Stress and Accent

204 Pages·2012·1.8 MB·English

Stress and Accent
Prominence relations in
Southern Standard British English
PhD thesis
Submitted to the University of Copenhagen
Februar y2004
'Christian Jensen 2004
Typesetingrof f,12ptLegacy
Printed Sun May1610:41:14 CEST 2004
(only minor typographical corrections)
Awebpage accompanies this thesis.
It can be found at the following twolocations:
http://www.cphling.dk/pers/chrjen/thesis/
http://www.fonetik .dk/thesis/
Abstract
This thesis examines the (perceived) prominence relations (stress and accent levels)
in Southern StandardBritish English. In line with previous descriptions it was found
that stressed words adjacent to utterance boundaries, or major phrase boundaries,
are perceived as more prominent than stressed words in intermediate positions, but
contrar yto manydescriptions in the traditional British school of intonation analysis
it was notfound that the (cid:140)nal stressed word(the so-called nucleus) is generally more
prominent than all other words in the phrase. Only 15-20% of the neutral, context-
free utterances had a clearly more prominent (cid:140)nal item.
In utterances where one word is emphasised due to some type of narrow focus
there is both a local effect on the emphasised item, resulting in signi(cid:140)cantly higher
perceived prominence on this word, and a global effect on surrounding words, which
are perceived as less prominent;inother words a combined foregrounding and back-
grounding of items inside and outside of the focus domain, respectively.
The backgrounding effect is largest in post-focal position, where the promi-
nence level of all stressed words is reduced, regardless of their distance from the focal
accent. In pre-focal position the reduction in prominence level is inversely propor-
tional to the distance from the focal accent: immediately adjacent items are reduced
the most.
The relevance of these observations was demonstrated in an experiment which
ex amined the relation between perceived prominence and perceived information
structure. As expected, listeners perceived utterances in which one item was particu-
larly prominent as responses to questions about this single constituent › they heard
the item as being in (narrow) focus. This was true of intended neutral and intended
focused utterances alike and in all positions in the utterance, even when the most
prominent item was in utterance (cid:140)nal position, that is, the default location of the
nucleus.
The relative reduction of non-focal items contributed to the perception of
focus, and the results suggested that post-focal reduction is more impor t ant than
pre-focal reduction, in accordance with the results of the prominence perception
experiments.
A brief account of the acoustic parameters F and duration is presented. The
0
variation in F mirrored the perceived prominence in a fairly direct way in both neu-
0
tral utterances and in pre-focal, focal and post-focal position: F movements were
0
almost absent in post-focal position but were reduced in inverse proportion to the
dist ance from the focal accent in pre-focal position. The duration data indicated a
larger pre-focal shortening effect and only a ver y modest post-focal effect. The two
acoustic parameters thus seem to operate differently depending on their position rel-
ativetothe focal accent.
iii
Dansk resumØ (Danish summary)
Denne afhandling undersłger de opfattede prominensforhold (eller opfattet grad af
tr yk) i Southern Standard British English. Result aterne af en r(cid:230)kke lytteforsłg viste,
i lighed med tidligere beskrivelser, at trykst(cid:230)rke ord i umiddelbar n(cid:230)rhed af
ytringsgr(cid:230)nser, eller stłrre frasegr(cid:230)nser, opfattes som mere prominente end
tr ykst(cid:230)rke ord i mellemliggende positioner, men i mods(cid:230)tning til mange beskrivel-
ser i den traditionelle britiske intonationsskole kunne det ikke demonstreres at et
frase(cid:140)nalt trykst(cid:230)rkt ord(den s(cid:229)kaldte ‘nucleus’) generelt setermere prominent end
alle andre ordifrasen. Kun 15›20% af de neutrale, kontekstfri ytringer havde etklar t
mere prominent (cid:140)nalt ord.
I ytringer hvor et ord er fremh(cid:230)vet for at signalere sn(cid:230)ver fokus (semantisk
fokus eller kontrastfokus), er der b(cid:229)de en lokal effekt p(cid:229) detfremh(cid:230)vede ord, hvilket
ses ved at dette ord opfattes som betydeligt mere prominent, og en global effekt p(cid:229)
omkringliggende ord, som opfattes som mindre prominente. Der er med andre ord
tale om en kombineret effekt af at fremh(cid:230)ve ord indenfor fokusdom(cid:230)net og nedt-
one ordudenfor fokusdom(cid:230)net.
Ef fekten af at nedtone ord udenfor fokus er stłrst postfok alt, hvor prominens-
niveauet p(cid:229) alle trykst(cid:230)rke ord reduceres, uafh(cid:230)ngigt af deres afstand til den fok ale
accent (det fokuserede eller kontrasterede ord). Pr(cid:230)fok alt er reduktionen i promi-
nens omvendt proportional med afstanden til den fok ale accent : umiddelbar t
tilstłdende ord reduceres mest. Relevansen af disse observationer blev demonstreret
i et eksperiment som undersłgte forholdet mellem opfattet prominens og opfattel-
sen af informationsstruktur. Som ventet opfattede lytterne ytringer hvor et ord var
s(cid:230)rligt prominent som svar p(cid:229) etspłrgsm(cid:229)l om netop dette element › de hłrte dette
ord som fokuseret. Dette var tilf(cid:230)ldet b(cid:229)de med ytringer som var ment som neutrale
fra talerens side, og med ytringer som var ment som fokuserede, og det gjaldt alle
positioner i ytringen › selv n(cid:229)r det mest prominente ord var i ytrings(cid:140)nal position,
dvs. den for ventede placering af ‘nucleus’.
Reduktion af ikke-fok ale ord medvirkede til opfattelsen af fokus, og resulta-
terne antydede at postfok al reduktion er mere v(cid:230)sentlig end pr(cid:230)fok al reduktion,
hvilket erioverensstemmelse med resultaterne af undersłgelserne af opfattetpromi-
nens.
De akustiskeparametre F og varighed pr(cid:230)senteres kor t.Variationen i F afspej-
0 0
ler den opfattede prominens p(cid:229) en forholdsvis direkte m(cid:229)de i b(cid:229)de neutrale ytringer
og i pr(cid:230)fok al, fokal og postfok al position: F -bev(cid:230)gelser var stort set frav(cid:230)rende
0
postfok alt, men i pr(cid:230)fok al position var de reduceret omv endt proportionalt med
afst anden til det fremh(cid:230)vede ord. Varighedsdat aene pegede p(cid:229) en stłrre pr(cid:230)fok al
forkor telse og kun en beskeden postfok al ef fekt. Dermed synes de to akustiske para-
metre at virke forskelligt afh(cid:230)ngigt af deres position i forhold til den fok aleaccent.
iv
Ac knowledgements
This project was supported by a grant from the Danish
Research Council for the Humanities and was carried out at
the Department of General and Applied Linguistics, University
of Copenhagen; I greatly appreciate both the (cid:140)nancial support
of the Research Council and the hospitality of the department.
I would like to thank all who have helped me in one way or
another overthe years, and particularly ever ybodywho acted as
aspeaker or listener in myexperiments.
v
Cont ents
Abstract pageiii
Dansk resumØ (Danish summary) iv
Acknowledgements v
List of (cid:140)gures ix
List of tables x
Intr oduction 1
1 Str ess,accent and prominence 3
1.1 Introduction 3
1.2 Earlier accounts of stress 4
1.3 Early experimental workonstress 6
1.4 More recent investigations 11
1.4.1 Pitch and duration 12
1.4.2 Spectraltilt/balance/emphasis 15
1.4.3 Prominenceand wordclass 17
1.4.4 Prominencescales 18
1.5 Stress and the British school of intonation 19
1.6 Stress and the autosegmental-metrical approach 24
1.7 Stress in Danish 25
1.8 Comments on terminology and de(cid:140)nitions 26
2 Aninves tigationof prominence › collecting data 30
2.1 Introduction 30
2.2 Datacollection 31
2.2.1 Text material 31
2.2.2 Recordings 37
2.3 Selectionof utterances 37
2.4 Segmentation 40
2.5 Acousticcharacterisation of utterances 42
2.5.1 F range variability among speakers 43
0
2.5.2 Generalobser vationsabout F 44
0
2.5.2.1 Neutral utterances with stress clash 45
2.5.2.2 Neutral utterances without stress clash 47
2.5.2.3 Markedinformation structure 49
2.5.2.4 Interrogativesentences 52
2.5.3 Durationand stress 54
3 Validating prominence ratings 58
3.1 Introduction 58
3.2 Material 59
3.3 Tests 1›3 › perception of prominence 60
3.4 Test 1 › Danish listeners 60
3.4.1 Subjects 60
3.4.2 Purpose of the listening test 61
3.4.3 Instructionsto the raters 62
3.4.4 Listenerfeedback on the test 63
3.4.5 Data 64
3.4.6 Testing reliability and agreement 65
3.4.6.1 Reliability 66
3.4.6.2 Agreement 67
3.4.7 Theobser ved reliability 69
3.4.8 Theobser ved agreement 70
3.4.8.1 Where do the raters disagree? 73
3.4.8.2 Lexicalversus grammatical words 75
3.4.8.3 Effect of experience and background 77
3.4.9 Conclusion 79
4 Perceived prominence levels in utter ances›Danish listener s 81
4.1 Selecting and grouping utterances 81
4.2 Context-freeutterances 83
4.2.1 First and last lexical item › or onsetand nucleus 84
4.2.2 Intervening lexical items: strong › weak alternation 90
4.3 Utteranceswith markedinformation structure 91
4.3.1 Focal stress/accent 94
4.3.2 Non-focal stress 94
4.4 Effect of experience and background on perceived prominence 97
4.5 Preliminaryconclusions 98
5 Tes t2›English listener s 99
5.1 Introduction 99
5.2 Subjects 99
5.3 Instructionsto the raters 99
5.4 Feedback from the raters 100
5.5 Reliability 101
5.6 Agreement 101
5.7 Prominencelevels › English listeners 104
5.7.1 Context-free utterances 104
5.7.1 .1 First and last lexical item 106
5.7.1 .2 Inter vening lexical items 107
5.7.2 Utteranceswith markedinformation structure 107
5.8 SummaryofTests 1 and 2 110
6 Tes t3›British school of intonation analysis 112
6.1 Introduction 112
6.2 Stress/accent levels 112
6.2.1 Aims of Test 3 114
6.3 Subjectsand instructions 115
6.4 Feedback from the raters 115
6.5 Data 116
vii
6.6 Reliability 116
6.7 Agreement 117
6.8 Prominencelevels › British school of intonation 119
6.8.1 Context-free utterances 119
6.8.1.1 First and last lexical item versus onsetand nucleus 119
6.8.1.2 Inter vening lexical items 125
6.8.2 Markedinformation structure › British tradition 125
6.8.2.1 Problems concerning individual words 126
6.8.2.2 Comparison with Tests 1 and 2 128
6.8.2.3 Defaultnucleus versus (cid:140)nal focus 129
6.8.3 TheBritish system and prominence ratings 130
6.9 Highpreheads › accented or not? 132
7 Tes t4›Perceived information structur e 137
7. 1 Introduction 137
7.2 Method 137
7.3 Test setup 138
7.3.1 Internettest 138
7.3.2 Recorded test 139
7.4 Subjects 139
7.5 Results 139
7.5.1 Listener reliability and agreement 139
7.5.2 Overall identi(cid:140)cation of contexts 141
7.5.3 Differences between listener groups 141
7.5.3.1 Average number of errors 142
7.5.3.2 Distributionof errors › response patterns 143
7.5.4 Identi(cid:140)edand misidenti(cid:140)ed contexts 144
7.5.4.1 Utterances with a speci(cid:140)c focus 147
7.5.4.2 Neutral, context-free utterances 149
7.6 Inferring information structure from prominence relations 152
7.6.1 Hypothesis 1 153
7.6.2 Hypothesis 2 154
7.6.3 Hypothesis 3 156
7.7 Conclusion 162
Futur eresearch 164
Bibliogr aphy 167
APPENDICES
A Information about experiment alpr ocedur es 175
B Dat alis tings 182
Bibliogr aphy(Appendix) 192
viii
Figur es
Fig. 1.1 Armstrong/WardTune I 20
Fig. 1.2 Stress/accent hierarchyinCruttenden’s analysis 23
Fig. 2.1 F turning points and simpli(cid:140)ed traces 39
0
Fig. 2.2 Minor deviations in F traces 40
0
Fig. 2.3 Segment ationand annotation view 41
Fig. 2.4 Av erage trace ofbsa n2F 43
Fig. 2.5 F range variation 44
0
Fig. 2.6 Av eraged traces › sentenceps n,speakers 5M and 2F 45
Fig. 2.7 Av eraged traces ›bsa n,speakers 2F and 4M 46
Fig. 2.8 Av eraged traces ›jkf tn,speakers 5M and 6M 46
Fig. 2.9 Av eraged traces › sentencepc n,speakers 5M and 2F 47
Fig. 2.10 Av eraged traces ›css n,speakers 5M and 6M 48
Fig. 2.11 Av eraged traces ›sepc n,speakers 5M and 2F 49
Fig. 2.12 Sentencebsa,three focus conditions, speaker 4M 50
Fig. 2.13 Sentencesepc,four focus conditions, speaker 1F 51
Fig. 2.14 Neutral versions of the interrogativesentencepdp 53
Fig. 2.15 Sentencepdpwith focus on the (cid:140)rst word 54
Fig. 4.1 Prominence ratings, neutral sentences, Danish raters 84
Fig. 4.2 Prominence ratings, (cid:140)rst and last lexical item 86
Fig. 4.3 Prominence ratings, markedinformation structure 93
Fig. 4.4 Prominence ratings, sentencejkf t 96
Fig. 5.1 Prominence ratings, neutral utterances, English raters 105
Fig. 5.2 Prominence ratings, marked information structure, English 109
raters
Fig. 6.1 Stress/accent possibilities in the British system 113
Fig. 6.2 Prominence ratings, neutral sentences, British school 120
Fig. 6.3 Pitch contour,utteranceps n4M 121
Fig. 6.4 Distribution of ratings for onsets and nuclei 124
Fig. 6.5 Ratings, markedinformation structure, British school 127
Fig. 6.6 F and duration › two utterances with prominent grammati- 135
0
cal words
Fig. 7.1 Distribution of correctly identi(cid:140)ed contexts 141
Fig. 7.2 Distribution of correct contexts, three test setups 142
Fig. 7.3 Graph of expected and observedresponses for one listener 143
Fig. 7.4 Utterances with good listener recognition of intended context 147
Fig. 7.5 Tw outterances with poor identi(cid:140)cation of intended context 148
ix
Fig. 7.6 Utterance with higher proportion of ‘neutral’-responses than 148
expected
Fig. 7.7 Focus and neutral version of jkf t › similar prominence, differ- 149
ent information structure
Fig. 7.8 Sentencespcandbsa,neutral andf1versions 150
Fig. 7.9 Illustration of the connection between perceived prominence 151
and the perception of information structure
Fig. 7.10 Hypothesis 1 › Most prominent item (MaxProm) 153
Fig. 7.11 Hypothesis 2 › MaxProm minus mean of rest 155
Fig. 7.12 Contribution of pre-focal maximum prominence (PrefocMax) 158
Fig. 7.13 Contribution of post-focal maximum prominence (Postfoc- 158
Max)
Fig. 7.14 Pre- and post-focal reduction incl. interaction with MaxProm 161
Fig. B.1 Prominence ratings, neutral utterances, three groups 187
Fig. B.2 Prominence ratings for Group 1, marked information struc- 188
ture
Fig. B.3 Prominence ratings for Group 2, marked information struc- 189
ture
Fig. B.4 Prominence ratings for Group 3, marked information struc- 190
ture
Tables
Table 2.1 Segment duration ratios 56
Table 3.1 Excer ptfrom rawdat a(cid:140)le, perception experiment 64
Table 3.2 Reliability coef(cid:140)cient (Cronbach’s alpha), Danish raters 70
Table 3.3 Distribution of ratings, Danish raters 70
Table 3.4 Distribution matrix for ratersr1andr2 71
Table 3.5 Distribution matrix, all Danish raters 72
Table 3.6 Agreement measurements, Danish raters 72
Table 3.7 Agreements and disagreements between one rater and nine 74
other Danish raters
Table 3.8 Number of occurrences of all possible response pairs, Danish 75
raters
Table 3.9 Reliability coef(cid:140)cients for lexical and grammatical words sepa- 76
rately,Danish raters
x