereismorescope1121forarelativelylongvoweltoconveyastrongerrhythmicbeat(indicativeofthe1122prefix form),byvirtueofamplitude, f0and formantspacing, than for theshort1123vowel that tends to accompany the non-prefix form. In sum, the rhythmic1124hypothesiswould be supported if the internal acoustic structure provedmore1125decisive in indicatingprefixstatusthanoverallsyllabledurationalone,without1126anychangeins:ɪratio.Thisexperimentisplanned.1127

1128

4.4 Concludingsummary1129

This study confirms that phonetic detail associated with prefixes and pseudo1130prefixescanaidpredictionoftheupcomingword’sidentity.Ourresults,especially1131


page31of44

from early trials, indicate that such phonetic detail will typically be used1132predictivelyinreal-worldlisteningconditions.However,wehavealsoshownthat1133over the course of our experiment many of the effects weakened, indicating1134relatively fast adaptation to the experimental conditions. Therewas also some1135evidencethatprefixesinfluencepredictivebehaviormorestronglythanpseudo1136prefixes do. We suggest that our findings support the hypothesis that speech1137rhythm (more properly, themetre of speech) provides a fundamental binding1138principleof speechprocessing, enabling linguistic structures tobe created and1139matchedwithsimilarstructuresinmemorytoallowrapidmatchingofcomplex1140soundpatternstomeaning.1141

SupplementaryMaterials1142Stimuli,originaldata,supplementaryfigures,Rcodefortheanalyses,and1143additionalacousticanalysesofthestimulinotreportedinthemanuscriptare1144availableathttps://osf.io/dsyxu. DOI 10.17605/OSF.IO/DSYXU. 1145 114611471148Acknowledgements1149

TheauthorswouldliketothankGerryAltmannfortheuseoftheeye-trackersand1150lab space; Elaine Tham for data collection; and Marco Piccolino-Boniforti for1151extensive help with stimulus construction and for many helpful discussions.1152Portions of this work were presented at the 17th International Congress of1153Phonetic Sciences, 2011, Hong Kong, the Colloquium of British Association of1154AcademicPhoneticians,Leeds,UK,2012,andAMLaP2012,RivadelGarda,Italy.1155This research was supported by the Marie Curie Research Training Network1156Sound to Sense,MRTN-CT-2006-035561, including fellowships toClayards and1157Piccolino-Boniforti. 1158


page32of44

AppendixA1159

Theterm‘phoneticdetail’asusedinthispaperhasaveryparticularmeaning.It1160refers to acoustic-phonetic properties that are systematically distributed and1161communicativelysignificantbutthatarenotessentialtodifferentiatephonemes,1162andhencetodistinguishthephonologicalformofwords.Thisdefinitionhidesa1163numberofcomplexities,discussedforexamplebyCarlsonandHawkins(2007)1164andHawkinsandLocal(2007),butthemainpointsforpresentpurposesare(1)1165thatspeechsoundscansystematicallydistinguishmeaningsandcommunicative1166functionswithouttherebeingadifferenceinphonemicstructure;and(2)thatthe1167perceptualsignificanceandhencemeaningofanysinglepartofthespeechsignal1168depends on the situational andphonetic context inwhich it is heard. The first1169claim is widely accepted, although the meanings concerned (e.g. question vs.1170statement intonation, expressions of doubt) were traditionally designated1171‘paralinguistic’ and treated separately both from the types of lexical meaning1172distinguished in phonemic analysis, and from mainstream psycholinguistic1173theoriesofspokenwordrecognition;theaccountproposedheremakesnosuch1174restriction, as exemplified by our focus on grammatical (specifically1175morphological)linguisticstructure.Thesecondclaimislikewisewidelyaccepted,1176but its implications do not always drive the theoretical interpretation of1177experimentalresults,northedesignofexperimentstestingtheroleofphonetic1178detailinspeechperception.11791180

1181

1182

1183


page33of44

AppendixB1184

Sentencepairsusedintheexperiment.Thefirstsentenceofeachpaircontainsthe1185pseudoprefixandthesecondsentenceofeachpaircontainsthetrueprefix.1186

Iwouldn’tbesurprisediftheboysdiscoverthem.1187Iwouldn’tbesurprisediftheboysdiscolourthem.11881189Hefellasleepdespiteallthediscussion.1190Hefellasleepdespiteallthediscomfort.11911192Theteacherhasaverydiscursivestyle.1193Theteacherhasaverydiscourteousstyle.(3-syllablepronunciationoftargetword)11941195Herpaintingsaresodistinctive.1196Herpaintingsaresodistasteful.11971198ItwasdifficultbecauseSamdistractedhim.1199ItwasdifficultbecauseSamdistrustedhim.12001201Aswandisplaysitsplumagetoitsmate.1202Aswandisplaceswaterwhenitlands.12031204Alextypicallydiscardsthefruit.1205Alextypicallydiscountstherisk.12061207I'dbesurprisedifTessmistakestheletters.1208I'dbesurprisedifTessmistypestheletters.12091210Wefeltuncomfortableabouthismysteriousdemeanour.(3-syllablepronunciation)1211WefeltuncomfortableabouthismistreatmentofAmina.12121213ThegirlswerespellboundbytalesofJo'smystique.1214ThegirlswerespellboundbytalesofJo'smisdeeds.12151216WethinkJeffmistookthetreeforaperson.1217WethinkJeffmistimedtheturningonpurpose.12181219Jostruggledtorecoverherbalance.1220Joestruggledtore-coverthesofa.12211222Theyagreedtheyshouldrepealtheverdict.1223Theyagreedtheyshouldre-peelthecarrots.12241225That'sOscarWilde.Helovedreposingquietly.1226That'sOscarWilde.Helovedre-posingquestions.12271228Afterthemassacre,thearmedforcesreformedtheirprocedures.1229Afterthemassacre,thearmedforcesre-formedonthehillside.12301231


page34of44

Afterthenoisylawnparty,Joshrecededbehindtheshed.1232Afterthenoisylawnparty,Joshre-seededthetrampledlawn.12331234EveryonewashappierafterGeoffrestrainedthebrute.1235EveryonewashappierafterGeoffre-strainedthefruit.12361237Hehurriedtorelaythemessage.1238Hehurriedtore-laythecarpet.12391240They'restartingtoredressthewrong.1241They'restartingtore-dressthewound.12421243Wehopedhe'dreleasethecatchsoon.1244Wehopedhe'dre-leasethehousesoon.12451246Thenextjobwastorepairthesocks.1247Theynextjobwastore-pairthesocks.12481249WewatchedJessrestorethem.1250WewatchedJessre-storethem.12511252Hewaspunishedforrefusingsorudely.1253Hewaspunishedforre-fuellingsoslowly.12541255Themanrejoicedashefinishedtherace.1256Themanre-joinedtheendsoftherope.12571258Toddrebuttedtheargumentsuccessfully.1259Toddre-baitedthefishinglinesuccessfully.12601261Harry'sparentsrevokedhisprivileges.1262Harry'sparentsrevoicedhisworriesagain.12631264WeknowthatDaverestrictshisarmmovementswhennecessary.1265WeknowthatDavere-stringshisinstrumentwhennecessary.12661267It'saperfectexampleofextravaganceinpublicspending.1268It'saperfectexampleofex-trampoliners'senseofbalance.12691270Wewereamusedtohearthoseexpletiveshadbeencensored.1271Wewereamusedtohearthoseex-policemenhadbeenhonoured.12721273Thereareconflictingviewsabouttheseexpanders'rolesinorthodontistry.1274Thereareconflictingviewsabouttheseexpatriates'rolesinthissociety.1275

(3-syllablepronunciationoftargetword)12761277Theydidn'tunderstandwhytheseexponencytermswereimportant.1278Theydidn'tunderstandwhytheseex-ponyclubgirlswereimportant.12791280


page35of44

Thejudgedecidedhe'dexpatiateatsomelength.1281Thejudgedecidedhe'dexpatriatethepoorkids.128212831284AppendixC128537pairsofFiller itemsspecific tothisexperiment(Fortheother30pairsof filler1286items,seeHeinrich,FloryandHawkins,2010.)Wordsinboldcontaineitheratrue1287(tr)orapseudo(ps)prefix12881289Wetriedtodistinguishbetweenthetwins(ps)1290Wetriedtodistinguishbetweenthebins12911292Youpurifywaterbydistillingit(ps)1293Youpurifywhiskybydistillingit12941295Welikedthedescriptionoftheballoonsovermountains(ps)1296Welikedthedescriptionofthefantasticaldragon12971298Theconductorloveshisjobdespitebeingpronetobackache(ps)1299Theconductorlikeshisjobdespitebeingpronetomotionsickness13001301Wecouldjustdiscernthebridgesinthefog(ps)1302Wecouldjustdiscerntheridgesinthefog13031304Suedisturbedthecows(ps)1305Suedisturbedthesheep13061307Hebegantodestroythedoor(ps)1308Hebegantodestroythecube13091310Thevandalsdistortedtheframetogetrevenge(ps)1311Thevandalsdistortedthewheeltogetrevenge13121313Theywereallimpressedwiththedisabledgirl'sspirit(ambiguouslytr)1314Theywereallimpressedwiththedisabledboy'sspirit13151316Sugardissolvesfasterinhotliquids(ps+pronunciationchange)1317Sugardissolvesfasterwhenyoustirit13181319Themanhadseriouslymistreatedthedonkey(tr)1320Themanhadseriouslymistreatedthedog13211322Thedrawerwasmisaligned(tr)1323Thedoorwasmisaligned13241325TheLordofMisruleasapuppet(tr)1326TheLordofMisruleasacarving13271328Amistrialisarollerblader'strick(tr)1329


page36of44

Amistrialisalawyer’slastresort13301331Jodyhadmiscalculatedwhentotakethetoastout(tr)1332Jodyhadmiscalculatedwhentojumpforthefrisbee13331334Thesheepdidnotreacttothefox'spresence(tr)1335Thesheepdidnotreacttothedog'spresence13361337Arefectory'swheremonkseat(ps)1338Arefectory'swheremonkssleep13391340Wewereslowtorepackbecauseofthebaby(tr)1341Wewereslowtorepackbecauseofthedog13421343Thedogwasreluctanttorelinquishtheball(ps)1344Thedogwasreluctanttorelinquishthering13451346Theclassredrewclassicalcartoons(tr)1347Theclassredrewclassicalplans13481349Katerefusedtobuythecoat(ps)1350Katerefusedtobuythedress13511352We'llrepaintthedhow(tr)1353We'llrepaintthebow13541355Theydidagoodjobofre-creatingtheoriginalinstruments(tr)1356Theydidagoodjobofre-creatingtheVictorianatmosphere13571358Herecitedthepoemperfectly(ps)1359Herecitedthepoempassionately13601361Hisfavoritewasthisrepeatingrifle(ps)1362Hisfavoritewasthisrepeatingpattern13631364Theyplannedtoreheattherisotto(tr)1365Theyplannedtoreheatthelasagna13661367Cameronre-setsthestone(tr)1368Cameronre-setsthebone13691370Aliexaminedthebookintently(ps)1371Aliexaminedthepaintintently13721373Herex-husbandisadiver(tr)1374Herex-husbandisadriver13751376Themenexchangedlooks(ps)1377Themenexchangedbooks1378


page37of44

1379She'sareallyexcellentmusician(ps)1380She'sareallyexcellentclinician13811382Nothinglikeextortingpromises(ps)1383Nothinglikeextortingmoney13841385Geoffextractedthetooth(ps)1386Geoffextractedthejuice13871388Therewasamountainofplasticrecycling(tr?)1389Therewasamountainofplasticforplayingin13901391Sallylikedmeetingalltherelations(ps)1392Sallylikeseeingthecelebrations13931394Eddiealwaystakesrevisingseriously(ps)1395Eddiealwaystakeshisdrivingseriously13961397Luketriedhardnottoeatthecake1398Luketriedhardnottoeattheicecream139914001401References14021403Allopenna,P.D.,Magnuson,J.S.,&Tanenhaus,M.K.(1998).Trackingthetime1404

courseofspokenwordrecognitionusingeyemovements:Evidencefor1405continuousmappingmodels.JournalofMemoryandLanguage,38,419-1406439.https://doi.org/10.1006/jmla.1997.25581407

Arnal,L.H.,&Giraud,A.-L.(2012).Corticaloscillationsandsensorypredictions.1408TRENDSinCognitiveSciences,16(7),390-398.1409doi:10.1016/j.tics.2012.05.0031410

Baayen,R.H.,Levelt,W.M.J.,Schreuder,R.,&Ernestus,M.(2008).Paradigmatic1411structureinspeechproduction.ChicagoLinguisticsSociety,43(1),1-29.1412

Baker,R.(2008).Theproductionandperceptionofmorphologicallyand1413grammaticallyconditionedphoneticdetail.UniversityofCambridge,1414Cambridge.1415

Barden,K.J.(2011).Perceptuallearningofcontext-sensitivephoneticdetail.1416Doctoralthesis,UniversityofCambridge).doi:10.17863/CAM.165011417

Barden,K.J.,&Hawkins,S.(2013).Perceptuallearningofphoneticinformation1418thatindicatesmorphologicalstructure.Phonetica,70(4),323-342.1419https://doi.org/10.1159/0003572331420

BatesD,MächlerM,BolkerB,WalkerS(2015).FittingLinearMixed-Effects1421ModelsUsinglme4.JournalofStatisticalSoftware,67(1),1–48.doi:142210.18637/jss.v067.i01.1423

Beckman,M.E.,&Pierrehumbert,J.B.(1986).IntonationalstructureinJapanese1424andEnglish.PhonologyYearbook,3,255-309.1425doi:10.1017/S095267570000066X1426


page38of44

BenHedia,S.&Plag,I.(2017).GeminationanddegeminationinEnglish1427prefixation:Phoneticevidenceformorphologicalorganization.Journalof1428Phonetics,62,34-49.https://doi.org/10.1016/j.wocn.2017.02.0021429

Blazej,L.J.,&Cohen-Goldberg,A.M.(2015).Canwehearmorphological1430complexitybeforewordsarecomplex?JournalofExperimental1431Psychology:HumanPerceptionandPerformance,41(1),50-68.1432doi:10.1037/a00385091433

Borsky,S.,Shapiro,L.P.,&Tuller,B.(2000).Thetemporalunfoldingoflocal1434acousticinformationandsentencecontext.JournalofPsycholinguistic1435Research,29(2),155-168.https://doi.org/10.1023/A:10051409274421436

Borsky,S.,Tuller,B.,&Shapiro,L.P.(1998)."Howtomilkacoat:"Theeffectsof1437semanticandacousticinformationonphonemecategorization.Journalof1438theAcousticalSocietyofAmerica,103,2670-2676.1439https://doi.org/10.1121/1.4227871440

Bradlow,A.R.,&Bent,T.(2008).Perceptualadaptationtonon-native1441speech.Cognition,106(2),707-729.1442https://doi.org/10.1016/j.cognition.2007.04.0051443

Breen,M.,Dilley,L.C.,McAuley,J.D.,&Sanders,L.D.(2014).Auditoryevoked1444potentialsrevealearlyperceptualeffectsofdistalprosodyonspeech1445segmentation.Language,CognitionandNeuroscience,29(9),1132-1146.1446doi:10.1080/23273798.2014.8946421447

Brown,M.,Salverda,A.P.,Dilley,L.C.,&Tanenhaus,M.K.(2015).Metrical1448expectationsfromprecedingprosodyinfluenceperceptionoflexical1449stress.JournalofExperimentalPsychology:HumanPerceptionand1450Performance,41(2),306-323.doi:10.1037/a00386891451

Calderone,D.J.,Lakatos,P.,Butler,P.D.,&Castellanos,F.X.(2014).Entrainment1452ofneuraloscillationsasamodifiablesubstrateofattention.TRENDSin1453CognitiveSciences,18(6),300-309.1454https://doi.org/10.1016/j.tics.2014.02.0051455

Carlson,R.,&Hawkins,S.(2007).Whenisfinephoneticdetailadetail?InW.J.1456Barry&J.Trouvain(Eds.),16thInternationalCongressofPhoneticSciences1457(pp.211-214,PaperID1721).Saarbrücken:http://www.icphs2007.de/.1458

Cho,T.(2001).Effectsofmorphemeboundariesonintergesturaltiming:1459EvidencefromKorean.Phonetica,58,129-162.1460https://doi.org/10.1159/0000561961461

Cho,T.(2016).Prosodicboundarystrengtheninginthephonetics–prosody1462interface.LanguageandLinguisticsCompass,10(3),120-141.1463doi:10.1111/lnc3.121781464

Cirelli,L.K.,Wan,S.J.,&Trainor,L.J.(2016).Socialeffectsofmovement1465synchrony:Increasedinfanthelpfulnessonlytransferstoaffiliatesof1466synchronouslymovingpartners.Infancy,21(6),807-821.1467doi:10.1111/infa.12140.1468

Clayards,M.,Niebuhr,O.,&Gaskell,M.G.(2015).Thetimecourseofauditoryand1469language-specificmechanismsincompensationforsibilant1470assimilation.Attention,Perception,&Psychophysics,77(1),311-328.1471https://doi.org/10.3758/s13414-014-0750-z1472

Cutler,A.,&Butterfield,S.(1992).Rhythmiccuestospeechsegmentation:1473Evidencefromjuncturemisperception.JournalofMemoryandLanguage,147431,218-236.https://doi.org/10.1016/0749-596X(92)90012-M1475


page39of44

Dahan,D.,Magnuson,J.S.,Tanenhaus,M.K.,&Hogan,E.M.(2001).1476Subcategoricalmismatchesandthetimecourseoflexicalaccess:Evidence1477forlexicalcompetition.LanguageandCognitiveProcesses,16(5/6),507-1478534.https://doi.org/10.1080/016909601430000741479

Dahan,D.,&Mead,R.L.(2010).Context-conditionedgeneralizationinadaptation1480todistortedspeech.JournalofExperimentalPsychology:HumanLearning1481andMemory,36(3),704-728.https://doi.org/10.1037/a00174491482

Davis,M.H.,DiBetta,A.M.,Macdonald,M.J.E.,&Gaskell,M.G.(2009).Learning1483andconsolidationofnovelspokenwords.JournalofCognitive1484Neuroscience,21(4),803-820.https://doi.org/10.1162/jocn.2009.210591485

Davis,M.H.,Marslen-Wilson,W.,&Gaskell,M.G.(2002).Leadingupthelexical1486garden-path:segmentationandambiguityinspokenwordrecognition.1487JournalofExperimentalPsychology-HumanPerceptionandPerformance,148828,218-244.https://doi.org/10.1037/0096-1523.28.1.2181489

Dilley,L.C.,Mattys,S.L.,&Vinke,L.(2010).Potentprosody:Comparingthe1490effectsofdistalprosody,proximalprosody,andsemanticcontextonword1491segmentation.JournalofMemoryandLanguage,63(3),274-294.1492doi:http://dx.doi.org/10.1016/j.jml.2010.06.0031493

Ernestus,M.(2014).Acousticreductionandtherolesofabstractionsand1494exemplarsinspeechprocessing.Lingua,142,27-41.1495

Ernestus,M.,Baayen,H.,&Schreuder,R.(2002).Therecognitionofreduced1496wordforms.BrainandLanguage,81(1-3),162-173.1497

Ernestus,M.,Hanique,I.,&Verboom,E.(2015).Theeffectofspeechsituationon1498theoccurrenceofreducedwordpronunciationvariants.Journalof1499Phonetics,48,60-75.doi:http://dx.doi.org/10.1016/j.wocn.2014.08.0011500

Fujioka,T.,Trainor,L.J.,Large,E.W.,&Ross,B.(2012).Internalizedtimingof1501isochronoussoundsisrepresentedinneuromagneticBetaoscillations.1502JournalofNeuroscience,32(5),1791-1802.1503

Giraud,A.-L.,&Poeppel,D.(2012).Corticaloscillationsandspeechprocessing:1504emergingcomputationalprinciplesandoperations.NatureNeuroscience,150515(4),511-517.doi:10.1038/nn.30631506

Gow,D.W.,&McMurray,B.(2007).Wordrecognitionandphonology:Thecaseof1507Englishcoronalplaceassimilation.InJ.Cole&J.I.Hualde(Eds.),Papersin1508LaboratoryPhonology9:PhonologyandPhonetics(pp.173-200).Berlin:1509MoutondeGruyter.1510

Hannon,E.E.,&Trehub,S.E.(2005).Metricalcategoriesininfancyand1511adulthood.PsychologicalScience,16(1),48-55.1512

Hasson,U.,Ghazanfar,A.A.,Galantucci,B.,Garrod,S.,&Keysers,C.(2012).Brain-1513to-braincoupling:Amechanismforcreatingandsharingasocialworld.1514TRENDSinCognitiveSciences,16(2),114-121.1515doi:10.1016/j.tics.2011.12.0071516

Hawkins,S.(2003).Rolesandrepresentationsofsystematicfinephoneticdetail1517inspeechunderstanding.JournalofPhonetics,31,373-405.1518

Hawkins,S.(2010).Phoneticvariationascommunicativesystem:Perceptionof1519theparticularandtheabstract.InC.Fougeron,B.Kühnert,M.d'Imperio&1520N.Vallée(Eds.),LaboratoryPhonology10:Variability,PhoneticDetailand1521PhonologicalRepresentation(pp.479-510).Berlin:MoutondeGruyter.1522

Hawkins,S.(2011).Doesphoneticdetailguidesituation-specificspeech1523recognition?Keynoteaddress,Proceedingsofthe17thInternational1524


page40of44

CongressofPhoneticSciences.HongKong:1525http://www.icphs2011.hk/ICPHS_CongressProceedings.htm1526

Hawkins,S.(2014).Situationalinfluencesonrhythmicityinspeech,music,and1527theirinteraction.InR.Smith,T.Rathcke,F.Cummins,K.Overy,&S.K.1528Scott(Eds.),CommunicativeRhythmsinBrainandBehaviour(Vol.369,pp.152920130398).London:PhilosophicalTransactionsoftheRoyalSocietyB153020130398.1531

Hawkins,S.,Cross,I.,&Ogden,R.(2013).Communicativeinteractionin1532spontaneousmusicandspeech.InM.Orwin,C.Howes,&R.Kempson1533(Eds.),Language,MusicandInteraction(Vol.3,pp.285-329).London:1534CollegePublications.1535

Hawkins,S.,&Local,J.K.(2007).SoundtoSense:Introductiontothespecial1536session.InW.J.Barry&J.Trouvain(Eds.),16thInternationalCongressof1537PhoneticSciences(Vol.,pp.181-184,PaperID1726).Saarbrücken:1538http://www.icphs2007.de/.1539

Hawkins,S.,&Nguyen,N.(2001).Perceptionofcodavoicingfrompropertiesof1540theonsetandnucleusof'led'and'let'.InP.Dalsgaard,B.Lindberg&H.1541Benner(Eds.),Eurospeech2001:Proceedingsofthe7thInternational1542ConferenceonSpeechCommunicationandTechnology,(Vol.1,pp.407-1543410).Scandinavia.1544

Hawkins,S.,&Smith,R.H.(2001).Polysp:Apolysystemic,phonetically-rich1545approachtospeechunderstanding.ItalianJournalofLinguistics-Rivistadi1546Linguistica,13,99-188.1547http://linguistica.sns.it/RdL/13.1/Hawkins_Smith.pdf1548

Hay,J.B.(2003).Causesandconsequencesofwordstructure.NewYork:1549Routledge.1550

Hay,J.,&Baayen,R.H.(2005).Shiftingparadigms:gradientstructurein1551morphology.TRENDSinCognitiveSciences,9(7),342-348.1552

Hay,J.,&Drager,K.(2010).Stuffedtoysandspeechperception.Linguistics,48(4),1553865-892.1554

Hay,J.,Hawkins,S.,Stuart-Smith,J.Smith,R.,Fromont,R.(inpreparation)The1555morphophoneticsofword-initialdisandmisinfourvarietiesofEnglish.1556

Heffner,C.C.,Dilley,L.C.,McAuley,J.D.,&Pitt,M.A.(2013).Whencuescombine:1557Howdistalandproximalacousticcuesareintegratedinword1558segmentation.LanguageandCognitiveProcesses,28(9),1275-1302.1559doi:10.1080/01690965.2012.6722291560

Heinrich,A.,Flory,Y.,&Hawkins,S.(2010).InfluenceofEnglishr-resonanceson1561intelligibilityofspeechinnoisefornativeEnglishandGermanlisteners.1562SpeechCommunication,52,1038-1055.1563

Kemps,R.,Ernestus,M.,Schreuder,R.,&Baayen,R.H.(2005).Prosodiccuesfor1564morphologicalcomplexity.MemoryandCognition,33,430-446.1565

Kemps,R.,Wurm,L.H.,Ernestus,M.,Schreuder,R.,&Baayen,R.H.(2005).1566ProsodiccuesformorphologicalcomplexityinDutchandEnglish.1567LanguageandCognitiveProcesses,20(1/2),43-73.1568doi:10.1080/016909604440002231569

Kingston,J.,Levy,J.,Rysling,A.,&Staub,A.(2016).Eyemovementevidencefor1570animmediateGanongeffect.JournalofExperimentalPsychology:Human1571PerceptionandPerformance,42(12),1969-1988.1572doi.org/10.1037/xhp00002691573


page41of44

Krieger-Redwood,K.,Gaskell,M.G.,Lindsay,S.,&Jefferies,E.(2013).The1574selectiveroleofpremotorcortexinspeechperception:Acontributionto1575phonemejudgementsbutnotspeechcomprehension.JournalofCognitive1576Neuroscience,25(12),2179-2188.doi:10.1162/jocn_a_004631577

KuznetsovaA,BrockhoffPBandChristensenRHB(2017).“lmerTestPackage:1578TestsinLinearMixedEffectsModels.”JournalofStatistical1579Software,82(13),pp.1–26.doi:10.18637/jss.v082.i13.1580

Lakatos,P.,Shah,A.S.,Knuth,K.,H,Ulbert,I.,Karmos,G.,&Schroeder,C.E.1581(2005).Anoscillatoryhierarchycontrollingneuronalexcitabilityand1582stimulusprocessingintheauditorycortex.JournalofNeurophysiology,94,15831904-1911.doi:10.1152/jn.00263.20051584

Lakatos,P.,Chen,C.-M.,O'Connell,M.N.,Mills,A.,&Schroeder,C.E.(2007).1585NeuronalOscillationsandMultisensoryInteractioninPrimaryAuditory1586Cortex.Neuron,53(2),279-292.doi:10.1016/j.neuron.2006.12.011.1587

Lakatos,P.,Karmos,G.,Mehta,A.D.,Ulbert,I.,&Schroeder,C.E.(2008).1588EntrainmentofNeuronalOscillationsasaMechanismofAttentional1589Selection.Science,320(5872),110-113.doi:10.1126/science.11547351590

Lakatos,P.,O'Connell,M.N.,Barczak,A.,Mills,A.,Javitt,D.C.,&Schroeder,C.E.1591(2009).TheLeadingSense:SupramodalControlofNeurophysiological1592ContextbyAttention.Neuron,64(3),419-430.1593doi:10.1016/j.neuron.2009.10.0141594

Lee-Kim,S.I.,Davidson,L.,&Hwang,S.(2013).Morphologicaleffectsonthe1595darknessofEnglishintervocalic/l.LaboratoryPhonology,4(2),475-511.1596

Lehiste,I.(1972).Mannerofarticulation,parallelprocessing,andtheperception1597ofduration,ComputerandInformationScienceResearchCenterWorking1598PapersinLinguistics(Vol.12,pp.33-52):OhioStateUniversity.1599

Lenth,R.V.(2016).Least-SquaresMeans:TheRPackagelsmeans.Journalof1600StatisticalSoftware,69(1),1-33.doi:10.18637/jss.v069.i011601

London,J.(2012).HearinginTime(2nded.).Oxford:OxfordUniversityPress.1602Mackenzie,S.,Olson,E.,Clayards,M.,&Wagner,M.(2018).NorthAmerican/l/1603

bothdarkensandlightensdependingonmorphologicalconstituencyand1604segmentalcontext.LaboratoryPhonology:JournaloftheAssociationfor1605LaboratoryPhonology,9(1),13.DOI:http://doi.org/10.5334/labphon.1041606

Mai,G.,Minett,J.W.,&Wang,W.S.Y.(2016).Delta,theta,beta,andgammabrain1607oscillationsindexlevelsofauditorysentenceprocessing.Neuroimage,1608133,516-528.doi:10.1016/j.neuroimage.2016.02.0641609

Matin,E.,Shao,K.,&Boff,K.(1993).Saccadicoverhead:Informationprocessing1610timewithandwithoutsaccades.Perception&Psychophysics,53,372–380.1611

Mattys,S.L.,Jusczyk,P.W.,Luce,P.A.,&Morgan,J.L.(1999).Phonotacticand1612ProsodicEffectsonWordSegmentationinInfants.CognitivePsychology,161338(4),465-494.doi:https://doi.org/10.1006/cogp.1999.0721.1614

Mattys,S.,White,L.,&Melhorn,J.F.(2005).Integrationofmultiplespeech1615segmentationcues:Ahierarchicalframework.JournalofExperimental1616Psychology:General,134(4),477-500.1617

Maye,J.,Aslin,R.N.,&Tanenhaus,M.K.(2008).Theweckudwetchofthewast:1618Lexicaladaptationtoanovelaccent.CognitiveScience,32(3),543-562.1619

McMurray,B.,Tanenhaus,M.K.,Aslin,R.N.,&Spivey,M.J.(2003).Probabilistic1620constraintsatisfactionatthelexical/phoneticinterface:Evidencefor1621


page42of44

gradienteffectsofwithin-categoryVOTonlexicalaccess.Journalof1622PsycholinguisticResearch,32(1),77-97.1623

Meng,M.,Remus,D.A.,&Tong,F.(2005).Filling-inofvisualphantomsinthe1624humanbrain.NatureNeuroscience,8(9),1248-1254.doi:10.1038/nn15181625

Morrill,T.H.,Heffner,C.C.,&Dilley,L.C.(2015).Interactionsbetweendistal1626speechrate,linguisticknowledge,andspeechenvironment.Psychonomic1627Bulletin&Review,22,1451–1457.doi:10.3758/s13423-015-0820-91628

Murray,S.O.,Kersten,D.,Olshausen,B.A.,Schrater,P.,&Woods,D.L.(2002).1629Shapeperceptionreducesactivityinhumanprimaryvisualcortex.1630ProceedingsoftheNationalAcademyofSciences,99(23),15164–15169.1631doi:10.1073/pnas.1925793991632

Nozaradan,S.,Peretz,I.,Missal,M.,&Mouraux,A.(2011).Taggingtheneuronal1633entrainmenttobeatandmeter.JournalofNeuroscience,31(28),10234-163410240.DOI:10.1523/JNEUROSCI.0411-11.20111635

Nozaradan,S.,Peretz,I.,&Mouraux,A.(2012).Selectiveneuronalentrainmentto1636thebeatandmeterembeddedinamusicalrhythm.Journalof1637Neuroscience,32(49),17572-17581.DOI:10.1523/JNEUROSCI.3203-163812.20121639

Oh,G.E.,&Redford,M.A.(2012).Theproductionandphoneticrepresentationof1640fakegeminatesinEnglish.JournalofPhonetics,40(1),82-91.1641doi:http://dx.doi.org/10.1016/j.wocn.2011.08.0031642

Ogden,R.(2012).Thephoneticsoftalkininteraction--IntroductiontotheSpecial1643Issue.LanguageandSpeech,55(1),3-11.1644

Ogden,R.,Hawkins,S.,House,J.,Huckvale,M.,Local,J.K.,Carter,P.,Dankovicová,1645J.,&Heid,S.(2000).ProSynth:Anintegratedprosodicapproachtodevice-1646independent,natural-soundingspeechsynthesis.ComputerSpeechand1647Language,14,177-210.1648

Piccolino-Boniforti,M.A.(2014).Fromsoundtogrammar:Theory,1649representationsandacomputationalmodel.Doctoralthesis,Universityof1650Cambridge.https://www.slideshare.net/MarcoPiccolinoBonifo/from-1651sound-to-grammar-theory-representations-and-a-computational-model1652

Pickett,J.M.,&Pollack,I.(1963).Intelligibilityofexcerptsfromfluentspeech:1653Effectsofrateofutteranceanddurationofexcerpt.LanguageandSpeech,16546,151-164.1655

Pierrehumbert,J.B.andBeckman,M.E.(1988),JapaneseToneStructure,MIT1656Press.1657

Philips-Silver,J.,&Trainor,L.(2005).Feelingthebeat:movementinfluences1658infantrhythmperception.Science,308(5727),1430.1659

Plag,I.(2014).Phonologicalandphoneticvariabilityincomplexwords:An1660unchartedterritory.ItalianJournalofLinguistics,26(2),209-228.1661

Plag,I.,&benHedia,S.(2018).Thephoneticsofnewlyderivedwords:Testing1662theeffectofmorphologicalsegmentabilityonaffixduration.InS.Arndt-1663Lappe,A.Braun,C.Moulin,&E.Winter-Froemel(Eds.),Expandingthe1664Lexicon:LinguisticInnovation,MorphologicalProductivity,andtheRoleof1665Discourse-relatedFactors.Berlin,NewYork:deGruyterMouton.1666

Plag,I.,Homann,J.,Kunter,G.(2015/2017).Homophonyandmorphology:The1667acousticsofword-finalSinEnglish.Journal of Linguistics, 53(1), 181–216. 1668doi:10.1017/S00222267150001831669


page43of44

Post,B.,D'Imperio,M.,&Gussenhoven,C.(2007).Finephoneticdetailand1670intonationalmeaning.InW.J.Barry&J.Trouvain(Eds.),16th1671InternationalCongressofPhoneticSciences(pp.191-196,PaperID1723).1672Saarbrücken:http://www.icphs2007.de/1673

Salverda,A.P.,Dahan,D.,&McQueen,J.M.(2003).Theroleofprosodicword1674boundariesintheresolutionoflexicalembeddinginspeech1675comprehension.Cognition,90,51-89.1676

Schroeder,C.E.,&Lakatos,P.(2009).Low-frequencyneuronaloscillationsas1677instrumentsofsensoryselection.TRENDSinNeurosciences,32(1),9-18.1678doi:10.1016/j.tins.2008.09.0121679

Schroeder,C.E.,Lakatos,P.,Kajikawa,Y.,Partan,S.,&Puce,A.(2008).Neuronal1680oscillationsandvisualamplificationofspeech.TRENDSinCognitive1681Sciences,12(3),106-113.doi:10.1016/j.tics.2008.01.0021682

Seyfarth,S.,Garellek,M.,Gillingham,G.,Ackerman,F.,&Malouf,R.(2018).1683Acousticdifferencesinmorphologically-distincthomophones.Language,1684CognitionandNeuroscience,33(1),32-49.1685

Shahin,A.J.,Bishop,C.W.,&Miller,L.M.(2009).Neuralmechanismsforillusory1686filling-inofdegradedspeech.NeuroImage,44,1133-1143.doi:168710.1016/j.neuroimage.2008.09.045.1688

Simpson,A.P.(2005).'Fromagrammaticalangle':CongruenceinEileen1689Whitley'sPhonologyofEnglish.YorkPapersinLinguistics,2(4),49-90.1690

Smith,R.,Baker,R.,&Hawkins,S.(2012).Phoneticdetailthatdistinguishes1691prefixedfrompseudo-prefixedwords.JournalofPhonetics,689-705.1692

Smith,R.,&Hawkins,S.(2012).Productionandperceptionofspeaker-specific1693phoneticdetailatwordboundaries.JournalofPhonetics,40,213-233.1694

Sproat,R.,&Fujimura,O.(1993).AllophonicvariationinEnglish/l/andits1695implicationsforphoneticimplementation.JournalofPhonetics,21,291-1696311.1697

Strycharczuk,P.(2019).PhoneticDetailandPhoneticGradiencein1698MorphologicalProcesses.InOxfordResearchEncyclopediaofLinguistics.1699

Strycharczuk,P.,&Scobbie,J.(2016).Gradualorabrupt?Thephoneticpathto1700morphologisation.JournalofPhonetics,59,76–91.1701

Strycharczuk,P.,&Scobbie,J.M.(2017).Whencethefuzziness?Morphological1702effectsininteractingsoundchangesinSouthernBritishEnglish.1703LaboratoryPhonology:JournaloftheAssociationforLaboratoryPhonology,17048,7.1705

Sugahara,M.,&Turk,A.(2009).Durationalcorrelatesofsublexicalconstituent1706structure.Phonology,26,477-524.1707

Turton,D.(2017).Categoricalorgradient?Anultrasoundinvestigationof/l/-1708darkeningandvocalizationinvarietiesofEnglish.LaboratoryPhonology:1709JournaloftheAssociationforLaboratoryPhonology,8.1710

Warren,P.,&Marslen-Wilson,W.(1987).Continuousuptakeofacousticcuesin1711spokenwordrecognition.Perception&Psychophysics,41,262-275.1712

Zuraw,K.&Peperkamp,S.(2015).Aspirationandthegradientstructureof1713Englishprefixedwords.InTheScottishConsortiumforICPhS20151714(Ed.),Proceedingsofthe18thInternationalCongressofPhonetic1715Sciences.Glasgow,UK:theUniversityofGlasgow.Papernumber0382,1-5.1716

1717


page44of44

1718

1719

Phonetic detail is used to predict a word’s morphological ...

Documents

Transcript of Phonetic detail is used to predict a word’s morphological ...