danpass.hum.ku.dk › ng › papers › aripuc18_1984_185-242.pdf · . aripuc 18, 1984, p. 185-242...

.

ARIPUC 18, 1984, p. 185-242

INTONATION AND TEXT IN STANDARD DANISH

NINA GRØNNUM THORSEN

Acoustic analgsis of recordings by four StandardDanish speakers shows that each declarative sen-tence in a text is associated with its own decliningintonation contour, but together two or three suchcontours describe an overall falling slope. Indi-vidual sentence intonation contours are steeper anddemonstrate greater amounts of resetting betweenthem in a succession of declarative terminal sen-tences than in a corresponding string of coordinatemain clauses. In other words, the cl oser relationbetween coordinate structures is reflected in a morecoherent or less segregated intonational structure.The results are compared with other languages, andthe implications for the abstract representation ofDanish intonation are discussed.

'\

'.

I. INTRODUCTION

Lehiste (1975, p. 195) hypothesizes

II... that pamt)r-aphs possess a supr-asegmental str-uature thatindiaates the beginning and end of par-agr-aphs and ahar-aater-izethe body of the par-agr-aph. For- exampteJ it may be that the in-tonation aontoU!" applied to a sentenae pr-oduaed in isolation(aonstituting a one-sentenae par-agr-aphJ tWill differ- from theintonation aontoU.?' apptied to the same sentenae in the begin-ningJ middZe and end of a par-agr-aph. In other- wo~dsJ it ispossibLe that a par-agr-aph is c:!ha~aater-iz~d by an overall in-tonation str-uatu~e to whiah the intonation aontOU!"B of itsaonstitu~nt sentenaes are B2Øordinated. ... A ao1'ollary hypoth-esis is that sinae speakers and Zistene~s share the same aodeJlisteners a~e aapable of deaiding ~hether a sentenae has beenproduaed in isolation o~ as part of a tar-ge~ st2"Uatu~e.'1

r . -

,186 THORSEN

Both hypotheses are verified in Lehiste's (1975) experiments.Acoustic analysis of ane recording of paragraphs containingane, two, and three sentences (of greatly varying length) byane American English speaker, shows that an isolated sentenceis longer than the same sentence in a paragraph context. Fun-damental frequency (Fø) in the beginning of a sentence is lowerin an isolated sentence than in a paragraph initial ane, andhigher in paragraph initial position than in media1 and finalpositions. - In alistening experiment subjects were tojudge what position asentence - cut out from its paragraphcontext - had occupied in the paragraph. Their judgment wassignificantly better than chance, though not a1ways in accord-ance with the speakeris intention.

Bruce's (1982) point of departure is an informal listening ex-periment where two sentences, produced in isolation and intwo-sentence texts, are spliced together in various ways.Isolated sentences spliced together, and sentences from textsspliced together in reverse arder sound odd, and neither com-bination constitute a perceptually coherent text. Acoustic(Fø) analysis of the original recordings shows that two sen-tences in a text are characterized by a continuous overalldowndrift, though the sentence boundary is marked by a low,sentence final Fø minimum succeeded by a resetting to thefirst high Fø peak in the second sentence. This peak is lowerthan the corresponding Fø peak in the first sentence. Like-wise, the first Fø peak is lower in asentence produced inisolation than in the first sentence of two in a paragraph,whereas its last Fø minimum value is higher in isolation thanin the second of two sentences in a paragraph. The differencebetween a single sentence and a combination of two seems to besignalled globally [i.e. over a temporal scope correspondingto the whole text, NT], except that each sentence in a tex talso has a final, very low Fø value. - These results areconfirmed in acoustic analyses of a 1arger materia1 with aSouthern Swedish speaker. Three sentences (each containingtwo stress groups) are produced in isolation and in the ninepossib1e two- and three-sentence combinations. They are sepa-rated by periodst but each begins with "A" (lAndi). The veryfirst and very last Fø va1ues (which occur in unstressed sy1-lables) are 10w and constant across texts of different dura-tions. Apart from that, 10cal Fø maxima and minima are higherin the beginning of the text than at its end, and they arehigher in the beginning of a longer text than in the beginningof a shorter text, whereas the final va1ues are constant acrosstexts of different 1ength. Mu1ti-sentence texts were producedwith internal pause(s). Before each such pause Fø drops tothe same low Fø value as found in abso1ute initial and finalposition (the "floor") and resets at the anset of the succeed-ing sentence to smoothly continue the overall downdrift of10Col maxima and minima. When sentences in different positionsare compared, isolated sentences turn out to resemb1e textinitial sentences more than text final anes. As Bruce notes(1982t p. 284) this corresponds to Lehiste's (1975) observa-tion that listeners confuse iso1ated sentences with initialsentences more aften than with final anes.

.

.

.INTONATION AND TEXT 187

Cooper and Sorensen (1981, p. 85ff) analysed four differenttwo-clause sentences (each clause containing four stressedwords), among them a coordinate main clause construction.They found that individual clauses had separate declinationsassociated with them, but together these individual declina-tions describe an overall downdrift. They note specificallythat the resetting from the lower value at the end of thefirst clause to the higher value of the beginning of the sec-ond clause is independent of any accompanying breathing pause.They hypothesize that resetting between clauses in a complexsentence is more likely to occur in the environment of longerclauses, slower rates of speech, and lower semantic related-ness between the two clauses (p. 97).

Uyeno et al. (1979) investigated Fø in declarative complexsentences in Japanese: a construction with an embedded rela-tive clause, a coordinate main clause construction, as wellas a sequence of two simple sentences in succession. Theirdata show that all three types are characterized by an over-all declination with a local resetting at the anset of thesecond clause or sentence, except when the relative clauseis utterance initial, in which case the utterance declinessmoothly from the initial high rise. The resetting is greatestbefare a center embedded relative clause. The point relevantto the present investigation is that coordinate main clauseconstructions and successions of simple terminal sentencesdid not differ among themselves.

These results (especially those of Lehiste, 1975, and Bruce,1982) are all in sharp contradistinction to Nakatani (1975)who reports that listeners, presented with stories read nor-mally and stories spliced together from isolated sentences,had difficulty in distinguishing the stories as normal orspliced. He concludes that speech features observed in iso-lated sentences can be reasonably generalized to sentencesin a coherent context.

The present study is to a large extent similar to those ofLehiste (1975) and Bruce (1982) to whom I owe the idea andgeneraloutline of my own procedures. Its aim was to es tab-lish how and to what ex ten t sentences in a short tex t arecoupled intonationally, and furthermore to see whether thispresumed coupling is dependent upon the (syntactic) relationbetween the sentences in the text. I shall mainly be con-cerned here with fundamental frequency, though there is evi-dence from American English and Japanese that other factorssuch as length, laryngealization and pause duration are alsoinvolved in the production and perception of sentence andclause boundaries, cf. Lehiste (1975, 1979, 1980), Lehisteand Wang (1977), Cooper and Sorensen (1981), and Uyeno et al.( 1981 ) .

r . -

.188 THORSEN

II. MATERIAL~ SUBJECTS~ AND PROCEDURES

A. MATERIAL

The base is the following three declarative sentences (stres-sed vowels are indicated orthographically with acute accentshere):

(1) Amanda skal afsted på camping.'Amanda is going away camping. I

(2) Hendes mor skal på kursus i '1:Ysktand.

'Her mother is taking a course in Germany.1

(3) Hendes far skal vandroe i Lapland.

tHer father is goirig to hike in Lapland.'

In isolation as well as combined in texts of two and threesthese sentences could all be uttered in answer to a questionabout what Amanda and her family are going to do during thesummer holidays.

The total number of possible texts that can be constructedfrom these three sentences amounts to fifteens which doublesto thirty when two different boundary conditions are intro-duced. This was an unmanageably large number for subjectsto records if the material was to be mixed with filler utter-ances of a different syntactic and semantic make-ups whichwas highly desirable. I decideds therefores to concentrateon (1) aboves i.e. the Amanda-sentences and be sure that aneoccurred in all the relevant positions.

The following combinations of sentences constitute the mate-rial:

l

l 3

2 l

l 3 2

2 l 3

3 2 l

The Amanda-sentence is the only ane to occur in isolation.Note, however, that both sentence (2} (referred to as the mor-sentence in the following) and sentence (3) (referred to asthe far-sentence in the following) occur in all three positionsin three-sentence texts.

Two different types of text result when the sentences in multi-sentence texts occur in sequences of declarative terminal mainclauses versus coordinate main clause constructionss joined

-

-.


by og ('and'). Examples are: Amanda $kal af$ted på camping.Hendes far skal vandre i Lapland. Hendes mor skal på kursu$i TYskland. versus Hendes far skal vandre i Lapland, og hendesmor skal på kursus i TYskland, og Amanda skal afsted på ccimp-ing.

The five terminal sentence texts and the isolated sentencewere randomized three times, and the total of 18 sentences/texts were mixed with two materials recorded for different ana-lysis purposes, so they were evenly spread over six pages ofreading material. The same procedure was applied to the fivecoordinate clause texts and the isolated sentence. To avoidany direct comparison of the two types of text they were re-corded in separate sessions at least ane day apart. In thisway the isolated sentence was recorded six times by eachspeaker, but the texts only three times each. . This is a com-promise between the demand for as comprehensive a material aspossible, a sufficiently large number of recordings of eachitem by each speaker, and the wish to avoid fatigue effectson the part of the subjects. On a fair number of previousoccasions (among them recordings of rather lang simple sen-tences, cf. Thorsen 1983a) subjects' repetitions of the sameitem have always been rather remarkably constant, with standarddeviations on calculated Fø mean values generally below 5% ofthe mean. I therefore decided that the demand for (the stand-ard, minimum) six recordings of each item would have to yieldto the possibility of presenting the texts under two differentboundary conditions.

My material differs from Lehiste's (1975) texts ("paragraphs"in her terminology) in that the sentences are all equallylang (in terms of the number of prosodic stress groups ineach) and in the addition of coordinate main clause construc-tions. It differs from Bruce's (1982) material in the greaterlength of each sentence (three versus two prosodi c stressgroups). And again, where Bruce (1982) so to speak combined aterminal with a coordinate clause construction (by separationof the sentences with periods but commencing each sentencewith 'and'), these two conditions are differentiated in thepresent material.

There is no conflict in Lehiste's (1975) choice of term ("para-graph") and mine (and Bruce'st 1982) ("text"). The presentsentence combinations are both paragraphs and texts; at leastif ane claes not claim that the total six pages of readingmaterial constitutes a text. In that case my Iltexts" wouldonly have the status of paragraphs. However, the six pagescould lay no claim on any semantic or pragmatic coherencetand hardly deserve the designation as a tex t properly speaking.

B. SUBJECTS

Four phoneticians, two males (NRP and JBC) and two females(GB and NT (the author)) read the material, in two sessionson separate days, as mentioned above. They all speak a form

r ' -

,190 THORSEN

of Standard Copenhagen Danish. Their style of speech duringthe recordings can be characterized as fluent and conversa-tional but distinct.

C. TECHNICAL PROCEDURES

The recordings were made with semi-professional equipment(Revox A-77 tape recorder, Sennheiser MD21 microphone) in aquasi-damped room at the Institute of Phonetics on Agfa PE39tape, at 7~ i.p.s.

The tapes were processed by hardware intensity and fundamentalfrequency meters (F-J Electronics) and registered on a mingo-graph (Elema 800) at a paper speed of 100 mm/s. Byadjustmentof the Fø meter zero-line to the lower limit of the subject'svoice range and full exploitation of the record space (about80 mm) of the mingograph galvanometer, a measuring accuracyof 1 Hz for the males and 2 Hz for the female speakers isattained.

In unidirectional Fø movements in the stressed vowels, the be-ginning and end point were measured, according to a procedureoutlined in Thorsen (1979, p. 63-66). In bidirectional move-ments, the turning point was measured as well. The unstressedvowels in the sentences were represented by a single Fø point,the midpoint (in time as well as frequency), which was an un-controversial procedure since most of these vowels have mono-tonically falling Fø movements. (The first post-tonic syl-lable may be rising-falling, in which case Fø is measured atthe peak.) The Fø movements are so short and slight in theunstressed syllables that it does not much matter which pointyou choose as the ane measuring point: the initial, medial, orfinal (or any other) value. Changing the location of themeasuring point will transpose all the unstressed vowels byvery nearly the same (negligible) amount upwards or downwards,relative to the stressed vowels in the figures presented below.There is ane instance where two unstressed vowels cannot besegmented. skal af(sted) is pronounced without any inter-vocalic consonant [sga a], and the measuring points are as-signed time coordinates ane quarter of the distance from theanset and offset of voicing in the lang vowel sound. Thedistance in time of each measuring point from the first Føvalue in the sentence or text was likewise measured. In theterminal sentence texts, time was set to zero at the firstmeasuring point in each sentence, and the pause between sen-tences (defined as the time interval where no acoustic energyis registered) measured separately. In the coordinate mainclause texts, time was measured cumulatively from the ansetof the first sentence. Fø and time measurements were averagedover the three recordings by each subject (six recordings inthe case of the isolated sentence). Average Fø values wereconverted to semitones (re 100 Hz) and average tracings drawn.No correction was attempted for intrinsic Fø level differencesbetween the high stressed vowels of the mO1'-sentences and thelow stressed vowels of the fa1'- and Amanda-sentences.

.

..

INT

ON

AT

ION

AND

TE

XT

191

a..~~

.

, .

I I

/g'

t)(lJ(lJ(lJ00.

o ~

U

) ~

E

~

'.-1 ~

~

. ..

o (lJ

.j.J ~

(lJ

U)

u g'

u)~'""i~

(lJ...

0- ~

~

t)

'""i~

a.

. ~

~

~

(lJ~-O

..

'""i '""i

(lJ ~

E

~/

u t)t).j.J~

." ...

(lJ ~

O

' a.

q)~~

U):i

. -O

t)..-I

'""i .j.J

-; .

~

~1I-I~

(lJ.

~

.. q)O

~~

"..

.j.J oou)

I~

. ~

U)o~

~c

-; (lJ(lJt)

(lJ.~

~

U

) t)

~~

U)

. <

/2"

~~

(lJ~.c:

'" ~

~

q)(lJ~U

)(lJQ,.

u c

E~

~U

)~~

~

... ~

o. (lJU

)~.

~

U)

(lJ--~

bitl)U

).j.J~g

(lJ (lJ

U)

~

11-1.

. .c:

t)~O

O-O

..;

.j.J .~

~~

c c

-- <

11 .j.J

~

~

(lJ (lJ.j.J..

(lJJO

~

.,.'"

~

~~

~U

).U)

~

~

(lJ '.-1 q)

.c: N

f- f-

~U

)U)Q

,.:l":.cc:.

~ 0-

. .-

~

~

t)..

.. 'tJ~

~O

~.

e". ,,"

~ (lJ biO

(lJ..

.. U

) ~

(lJ ...,

>5

-- 5

.j.J.c::i(lJ 11-1

~

~

><

.j.J.c:oo

... ...

(lJ .j.J

(lJ .j.J

.j.Ja.

a. .j.J.c:.Q

.j.J

- -

bi ~

U).cc:

. ..'

.. o

'.-1 (lJ 0.-1 ~

bi

. ~

.

~:i

~

U)

~

'.-1Lo

.j.J--~U

)O~

e ---

~

II-I~~

~~

~..

-.. ..

o o

-.-I (lJ

oq)

q)'.-I (lJ

'""i ~

.j.J

'" -g

-g --

.j.J .c: ~

~

. q)

~

. q)...,

U)'.-I.j.J'""iO

(lJ~

z

~

z blU

) '""it).c:

~

~

(lJ ~

Oll-l~

.j.J

~I

I "

I I

II ,

I I

I I

~I

. .

, ,

I . . . . I

, . ~

'.-I Q

,. o 11-1 (lJ

, ,

" I

' .

, .

, .

. .

~

~

'""i.j.JIn

o '?

In o

'? bi

~'""i~

.c:~~

'.-1 o

~

o .j.J

t)~

t)

~

'.-1 '.-1

U)

~-O

(lJ

'.-1 .j.J

~

(lJ.

c ~

11-1 ~

::J)'.-I

~

~~

t)1I-I.-/

~

(lJ.j.J ~

(lJ

~'.-I

-J (lJ>

<~

.j.J(lJ.j.J.

~

(lJ ~

~

~

. 0-

. -o

.c: .j.J (lJ

t) o.

(lJ.

q) :;

.j.J .c:'.-I(lJ~

-'ti/

- ~

.j.J

~

~

'.-1c

~

~~

~

II-I..

-J (lJ

~

11-1 '.-1 I

~

>

.. ,-

~

o t)

-; o

-- (lJ

o .-I --

. ~

~

(lJ~

~E

E"i

.. ~

"O

(lJ~

O~

.c:~Lo

0"':; 0\ . .-I . .-I

.j.J?

~

>

~

U).j.J

U)

'.-1 ~

. -

~

~'""i~

~

~

~

~.j.Jt)~

~~

-o ..

... 11-1 '.-1

~

V1

. ~

Lo

~(lJ~

OO

~Z

"""..

--'""i~~

'""i~...

--'.-1--

~

.. U

) ~

(lJ

~

q) bI~

.c:(lJ.c:U)

. -g

~O

.j.JU).j.J(lJ

q) '.-1

'.-1".-1 U

) '""i

. .c

t).j.J~(lJ~

~.

. g

~

'.-I -- ~

o

E~

U

) .j.J

(lJo

. .j.JO

(lJtI)OIl-l

g'/

g' Q

,.~

~/

0- 0-

:J\ o

(lJ o~

~

t)'""i.Q

.N:i

.. ..

~

~

~

~

.j.J.

U'

u (lJ

'.-1 o

.", ...

~.j.J

(lJ '""i .

~~

a.

..- a.

0.'.-1 t)

(lJ. U

) ~

r -o

~

-o (lJ~

~.Q

.j.J~~

~

~

'.-1

(lJ ~

. ..

. ..

1I-I.j.JU)'.-I--

. ~

. .j.J~

~O

CJ

'""i>«lJO

Q,.~

- -

~

(lJ U

) 0.-1

~

.. ..

. ~

.

~

.j.J.j.J.j.J:J\..

.. ~

'""i

t).Q~

.. ..

(lJ ~

~

~

~

"" -g

"-1 -g

E

'.-1 ~

~

~

~

'" 'V

..

~".-I

.j.J (lJ

~

. ~

i.

~

~U

)EU

).j.Jo..

E

~~

~~

~~

~

~

~

(lJ O

(lJ

~I

I I

II I

, II

I j

I j

I. I

I I

I I

" I

I I

I I

~

t) .j.J

t) U

) --

In o

In In

o In

I I

r .

-,192

o T

HO

RS

EN

~

o ./

/ g'

o

a 'i

,/ ,

U

0\c:

/ o~

aO

'~

"tO

mQ

) u

o ..

a 0/1

.m"-

a Q

.m

"tO-

Q)

m

..a

~

i.::m

I ~

o

-c:

mm

o

~a

~

0/1

mII)

"tOU

c

PI

m...

E.o:0\O.

a."tO

"tO

c: c:

... m

m

r-'" -

~

r ~

0/1 0/1

>-

>-

a I-

I-a

.- .-

a 0/1

a' 0/1

" "

.>

~)

~~

~

om

""a

Q.

Q.

a-

-a

m

a m

~

~0/1

..

~

~

'7 S

0/1 ..

Q)

Q)

a "tO

o

"tOc:

c:

~

a ~

~

a

~

i, i

I I

I II

I I

I I

i, I

.I I

I I

I I

I I

I'"

O

'" O

a

a "tO

...;::. i

a >

o .

a.- "tO

a Q

) c:

L. m

>-g

.>

jm

-'

>

a

a ...~

Q

)..

L.

'>L.

"-] ~

m

>"-0/1

a ...

a ~

~

c: a

0/1

aO

Q)

L.

:r a

m

"-II)"

..ID

Q

)"tOc:Q

).I:

o0\

a O

0\ /.;.

I.: .:

o.S

o. E

a

E

'""'

m

m

r.-o

U

U

vQ)

omI

om

;3I

o. o.

~"tO

"tO

'ooi

Q)

Q)

.. o

..-I.Ja

.:: a

.:: ~

a m

m

O

- -

tJm

a

ma

~

~

0/1 ..

'llO

>lO

Q

)

"tO

"tO

t.

c: c:...

io

~

~

o ~

.~

~

11..1I

I. I

II III

II ~

I I

I II

I I

I I

I I

'-'"

O

'" O

. ,

-.

.IN

TO

NA

TIO

N AN

D T

EX

T

193

CX

:I~

.

0\co.-

..--

o. .

E

O-

~

.~."'"

Eo.

..I

"U

. U

.. "'"

:;: /

o..'"

"U."

....-

. In

.. .

....O

L ..

. In".

.."U

.O

Lco

In

. ~

~

..~

"UC

O..

\I) .

EU

.

~O

)ro

O-

O

"U

"U.

CO

.

CO

.. ..

~

~.-/

~

~

~.-

.-.

.In

In"

".

In. '"

.. ..

/" 1.

~

1."'"

"'" "'"

o. o.

. -

-..'

...O

L .O

L.

In .

In..

..2"~

2

In In

.. ..

"U

"U~

. ~

~

.

~il

I I

I I

I I

I I

I I

I I

I I

il I

I I

I I

I I

I I

I I

I I

II!J

R

Ln ~

R

Ln

"UCO

...

o... .

-' "UC

O'-

...'!O

l o.

." ..

. -g

~

..J

.. .-'"

.->

. ..

ro .

-'o.

.OL

CO

In ..

--;;00

:;; /.:

... .

...OL

In In

.."U

'--;:o...

CO

..

. ~

...'"..

'" ."U

U

CO

o'" ."

.r:

. O

-O

. O

- .

.;,C

O

CO

.- .-

o. o.

~

~--

U

.-- U

~<

1J.

.~.

~

~~I

] -'"

],ro!

...'" "'.j..I

;::!. ;::!.

~...,

.. O

. -.

- (.)

.. ..

.OL

.OL

. In

. In

.. ..

II,---

"U

-- "U

'"

-- C

O

CO

I..j

i.

J ~

.

~.~

I 1.1 I!

1 I I

I I

I I

II I

i, I

I I

! I

I I

I I

I I

I I

I ~

I!J ~

Ln

I!J ~

Ln

r ~ -

.194 THORSEN

f- oZ .

. s c- ~.E c-

o" CU .-o C-~... EC- ..

] / ~... C-

o 1ft... "O

o .. o GI

...- o 1ft.. ...

o ~ ..1ft.. o -;a

/ "0 ~C 1ft

o ~ \ ~c

1ft o ..u o E. <. c-oo

o"; .;c c.. ..

~ :a --- :a>- >-

f- f-o ~ .- o .-

S ~ o ~o 1ft 1ft

./ ~ / ~... ...C- C-

oo ~ -

.. ..~ o ~o 1ft 1ft

~ ~ ~ g1ft 1ftGI GI

"O "Oi o ~ ! o ~

iI I I I I I I I I I I I I I I I I I I I~ I I I . I I I I I I I I I~ ~ Il) ~ ~ Il)

o "OC..- .c-

.., j.- ~ "O

o -- co ~ ..

"O o "Q...c o ..

-J~ .-o ..

~ GI ~ I-

1ft "O

C> I- o .... >

"-

1ft > -;a ~ ~"O 1ftC

o GI l-C X ..

...~ o 1ft

CO) GICO) o -g

GI.c

o c-o o

.-- .s .~c- c-E o ; o" U U v

... \. ... Qj./ c C- :3"O "O ~GI GI '.-j... o ... {J

o 1ft 1ft... o ... ~

o .. .. o

- ~ ().. ..o ~ . ~1ft 1ft ...

' .g ' .g ~~ o ~ ~ o ~ &i, I I I I I I I I . I I I I I I . 1. I I I I I I I I I I I I I I I I ;.::

~ ~ '" 1!2 ~ Il) '-

.

-..

INTONATION AND TEXT 195

III. RESULTS

A. PRESENT AT ION OF THE DATA

In figure 1 is shown average tracings of two pairs of two-sen-tence texts, where the Amanda-sentence occurs in initial andfinal position, respectively. This is to illustrate the rela-tion between the averaged "raw'l data and the data reduced sim-plified representations in figure 2 (upon which the account ofthe results and the conclusions are based) and to give a morecomplete picture of similarities across (and differences be-tween) individual speakers. The stressed vowels are indicatedin figure 1 by lines connecting two or three points.

Although the main trends in the data are apparent in figure 1(and of course also in the tracings of the averaged "raw" three-sentence texts, which are not shown here), a simplification iscalled for for a number of reasons. The three different sen-tences are not immediately comparable, due to the distributionof stressed vowel qualities, i.e. rather high vowels in themor-sentence and rather low vowels in the Amanda- and far-sentences, which makes the intonation contour tao high in themor-sentence relative to the other two, ceteris paribus. Dif-ferent initial consonants in the stressed syl lab les within eachsentence also obscure the picture. The [ro] in mor and Amandalowers the initial Fø value compared to the vowels in the fol-lowing two stressed syllables which begin with unvoiced ob-struents. Contrarily, the [f] in far raises the initial Føvalue considerably compared to the other two stressed vowelsin that sentence. (See further Jeel, 1975, and the referencesthere.) Both these difficulties could of course have beenavoided in the construction of the material, but only - Ifeared - at the expense of the naturalness of the sentencesand texts (to the extent that they can lay any claim to natural-ness at all). As it is, these "extraneous" factors are allevi-ated if I Gut and sp1ice the texts (in the i1lustrations) sothat initial, media1 and final Amanda-sentences are presentedtogether in pseudo-texts, and likewise for the mor- and far-sentences. In this way we get a clearer impression of therelation between sentence intonation contours (a) in differentpositions in a text, (b) in texts of different length, and(c) in texts of different types, and it is the relations ratherthan the abso1ute slope shapes and va1ues which are interestingin this connection. This last contention also justifies therather drastic reduction of each sentence into two straightlines, an upper and a 10wer ane. The upper line in figure 2is the connection between the first and last post-tonic syl-lable in each sentence. The lower line is the connection ofthe Fø minimum in the first stressed vowel (the initial valuein pure1y rising movements. the turning point in falling-risingmovements) and the final va1ue in the last stressed vowel.This lower line is a reflexion af, and related to, the intona-tion contour as I have chosen to define it, namely as the lineor figure described by the stressed syllables of an utterance.It will be tao steep in the far-sentence, and not steep enough

.

, -..

196 THORSEN

in the Amanda- and mor-sentences, due to influence from theinitial consonant in the first stressed syllable. - When Isay about vowels that their initial Fø value is "too high" or"tao low" and therefore intonation contours will seem "toosteep" or "not steep enough", I mean that their physical mani-festation does not, presumably, correspond to the way they areperceived. We know that differences in duration and Fø levelbetween stressed vowels of different tongue height, ceterisparibus, are overheard or compensated for by the listener, seee.g. Reinholt Petersen (1974) and Hombert (1977). This com-pensation is explained with reference to the listener's "know-ledge" of those constraints inherent in the speech productionapparatus which are responsible for physical differences insounds which are intended by the speaker to be the "same".From this knowledge presumably derives an expectation to findcertain acoustic differences coupled with certain differentspectral events, and these differences therefore carry noindependent significance and they are disregarded by thelistener in the interpretation of what (s)he hears. If thereference to the listeneris double function as both speakerand listener (which in a more general perspective is what themotor theory of speech perception is founded an, cf. Libermanet al., 1967) explains the (non) perception of Fø and durationdifferences between stressed vowels of different tongue height,I think it reasonable to assume that it will also make thelistener disregard differences in vowel anset Fø which arecaused by the different nature of the preceaTng consonant.It is certainly my own impression from the recordings thatthe three different sentences do indeed sound "the same" asfar as steepness and shape of the intonation contours are con-cerned when everything else is equal. This is of course noevidence, and formal experiments should be undertaken to verifyor falsify my assumption, but note again that the conclusionsto be drawn from the data do not hinge upon its verificationsince I am more concerned here with relations between sentenceintonation contours than with their absolute gestalt.

In figure 2 the simplified data are presented which describethe intonational structure of three pairs of pseudo texts (asdefined above), for each of the four speakers and their mean(mean of means). Sentences are identified by their originalposition; e.g. '11-1" is the first sentence of ane, "2-2" isthe second sentence of two, etc. The isolated sentence ("1-1")and the two-sentence text ("1-2+2-2") are both presented to-gether with the three-sentence tex t (the set is complete onlyin the case of Amanda). For ease of comparison "2-2" is showntwice: directly af ter "1-2" together with the medial sentence("2-3") and also together with the final ane ("3-3").

ane type of information is last in the process of piecing to-gether the figures from parts deriving from different actualtexts, namely pause duration. All speakers paused between theterminal declarative sentences, cf. figure 1. Pauses rangebetween 20 and 80 cs. ane speaker (JBC) also paused betweenthe coordinate main clauses, but his pauses were only abouthalf as lang there (20-50 cs) as between the terminals (50-80 cs).

.

..

INT

ON

AT

ION

AND

TE

XT

197

"I

Cl)

Q)

I0Q

):5. Q

)c.t)o",~

~.Q

""",""",~'""'tQ

)f\1Q)Q

)0~f\1.j..)

~~

t)~~

~

C

l) ""'I ""'I

f\1""'I

Q)

Q)

'""'t '""'t

Q)

~~

CI)t)

'""'t bi

~~

'""'tf\1.j..).j..)Q

)Q)'""'tt)f\1

><

~.j..).j..):3C

1)Q

)Q)~

.j..)~

~.j..)~

Q)0

~~

~

I Q

)CI)~

"t)f\1~

N

I&,

~~

I")

~-

.4/' ~

Ij..j '""'t

" I

Q)""",

I f\1""'If\1""I"):3E

"i

/ ~

~I+

OI~

I """'

""'I"" I")

Q)

/ I")Q

)~'-I

~~

/ IQ

)~

~~

~I

~~

Q)Q

)+

f\1"

'-.c: .j..)

t) I")

t)I

.j..) ~

I

""'I ~

/ '""'t

"Q)

~(.j

" f\1 Q

) Q) .j..) '-

.c: '-I

".; ~

~

~

.j..)

I ~

f\1 ""'I

Q)

Cl) ""'I

Cl)

/' Q

) .j..)

Cl)

.j..) ~

Q

),

I ~

Q

) f\1

~

æ.

'""'t,

~

~

~

Q)

VI

f\1

.j..)Q)f\1Q

)~0~

><

,!:::'""'t.j..) O

'""'t

Q)

, Q

)E-.jt)f\19.

~.j..)

Q)'""'t~

Q)

~00.c:0

". C

l) t)

.j..) ~

"""' .j..) ~

" "'I.j..).

1"»<0

Q)~

C

l)I

Q)

~Q

)0~.c:

.j..) C

l) f\1

~

~

c.Q

) .c:0f\1f\1

. /

"Q)t).c:.j..)~

~

'" .

"1'1 ~

.c:

~.j..)

Q)

"""' 0\

.=

g' ":'

I .j..)

Q)

""'I .c:

~

C.>

o. ,"-

I I

:3 ~

.j..) ftI

~~

'

~

/ I

'- Q

) O

l ""'I

t-:) O

U

I" I

I Q

)Q),,~

....

: U

/

I '""'t

Cl)

Cl)

Q)

I")~.j..)

o. ,,"

" f\1.j..).j..)

I ~

Q

)'tJ

I o.

/ I

~""'I

".c: Q

) >

< I")

f\1 C

l)G

I ,

'tJ /,

.j..) C

l) .j..) '""'t Q

):;:

, .!!

I I

Q)

°"'1 Q

) ""'I

9..j..) .c:

Q,.

.c:...

: lA

"

I ~

~

~

~

~

.j..)

~

t)..

I I,

:3""'1 ""'I

O

f\1 "~

~

f\1-:

" I

bi '""'t~

t) ~

,-Q)

!:. :

/, ""'I.j..)

Q)

~'"

, -'"

I I

~

><

~blC

I)~~

CI)~

.. :

lA

" Q

) Q

) ~

""'I f\1

Q)

Q)

O'tJ

, ..

I .j..)

~

f\1 .c:

'""'tC

O

I -g

, O

~.j..)

".j..)f\1.j..)~

:

~

I ~

'""'t~Q

)""",~E

-E.

..: O

f\1

Cl)

Q)

VI

V1

~

~

~

O

O

""'Ig'

~

~"

'""'I.j..) :i

~f\1C

1).'""'t .j..)

~

~.

~

~Q

) ~

:3 .~

~

"O

N

.j..) Q

) O

Q

)f\1C1).j..)

,~

t)~.j..)\j.c:

~~

"o°,!. f\1

p,j ~

~

C

l) I") Q

) Q

)00

~

:3 O

~

f\1

I ~

.c:

o I.j..)

t)'\j~~

~.j..)

~

I >

<~

~

Q

)

I Q

)O~

"':I: ".c:p..j..)

I O

~

.j..) C

l) f\1

, C

I).j..)Q)

I '"'1

I Q

)§

.j..),!:::~~

~~

, t)

Q)E

-.j+

:3 Q

)

'0" I

~O

Cl)

~~

O""'I

... I

Q)

t) IQ

)~~

," I

.j..)t)~

c: '"'1

, ~

f\1U

~,-.j..)Q

).j..)I

Q)f\10

Q)~

~I

CI)~

Q)'""'tC

l)blf\1Qj

, f\1

Q)

.j..) O

~

, ~

~

.Q~

.j..)Q)°"'l

I 0"'1

~

O

Q)

~I

'æ'

O

Cl)

~

" Q

) I

~

§ C

I)~~

Q)O

Q),!:::

ilq)

O

Cl) 9.

t) E

-.j """'

~

.j..):3 ~

'"'1

Cl)

I I

1 1

I I

I I

I I

I II

I I

I I

I I

I I

, I

I I

I '"'1'

f\1 O

~

~

In O

~

In

O

In '""'t"""'

Q) '""'t

t)..j..) .

f\1I

I").c:t) N

Q)

~

I.j..) Q

~

::t: ~

Q) I")

Q),sQ

)~

.j..).j..).j..).j..)O

~0"f\1f\1

t)OO

'""'t ~

~

.c: 0"'1

I ~

°"'l.j..)

~

~

~

~

Q) ~

o:;:! Q

) O

f\1

~

'-""'I ~

,s

~

.j..) Q

)f\1

~

O

E:

~

°"'1 O

.j..)

Cl)

Cl)

'-~

O

.j..)

t) >

<

'"'1 ~

Q)

'"'1 ~

Q

) ~

~

~.j..)Q

)~.j..)~

0f\1p,j""'I

~

~

I c.

Q)

~

Cl) '"'1

f\1 f\1

~

Cl)

E:

. C

C"~

..'c~c.--

-.

.198

TH

OR

SE

N

O-

o::=

ij

"" .

I: -g

ru ru

- -

~

~III

III>

- >

-I-

I-

III III

:I "

III III

Lo Lo

:I :I

~

~

.ru oIU

Q.

Q.

~

~III

III

'- '-

g g

III III

Ol

Ol

"" ""

I: I:

Ol

Ol

:I: .c0\o

I':'

I-

III I

I ,~

I

I '-

I,

' I

, /

,,

I ,

, I

I,

, I

--,

" -

, ,

I v

, ,

I Q

),

, I

...,

, I

...,

" ~

, I.~

, I"

, I

~I

' ~

, o

I t>

,I C

'\I,

~

'I

.-

0\~

~

I I

I I

I I

I I

I I

I I

I I

il I

I I

I I

I I

I .1

., ~

Il) o

Il) Il)

o Il)

'-I

I

.

..

INT

ON

AT

ION

AND

TE

XT

199

O-

Ct:

æ:

I N

I.

II

' I

IN

I ,

I ,

I I

I I

I ,

I ,

II

II

I,,

I,

,,II,

." ..;

c: c:

IV

IV"Q

. -

IV

a.-'

IV-'

0- 0-

I

IG

I G

I I

.. ..

I."

,'." 'C

;',i

I I;

INI

IV

I>

I

>

II

I-

I -

IIV

I

IV

".ot

'.ot I

.. I

.. I

I I

.. I

I'"

I ..

I...

I '"

II

...I

.. I

..G

I I

GI

." I

."I;

I C

~

I 1O

lo

'OQ)

:3~"'"i~g('-I

I ~

~

il III

II .I

II I

I II

II I

I I

I I

I I

I I

I I

I ~

Il) o

Il) Il)

o In

'-,

,

I ' -

.200 THORSEN

[.

u 1a1 '";'1-, ",I

II, I '? I

1 I CO)I I

1 I II 1 II 1 II I II , II 1 I1 I I1 , I1 , II I II 1 II , 1I , I1 , 1

, 1 I 1, 1 , I

, I , 1, I 1 1

I II ,1 I1 I,

11

It;-'"," , .'

0\' 0\ I 1,,' "I I

.- '", I .- 'I

~ i~ : ~ l:ru " ru l:u , I U 1 I, I ,

.ru , I .ru "Q. " Q. "

"" 'I ""':

QI /: Ol l I... , I ... Ion " on 1

... , 1 ... Iru 'I ru I

, I 1- " - Iru , I ru.:t / I .:t.. I ..

ru 'I ru"" 1 """ I "ru I ru

E Ec( <t

0\O

~:.'" :oL ::::r ~

: :: : : . ~. . QJ

: : :3! : ~: : ""-i, " ., . . ~, : : ..., " . ..., . . O,. ., : : U,. .I : C'I

! : ! I ~

i, I I I I I I I I I I i, I I I I I I I I I I S'" o '" o

.

-

,

INT

ON

AT

ION

AND

TE

X T

201

u!:Q-," "

c c

lO

lO-

-~

~

.. ..

M>

- >

- I

I- I-

NM

': ~

I':

:J :J

.. ..

L- L-

:J ~

~

~

..." .10

a. a.

- -

lO

lO~

~

.. ..

L- L-

a a

.. ..

.. ..

" "

c c

.. ..

:I: .cO

lo

II")II

;I:~

II I

I I

II

I I

I I

IIII -..

I ~

I Q

JI

:3I

I t:

I :

' I

I .j.I

I I

I I

...

I ,

oI

: (J

,,, C

'JI,

! ~

~

il I

.. I

I I

I I

. .

I i.

I I

I I

I I

I I

I ..

SIn

o In

o

.

-,

~~. ;,.202

TH

OR

SE

N

i1fI!.,~'.[,.'

'->~~

NI,

I

N,I

/ M

I

II

/ !

I

/ :

IIIIII

-O

.,;c:

c:!J.

!J.a.

a.'V

'V

-' -'

I N

'.-

M

I .-

,I.l.

I '"

NI

Ol

.. ~

O

l' I

... I'

... N

I

I"

IN

I "

"c:

I I

c: I

I'V

I

I 'V

I

I>

I

I >

I

II

I I

I-

I I

- I

I..

I I

'V

I I

.x I

.:( I

I'"

I '"

I I

I I

I...

I ...

I I

'V

I I

I

... 1/

~:

:'"

'" I

Ol

Ol

I"

",c:

c: I

Ol

Q)

I:I:

.cCl

O

-..~Q

):3t:.j.Jt:Qt)

~

~

~

.~

~

0\

i, l

I II

I l.

,I'I i,

I, I

. I

I I

III ~

~

o In

o .,.

(,.:l"j~.

-.

.IN

TO

NA

TIO

N AN

D T

EX

T

203

j,,t,,,ffi

N

iii...Ii,.,,

. !

. '

.,.,,,,,,,

,

, .

0\ 0\

tO

tO

'o. '"

'o. "

E

u E

,

.. '"

,U

/

U

",

, ,

.,. ,

.,. "

Co

1 C

o "

1 '

1"tO

1

"tO

,+,.,

.. 1

.. 'N

'...

1 ...

, 1

VI

1 V

I ,

1'+

- 1

'+-

, 1

.. 1

.. "

1,

, 1

- ,

-, 1

.. 1

.., 1

.ol 1

.ol, 1

'" 1

.., 1

1 "

,. 1

,., 1

"tO

, "tO

1

tO

1 tO

1

,. ,

,. 1

~

1 ~

1

0\o

'i'

.: : 'O

: ,

Q)

I :

:3, :

t:;, .

.-,, .

.~I .

4.JI: ,.

...,.

O':

(j,.J.E

C"I

.~

J

'" .

~

~

.t:;I.

. I

. I

I I

I I

I I

I I

I .

~I

I I

I I

I I

I I

I I

I I

I I

~~

~

In

~

~

In

f..

-.

.204

TH

OR

SE

N

~(.!J. .

-o -o

c c

." ."

:; :;

.. ..

>-

>-

~

~.-

.-..

..:J

:J..

..Lo

Lo:J

:J~

~

... .."

Co

...-

-."

."~

~

.. ..

Lo Lo

g g

.. ..

Q)

Q)

-o -o

c c

Q)

ol:I:

.cCl

o

I,I.I.,, I ~

, Q

JI

~:

~,

."",,

~,

~,

o,

: Q

,,, ('-I

,~

~

: .

- ~

bi

E

E

.""~

I t

I I

I I

I I

I I

I I

I I

I ~

I I

I. I

I I

I I

II I

I t

t ~

II!~

In II!

~

In "'"

-


I

,

f:Q~

,,'",,,,,,,,,, ,, ,, ",""

"

"OC

-o ~c a.'" '"- -Ja.'" .-

-JOl.- ...

Ol "O... C

"O , >'"c , ,'" , ,, - ,> , '" ,

, '" ,- ,.. ,'"' ,'" , .. ".. , '" ,

, '+- ,

...' ,, ,'" , .. ,'+- , Ol ,, "O ,.. , c 'Ol , Ol '"O , .c "c' ,Ol ' Oi '% o

~Qj:3~

""'\.j..)

§(.)

C'I

~ ~ "~i, I II I , , I, I I , II I ~I I I I II I I I I I , I I I ~

~ 52 In ~~ In

.

-:-

.

206 T

HO

RS

EN

I-z:

,,,

I'1' I

. N

'C

l C

l'C

,

C

, ,

.- ,.-

" ,

Q,

, Q

, ,

,e

' e

, ,

'" /

'" I

"U

,

U

,,

,.",

,"" "

Q,

, Q

. "

, I

,"tJ

, "tJ

"G

I '

GI

"...

, ...

I I

lA

" lA

"

/"-

, "-

"'"

I '"

",

I ,

- I

~

I"'"

, '"

,-"

I -"

I',lA

"lA

"I

,'"

I '"

,"tJ

' "tJ

IC

'

lO

,'"

' '"

,,

e I

~

' C

( ,

Cl

o

" ,I

'" "

'1' 'I I.

- /-I'

,I'

0" ,

I' I

I' ,

I.. ,

" ,

-..I

o' ,

" ,

vI..

, ~

" ,

..."

, ...

" ,

,.."

I ...

I 0'"

, °,""

'" I

. .

I' ,

I.. ,

~/

o' ,

O,"

,,,",

r,.,

vo'

I..0'

~

! 0'"

~

.~

~I

I ,I

I II

I II

I ,

I I

, I

I II!

I I

I ,

I I

I ,

I I

, I

I I

I ~

~

S2

'" V

2 S

2 '"

.

..


f-z

..; .c -o

IV c

- IV

~ ~.. ..>- >-

I- I-

.. .." ".. ..'- .." "

~ ~

.., ..,Q. Q.

IV ";ij~ ~.. ..'- ..

g a.. ..., .,-o -oc c., .,

:r .I::

0\o

I~.,' 'I' ". ~I" I

I "I I

" II I

I II I

I I -.,I I I I v

I I QJ" " :3I " ~I ""i

I ~

I' ~I o

I CJ

\"',j

! ! O;- ~ . ""i

i, I , I I , I I I I I I , I " , il I ' I I I , , I . 'I' I I'! ~~Q-- \li ~ ~ \li '-

:4";;",i

.

-.

.208

TH

OR

SE

N

I--z

,,'

,

." ..;

c C

IV

IV-

-Q

. Q

.IV

IV

-' -',

,.-

,'- ,

GI

GI

.. ..

." ~

co ,

CIV

,

IV

,>

,

>

,

" ,

- ,

- ,

IV

, IV

'

-" ,'~

'

.. ,..

',

"..

" ..

,IV

,

IV

''+

- ,

',

,..

" In

"G

I '

GI

,."

" "

,co

, c

,O

l '

GI

':c

' .t:O

lo

--'"QJ

:3~"I"oj~8C

'\I

~

! "~

~

E

~$1

I I

I I

I I

I I

I II

I II

II $1

I I

I I

I I

I I

I I

I I

I .

I I

'-~

~

In

~

~

In

.

. -


z<IL!~ ,

,

I. . N IOl Ol cl..:CO CO I.- .- I I

a. I a. ,:E : E I I.. I .. " IU , U I I

I I '... I ... I IIa. I a. II I ,~ ' ~ I:Ol : Ol I I... , ... I I'" ' '" I:"- ' "- ".. : .. I I, I I- , - I I.. I .. I I-'" , -,", II'" ' '" I ,, I.. : .. I :

~ I ~ ICO I CO ,.. ' ~ I

E '< M

OlO

-..~Q}~c:.j.Jc:OtJ

l'\1

5 ~ .~i, I I I I I I I I I I L I ~I I I I I I I I I I Il J . ~

~ In o ~ In o

-,

~

210 THORSEN

z<u:J~

" .c ".. c- ..

-" -"' -">- "'

I- >-I-

"':J "'"' :JLo UI:J Lo

~ :J-"

...C- ...

C-

.. --" .."' -"

"'

g ~E

"'., "'" .,C "., (;

% .,.c0\O

~, I I',' I '" I_I 'I

I -,I I II 1-

I II I

I II I

I II I

I I" I ""I I r.-

I I ...,I I n.

I / ..,

: :3: ~

'I"i~

§~

C\j

~ :il .;: (; ~= " ".;

~I I I I I I I I I I I I. ~I I II I I I I I I I I I ~$2 '" O $2 '" O '-"

.

-..

I~JTONATION AND TEXT 211

zo:(uJ~

,II

II

II

II

I I,I I, I

/I

II

I

, .-o -oc c.. ..

'Q. 'Q... ..

..J ..J,- ,-

GI GI... ...

-o ~ IC I C I.. I I> I" I

I > II I- I - I.. ,'.. "

~ I ~ I'" I '" I

I II I... I ... I

.. I .. I"- I "- I

I I

'" " '" "GI I GI I

-o I -o Ic " c IGI I GI I

X I .c

Clo

-..~Q):3t:

"'i-I.Jt:o(.)

("J

~ ~ ~j, I I 'I , I , I , , .I .. i. I , I , I I 1 , I 1 l J ;;::~ In o ~ In o "-

.

-

.212 THORSEN

NRP1

~ . "". ~II) ~ *..-. ". *; "" ~ :-..""--- * ":""". ""'"

~ ~ '*'~ *"'... .. .. .' .., ...., * * "".. .::..:: " """"'" * ".* *. -" "...'

1 2 3 1123111231123111231123

Amanda og Amanda Hendes mor og hendes mor Hendes far og hendes far

JBC

II)

II)

~, ~ ~ .IH. ~ ~ , :;~~< -. "-

-~'V"""'" .. ."." * ". ': ". ..,.. .."."", ..,. ...,. . .'" ...' '. .'

..

1231123111231123111231123

Amarlda og Amanda Hendes mor og hendes mor Hendes far og hendes far

Figure 3

Slopes (in semitones/second) of the upper and lower lines de-picted in figure 2. The three different sentences are arrangedhorizontally in pairs, withdeclarative terminals to the left(points) and coordinate clauses to the right (stars). Datapointspertaining to texts with two components and with threecomponents, respecti vely , åre connected. Upper line data pointsare connected with full lines, lower line data with dotted lines.Unconnected points in the mor- and får-sentences represent ini-tial and final items, respectively, in two-component texts.Where unconnected point s are not clearly associated - in the

graph - with either upper or lower line data, identification is

ensured by a short full or dotted line next to the dot or star.Four speakers, two males (NRPand JBC) and two females (GB andNT) and their grand mean (mean of means) - identified at the

top right of each graph.

.

-


GB

I

--~L-- , ~"', .00000.'..-In ' ~.,.:::o . .

~ -. o'In~ v "', J . .' . . .0 . '. : . '. o o'

. o o*,~ .'~ 00:

:"","'0 0,,'00'.:'.' '. ~ * .o. ~ "0 .,.., '.- ..0' ..' . 00'"

t::. ..0'"...'

1 2 3 11 23 U, 231123 n 123 1'23

Amanda og Amanda Hendes mor og hendes mor He~des far og hendes far

NT~1 . :J.. 000. ..'" .~ "'-- .*In *0 * .0".. . '. .! ~ o ~ / ~t :--'-- '. .

~..~. t.:" , """ ...,*. . ...".'

1 2 3 1'2 j 111231123111231123


MEAN

In . '.~ '-3~ "':-:o:! ~ : ",' ! v , ""-- ."'~~ ..a... ~..., ""'."';'.

..-~, """"' ., ..., . . .'.:'" , 0

.~.

1231123111231123 11'231123


(figure 3 continued)

I

-,

214 THOR SEN.

B. INTERPRETATION

I shall take up the following aspects of the data: upper lineslopes and lower line slopes; the frequency of beginning points(onset) and end points (offset) of upper lines as well as lowerlines; the frequency location of sentence and clause initialunstressed syllables. All these phenomena are evaluated alsoon the basis of the numerical information upon which the figuresare based. I shall state the results for the average over allsubjects, and note the extent to which individuals agree withthe grand mean. (The fact that subjects agree so well amongthemselves, qualitatively, is what permits the calculation ofa grand mean in the first place.)

1. UPPER AND LOWER LINE SLOPES

In figure 3 the slopes (in semitones/s) of upper and lower lines ~iin the three pairs of texts are depicted graphically for each -,

speaker and their mean (mean of means). The following pointscan be made from these figures:

1. U per lines are steeper than lower lines, and the differ-ence lS sma er 1n t e far-sentences, c . above about the

"overestimated" initial Fø value which makes the lower lineslope relatively tao steep in this sentence: all speakers.See also figure 2. This is a common observation and not aspecifically Danish phenomenon. Note also the difference be-tween JBC's less steeply declining lines (varying around 2!semitones/s) and those of the other speakers (varying around4 to 6 semitones/s).

2. Upper and lower lines are steeper in terminal declarativesentences than in coordinate main clauses, ceteris paribus:

all speakers.See also figure 2.3. between success' . and

1n a succeSS10n nin c nstruct10ns: a .figure 2. This is a natural consequence of the slope differ-ences noted under (2), granted that a speaker's total Fø rangedoes not change from ane type of text to another. ~Ji

Same further tendencies (but that is all they are) towards adifferentiationaccordlng to position in the tex t appear:

4. The u per and lower line slopes in an isolated sentenceare equa to or ver s 1 ht steeper t an 1n1t1a y 1n

mul t1-sentence texts re evant on y or term1 na sentences : .:

though with JBCupper and lower lines are less steep in theisolated sentence. Otherwise, upper and lower line slopestend to be more nearly identical in initial sentences and infinal sentences, irrespective of text length. I.e., 1-2 re-sembles 1-3 and 2-2 resembles 3-3 (rather than 2-3). See alsofigure 2.

!

-..


5. Upper and 10wer line slopes in the terminal sentences tendto be steeper in initial than in non-initial position in

three sentence texts: thereare random exceptions with allspeakers, typica11y where a final slope is as steep as theinitial ane. The two sentence text offers no uniform picture.See a1so figure 2.

6. and 10\'/er line sl es in co rdinate m in c1auses areteep ln me a -

wh are appro e sen ence ex s~an are steep y ln two sentencetexts: there are a few slight and random exceptions withevery speaker. See a1so figure 2.

The fu11 and dotted lines in figure 3 run approximate1y parallelto each other, considered frame by frame, with a coup1e ofapparent1y random exceptions. In other words, the differencein upper and 10wer line slope is near1y constant across a givenmu1ti-sentence text. I.e., given a certain change in 10werline slope across a two- or three-sentence text, the upper linesfo110w suit, and can be deduced from the behaviour of the 10werlines. (Arguments to the effect that the 10wer line - thestressed syllables of the utterance - is the independent vari-able are given in e.g. Thorsen 1980b and 1980c. Suffice ithere to say that every utterance has a 10wer line (one or morestressed syllables), but an utterance does not invariab1y havean upper line which depends for its existence upon the presenceof a post-tonic sy11ab1e in the prosodic stress group(s).)The steeper upper lines derive from the fact that the magnitudeof the Fø interval between the minimum in the stressed sy11ab1eand the maximum in the post-tonic varies with position on theintonation contour (ear1y - 1arger interval, 1ater - smallerinterval), cf. Thorsen (1980b, 1984).

From the figures and (1) to (6) above we 1earn that the indi-vidua1 components are subordinated different1y to the super-ordinate dec1ination which spans the who1e text. Individua1dec1inations are steeper and have greater amounts of resettingbetween them in a series of dec1arative terminal sentences thanin a corresponding sequence of coordinate main clauses. Further-more, in a series of three terminals the individua1 slopes tendto become successive1y 1ess steep, whereas coordinate c1auseslevel out in media1 position and a final dec1ination is a1wayssteeper than a preceding (initial or media1) ane. This can betaken to indicate a more integrated and coherent intonationa1structure accompanying the c10ser relation between coordinatemain clauses.

Finally, it appears that increasing text 1ength from two tothree sentences/c1auses ar, inversely, decreasing text 1engthfrom three to two, wi11 1eave the initial and final componentsunaffected. A component is squeezed in or removed media11ywith no apparent changes in the surroundings. As a resu1t,the second component in a text is different when it is simu1-taneous1y text final from when it is media1 (compare 2-2 and2-3 in figure 2). In other words, the production of the second

~

-,

.216

TH

OR

SE

N

-..IJ) Q

) I

. -1-1 Q)

o ~

C\l1::~

9.-I-IQ

)~"--i

~

~

~

'--i ~

t!j

Q)O

-l-l-l-lO~

-I-I--~

~

°--i

tJ:3--~

~

I

~~

IJ)b\

-1-1 ooO

Q)

°'"'i -1-1 °'"'i

11! :i

'""i4-t4-t~

-I-I-I-Ib\OI1!

Q)

11! ~

~IS

~

'""i 'O

'O

I::

°'"'i Q

) Q

)°'"'i

Q) ~

-1-1 °--i § ~

4-t

'O

~

I::..."'"'i o

Q)

-1-1 IJ) °'"'i

IJ) b\ . ~

.-1-1

-I-IO~

Q)~

-I-IIJ)

tJOI::~

Q)~

11! ~

°--i -1-1 Q

) -1-1

-1-1 'O

~'-

~

~

'O

"'"'i ~

IJ)

I:: 11!

0- /.'

'"/

*; '"

~

Q)

IJ) o

I:: Q

) ~

H

~.'

'. 'O

'""i

r'\Q)'""i""

~

biz

o °

on t;'

000

. O

l 11!

11! -1-1 -..

. o.'.

N

. ~

, N

"C

IJ) I:: O

... I:: °'"'i

t.) 4-t00

:0 ~

Q

)°'"'itJlJ)°'"'iQ)-I-I~

O,,'

: .c

I:: ~

Q

) 4-t O

I-:)

.° -

*' -

Cl

""'i~O

I:: ~

'O

-1-1O

'""i

Q)

:i °--i 'O

-1-1

'O

Q)

-1-1 -1-1 '""i I::

°'"'i Q

) I::

IJ)-

- -..

11!~~

11!~

Q)~

'""i -1-1

~O

~-I-I'""i'""i1

Q,tJ

/.

'"I

. """

'""i°'"'i°~:311!

0~11!

o :

~

Q)-I-I~

4-t°--i~-I-I~

Q)

.. .

~11!

-I-I~--

.. :

on --~

IJ)~°'"'iI1!-I-I

4-t.

!. N

.

!. N

.g 11! -1-1 -1-1 I::

~

><

IJ) O

o o

C

~

'""i >

< °'"'i

°--i bi

Q)

Q)

.: ~

~

92

tJ Q

) ~

I:: '""i-l-l

- ~

~

Q)-I-I

-1-1 Q)

11!.cL::

0'0 'O

1::~Q

)~bI

, '""i

O

Q)

Q)

-1-1 ~

°--i-

=

~

-1-1 -1-1 IJ) °--i

O

~'0-1-1

tJQ)I::'""i:i

... 1::°'"'ib\Q

j~°'"'i-l-l~

/:!o.'"

~.

"'g 11!~

1::1::~

'O

O.000

:0 °'"'i

I:: Q

j I

Qj

1-100°'

,o lA

-..

... ~

O

~

-1-1 IJ)

..,0 N

.0

N

~

Qj

IJ) °--i tJ

'O

-1-1 ~

Qj

000 ,°'

C

~

~

11! IJ)

Qj

O

Qj

~0,0

0,° i.

"" O

"'"'i-l-lQ)Q

j-l-l'O~

-I-I.0'.

~

... -

Cl

~

11! ~

~

tJ 11!

11!I

O

Qj

I1!~Q

jI1!I::°'"'i~Q

j-l-l~

--

~

QjtJO

PoI1!

- :3

I:: IJ) -1-1 O

IJ)

- b\

~°'"'iIJ)-I-II::IJ)'""i

'O""'i

Qj

-1-1 I:: Q

j IJ)

'""i ~

Q

j...

R.

~~

I::°--iIJ)I1!:3:3°'"'i/

o' '"

J:.

'" g

po '""i °'"'i

O

I 4-t

O

4-t00

: :3'""i0~

~~

R

. ° '"'i

,: :

on 11!

Q,

\~

'""i -1-1

-1-1~

N

,

N.g

Qj-l-l

b\t.=..~

~

I::

:' ..0

~

~1::11!1::

11!0.Qj

: ..:1:

-I-IO-l-l'--i'O

Qj~

W'O

. .0.

- ...

- W

I1!I::I::'""iIJ):X::°'"'i

4-t "'"'i A

I:: ~

tJ

O

~

°--i 11! o

I=

=

O

b\I-I-I

OIJ)~

.Q

j~O

~""'-"

-1-1 -..

~,~

~

~

IJ)

A.'"

11.

"'lO

~'O

IJ) E

:: O

l::,o"

: "C

"'"'iQ

j~

Q)'O

-I-I11!o'"

.: ~

o

b\ 11! .

Qj

~

Qj

Qj

..0'.. N

,..°.

N~

~

I::-I-I'O~

I1!~ti]~

° ,o

. o

~

IJ) Q

j -1-1 :3

'O':0.000

::0 C

l 'O

~---I-I

ti]1J)1::4-t'00"

:: O

I::~

tJ~

-I-I~O

O.

- -

~

Qj 11! -1-1 Qj '--i I:: Q

j po.cL::

I:: °'"'i

IJ) I::

'OQ

jblI::IJ)Oti]Q

j~-

- 1::~

°--iO-l-l~

"--i~Q

jtn

tn I1!I1!~

tJ~

~~

UJ

UJ

"--i 'O

I:: O

--~ F

. '"

~~

. '"

b\ti]Q)Q

jOQ

jotJ ~

o:

~:o

lO

~Q

j~~

~-I-I°--i

~

: :

"C

°'"'i tJ

-1-1 11! tJ

-1-1 Qj

11!Q

: ..

. N

ffi",

N

~

I:: ~

'O

Qj 11!'""i Q

jW

:

000 :&

:: ..

E

I:: Q

j O

... Q

j I::

tJ 11! ~

~

:0" O

:.. -o:

°--i-l-l-l-l~-I-II::°'"'itJ

~.

.00 . -

--J.. -

b\ I:: '""i tJ O

4-t IJ) 'O:

Q)

Q)

IJ) Q

j Q

j tJ

°'"'i I::

~IJ)Q

j~~

§-I-I~

11!,

I I

, I

I I

, I

I I

I I

, I

. I

I I

I IJ) °'"'i I::

~

tJ ~

'~-'o

'" c5f

4-t-l-l:3-1-10 Q

j~b\

I O

I::I1!

tJtJ Q

j'OQ

jS

3NO

1IW3S

S

3HO

1IW3S

IJ) ~

~ ~

:§ ~ °'"'i 6. o~

QjQ

j IJ)

~

...Q)Q

j:34-tQ

)Qj

~11!~

~'""i4-t-l-l~

' -1-14-t-l-l

11! °'"'i 11!

IJ) 11!

~'O

I::...Qj

.'OtJ~

"'"'i IJ)

~

IJ) °--i

I::~

Qj'O

-I-I°'"'i-l-l-l-l~~

tJQj~

I::'""i><

I::~I::

~

O

Q)

Qj

'--i -1-1 -..

Qj

~

O

I:: 'O

-1-1 O

°'"'i

E-t

:3-1-1 tJ

OQ

j ~

~~

bi ~

-I-I-I-I 11!

QjQ

j'O~

-I-II::'Ob\'O

~~

I::OO

QjI::O

~R

.I:"1l1!tJ'OI::Q

j'""i11!

,'.

.

-.


'-U ) . M J * M~ ~.. o -,.0 :..

. . Qj

~...N.". N-". . C

. . Qj. '.co .. - . - Cl

O

- -I . M ) . M'- . . ru o' .0 "-

00 : ... ~o. N . ,. . N Qj

. . -"" . C" . Qj.0: :I:- . -

- -'-} ..0. M I .., g 00 .. . Qj .. N N -" : C

: o Qj. ..c.. - * . - Cl

O

- -) .., j o:O.'" ~

o o Qj

~. N .. N -"o : C. . Qjo .:1:... - ..-

I- -/, . .., f..., ru f::~ N 't;.:o~ N ]:: : Cl 00 : O ~, - - Q)

:3~- - °--i

(/) (/) .j.JW W ~= ) " M :: A..'" o -J: -J: ru (J

ffi ~~ N (t:~. N ~~ ~Q. . o W..

.0 X..Q... O 1. .::) .. .. - -J - t)II "--i

"L I I I I I I 4-1"' o II~IIII '-

S:JNO1IW3S S:JNO1.-.3S

.

-

.218 THORSEN

L-

~ / ..M ) ... M~ c.!J. . ...' :::

. ,jO'.. N * .*.. * N-o. . c:. . Q)... c:

.. - * - '"O

- -/ . M ) . M L- . . ru

. i. N . L N;. .-0

. : c:. o Ol. .:1:. - .-

- -J ...'. M ) .0'... M ~..0' .0..0..

..' N ~. N ~. o c:

o : 1:-.0 ~ . - O-

O

- -L-) ! M / ! Mg : :..

. . Q). N . N -Oo o c:. . Ol

: ::1:. ...c ..-- -,) .* CO) J .. CO) ru . '-0

. .0' o.'" iij0,,0. .0.'. '"' o ...0. N ~

0. 0:000..0 g' ~..0- - .' - Q)

~c:- - o...

U) II) .j.JUJ UJ c:Z f . M ~ t . M O :i : ..J : (.)

ffi + N ~ \ io' N j "3'

Il. .' O . <.Il..' . ...:;) ... _..J .. - Vl11-1

~I , I I , I , I I I I I II I , I , I. I I I '-~ ~ O '"

S3~J.IW'" S3~J.1~3S

-

,


I- / ..'. M / .. M -.' .' L-

~.' ....,., ,'.... *..0.0' *... * ,..,.'. ... '". . O)

'. '-O, .' c:'. - ., - O).c:OlO- -~ ..' M I ..e M ...' .0.0 ~

. .:.' . .." . . '", . O)'. :-0. - . ~ c:

O):I:- -/ . M / . M .' . :' L-

o," : ~..0 ... . N. . '"

. . O). '-0: .. c:. .. - ... - O)

.c:Ol- - O/ ~ M / e M .' : L-

,: : ~.' N ; N, . '"o . Q). .-0

: : c:. .. - . . -O): :I:- -/-: .".". M ./ ..'. M

,..".,.,' .,...' ~.., " .'.. N c:. " ,,' ~

f..",,"."" f.,..""'..:*,".. - .'. - Ol -..O ~

QJ- - :3~

""'!U) . M U) k . M .j.) W " W : ~ Z .,' Z : C

... .' ... , C.>.J ~" .,,' N.J ,:,e N..

. ..' .-0IX .,.' IX " c: ~.- ."W ,'" W . . ..g: ? .,,".. . -; ~ . .,'..'. - ~ .J .J bi

""'!~""~"""'~"""!'" I'I'!I"I~I" --

S3NO11~3S S~NO.lI~3S

-...

220 THORSEN

~/ .* '" / .. "':; U.I ...' : ...

~. '..* f'" * N * i * N -8. . c:. . Ol

: :.c:* - .. - C\

O

- -/ . M / . ML. . . .. .. .. ...

. ~"'. N . ..'. N ~. : .": . c:. : Ol. ~ . -:I:

- -/ .. M ) * M8 ..,,'.. ' lO

...' N.." N -8. ' c:

.. .. Ol. ..c:.' - .. - C\O

- -/ :." M ) ;.. M 8. ."

,. N f N-8, . c:: : Ol. ':1:.. - .-

- -,/ .-* M J1 . M.. ..." .,..-g

«...".. N ...". N~i,...,.". f..,."". g'«~- - - ~

QI- - ~~

.'"

Wj . M ~ ~ . M ol.) UJ : Z . ~

Z .. _:.. O-. '." .,-J ... N -J ... N c: V. .0 . o ..!t: : o' !t: J. E ~UJ o.' UJ ~& . ~ ~ ... . - .:;) -J t)I

""'~""!"" II!'!"!, 't.

S3NOJJ~3S 53NOjJ~3S

.

-.


component is sensitive not only to what precedes it but alsoto upcoming events. This argument extends to isolated versustex t initial sentences, of course. See further section IIIbelow.

2. UPPER AND LOWER LINE ONSET AND OFFSET

In f~gure 4 beginning and end point frequencies (f~ll vs. dottedtrac1ngs) of upper and lower lines in the three pa1rs of textsare depicted graphically for each speaker and their grand mean(mean of means). From these figures the following points canbe made:

First of all, there is no qualitative difference between textswith terminal sentences and texts with coordinate main clausesbut

7. changes in anset and offset frequencies across a text arelaraer in coordinate main clause constructions:

all spea ers~ ut W1t N t e di erence etween terminal sen-tences and coordinate clauses is small in this respect. Thegeneral trend is a consequence of (2) and (3) above: clause con-tours are less slanted relative to the same overall decline, soonsets and offsets decline more through the text. See alsofigure 2.

With this difference in mind, I'clause'l can substitute for Ilsen-tence" at any point in the following four statements (8-11):

8. Up er line onsets decrease smoothly across each text:a spea ers, W1t spur10us except10ns to t e smoot ness

by GB and NT. A final upper line anset is 3-4 semitones lowerthan an initial ane, on the average. Upper line onsets arevery nearl identical across initial sentences, including the1S0 ate ane, 1rrespect1ve o text eng, an across 1na

-2which resembles 1-3, and 2-2 resembles 3-3: all speakers,except JBC where 1-1 resembles 2-2 (and 3-3). See also figure2. Initial u er line onsets are also identical across thetwo Dun ary con1t10ns: a spea ers.

9. U er line offsets decrease across each text but not quitesmoot y, S1nce t e rap rom to 1S smaller than

between 2-3 and 3-3: with JBC and NT we even get a I'rise-fall", i.e. the end point in 2-3 is higher than in 1-3 butlower in 3-3 than in both preceding anes, most notably in thecoordinate clauses, where it is due to the way upper line slop~sdevelop: a steep initial one, then a levelling out medially,and then again a steeper final slope, cf. (6) above. Offsetsalso decrease by 3-4 semitones (average) from initial to finalposition. Up er line offsets are very nearly identical acrossinitial sentences exc u 1ng t e 1S0 ate one , 1rrespec 1veof tex t length, and across final sentences, includin the iso-lated one: all spea ers. ee a so 19ure 1na upper 1ne

.

-.

.222 THORSEN

offsets are identical across the two boundary conditions:

all speakers, except GB where they are lower in coordinate

main clause texts.

10. Lower line onsets decrease across each text: NRP has ane

instance of a step up between 1-3 and 2-3, and GB gives

no uniform picture. A final lower line anset is 2 semitones

lower than an initial ane, on the average. In the lower line

the anset of an isolated sentence is intermediate between on-

sets in initial and finalsentences in multi-sentence texts

(relevant only in the terminals): with NRP and NT it is much

closer to text initial lower line onsets, and with JBC it is- conversely - closest to text final onsets. See also figure2. (If we compare upper and lower lines it is apparent thatlower line onsets and offsets span a smaller Fø range than inupper lines, and so the same SJ_i9_~_~Y lower beginning of upperand lower lines in an isolated sentence compared with initialsentences in multi-sentence texts will bring isolated lowerline onsets relatively nearer to the anset of lower lines infinal sentences.) Otherwise, lower line onsets are very near-ly identical across initial sentences (i.e. 1-2 equals 1-3)and across final sentences (i.e. 2=2 equals 3-3): all speakers,with a couple of spurious exceptions. See also figure 2.Initial lower line onsets are identical across the two bound-aryconditions: all speakers, except GB's mor-sentences.

11. Lower line offsets decrease across each text: no speakeris as regular as the grand mean. As with upper lines,

lower line offsets are very nearl identical across initialsentences exc u 1ng t e lS0 ate ane , 1rrespect1ve o textlength, and across finalsentences, including the i so la ted ane:with NRP the isolated sentence resembles initial rather thanfinal sentences. See also figure 2. Final lower line offsetsare also identical across the two boundary conditions: allspeakers, except GB where they are lower in final coordinateclauses (as are her upper line offsets, cf. (9) above).

With a reservation about the anset of the lower line in anisolated sentence (cf. (10)), the ensemble of upper and lowerlines in a text span almost the same Fø range, irrespective oftext length and boundary condition. The upper and lower lines,respectively, begin at the same frequency in tex t initial sen-tences/clauses and end at the same frequency in tex t finalsentences/clauses. Thus, the superordinate declination varieswith text length but it is the same in both boundary conditions.The fact that both onsets and offsets of upper and lower linesare identical in initial sentences (1-2 resembling 1-3) and infinal sentences (2-2 resembling 3-3) ties up with the conclusiondrawn from points (1)-(6) above. The second component in atext is different when it is simultaneously text final thanwhen it is medial, and the same is true of an isolated versusa text initial sentence. Together the upper and lower linesbear testimony to a considerable amount (in terms of temporalscope) of preplanning and look-ahead in the production of in-tonational events. It is possible - though it remains to be

tested - that only three different textual positions are

,

!

-..


distinguished, as far as sentence intonation anset and slopeare concerned: initial, final, and medial. That is to say,two ar more medial sentences ar clauses may not be furtherdifferentiated among themselves. Such an arrangement af indi-vidual text components would be less taxing an the speakerispre-planning operations (cf. section III).

3. INITIAL UNSTRESSED SYLLABLES

In text initial position the Amanda-sentences have only aneunstressed syllable, [a], the mor- and far-sentences have two(the second af which may be carried by a syllabic nasal as aresult af schwa-elision) but only the first ane, [hen], willbe dealt with here. In text medial position an unstressed og,[A], is added to the beginning af each coordinate main cl au se.This syllable, [A], is - with very few exceptions - higher thanthe succeeding [a] ar [e], and by nearly the same amount foreach subject (0.5 to 1.0 semitane with NRP and JBC; 1.0 to 2.0semitanes with GB and NT). Therefore, the conjunction can bederived from [a] and [e] and is not considered any furtherhere.

In figure 5a,b,c the frequency af sentence ar clause initialunstressed [a] (Amanda) and [e] (hendes mor, hendes far) ininitial sentences (Sa), final sentences (Sb) and across athree sentence text (Sc) is depicted, for each speaker andspeakers. grand mean (mean af means). Again, I shall statethe results for the mean and note the extent to which each sub-ject agrees with the overall tendencies.

12. The pretonic syllable(s) in tex t initial sentences areconstant across texts ofdifferent length and different

syntactic conditions: all subjects, with exceptions by GB,cf. figure 5a. This (near) constancy in text initial unstres-sed syllables is analogous to the constant anset af lowerlines (i.e. to the constant frequency af the first stressedsyllable) across tex t lengths and boundary conditions, cf.(10) above.

13. Sentence initial unstressed syllables in text final sen-tences are constant across different text lengths in

terminal sentence texts: there are spurious exceptions byNRP, GB and NT, cf.figure 5b. This corresponds to the be-haviour af the final lower line anset, i.e. the frequency afthe first stressed syllable (cf. (10)). In coordinate mainclause constructions the retonic syllables in tne last clauseten to e ower 1n t e anger text: lS an excep lon;. his final pretonicsare constant in this condition as well.The general tendency is likely to be due to a higher offset afthe declinations preceding 2-2 than the anes preceding 3-3,i.e. the final Fø minimum and the (derived) peak in the pre-ceding clause is higher in the first clause af two (1-2) thanin the second af three (2-3), see figure 4 ("lower lines"-data points connected by dotted line) and also figure 2.

.

-.

.224

TH

OR

SE

N

Q)

ti) +

.JQ

) 1\1

,~

0~'tJ~

~

. o-i +

.J ° o-i

Q)

II) Q

) t,j

'""i 'tJ+

.J ~

"-

ti) I.i

+.J

Q)

+.J

1.i'tJ001.i ti)

~1.i0'""i;3~

Q)

oSel)~

blO'""i

o ~

°o-i

1\1

'""i,Q)~

oe::J'I ti)

I.i I.i

Qj

~

'tJQ)+

.JI\1~Q

)~~

'-

'- ~

-!<

~

°o-i

~E

~

/O

1\1

°o-i ti)

O..:

... ::J'IO

Q)'tJ

~

1.i.Q~

tI)Q)

.+.J

Q)

;3 +

.J l)

~C

!JIl

C!J

. C

!J ~

'tJ~I\1+

.JII)'tJ-

- -

p.jQ)+

.J'""i0 ~

z ~

+.Jo';:!l)'""iQ

)1\1«

N

~

1\1 oS

~

I.iU

J .:.

- Q

)l)'tJ ;3

~

ti) °o-i

'tJ ~

Q

) bi

I:J

- O

'tJ

Q)

1\1 I.i

~~

~

,;. .;::

~

+.J

1\1 ...

~

° o-i

1\1 -""'

l)Q)tI)~

'tJti)

ti) °o-i

~

Q).o-i

~

.

lilli!! IIIIIII

IIIIIII Q

) 1\1'tJ

O

ti) 1\1

Q)

ol) '"

ol) l)

~

~'tJ

I.i~

'"o-i

'tJ 1\1

Q)

n. ~

Q)::J'I

Q)'""i+

.JP:;~

+.J

'""i Q

) +

.J l)

+.J

~

° o-i

~'""i1.i1\1

O"-~

Q)

1\1 1\1'""i'tJ'""i

1\~

1~

. ~

ti)

l) O

~~

tI)-G°o-i

ti) ti)

l1j Q

) C

j

I- ..

N

Q)

+.J

(\) °o-i

Q)

'""i 1\1

~

,;..:. ~

I.iCj

-""'1.i1\1Q)

, +

.JQ)~

Q)Q

)rue~

(\)~~

~

~

~

+.J+

.JOtl)O

O-

'o-i 'tJ

~

+.J

~

(\)(\)~'tJ>

<+

.J~tI)~

tI)~(\)Q

) O

(\) ~

° o-i

+.J

+.J

,+.J

'""i1\1'""i'tJtU

ti)

\1C

!JI

\~

.

~

.Q1.i1\1~

'""i+.JI.iQ

) -

... -

1\11.i~'""i0~

Q)~

'""i 1\1

° o-i C

j ti)

Q)

-!<

+.J

~

N

N

1\1 '""i

E~

"o-i~1\1

(.!;I .:.

.:. li)

:JI Q

) I.i

° o-i O

Q

) +

.J

tl)1.iQ),,-Q

)g.p.1\1-

(\) ru+

.J ~

etl),;.

I.i 'tJ

ti) +

.J O

'tJ

~

Q)tI)(\)Q

) l)1.i(\)

~

ti) Q

) ~

C

j ~

;3

'o-i

"o-i tl)l)'o-i~

~Q

)0~I

I I

I I

I .

. I

I II

..II I

I II

L. L_J

r... Q

) ~

+

.J (\)

°o-i Q

) r...

°o-i$!

$! $!

I.i (\)

1\1 +

.J 'tJ

I.i +

.J

+.J+

.JI.i~;3~

~

tI)~ruQ

j'""i+.J

.Q)

~(\)'""itl)l)

N'tJ

~

ti) l)

~

~

:x:: °o-i

(\) '""i

°o-i "o-i

~~

li~

.-

~

'""i +

.J 'tJ

1\1 "-

a I

ru ~

'o-i

ti) a

u °o-i

Qj

O

+.J

ti) e

~

'1 ~

+

.J I.i

+.J

.o-i Q

) (\)

ti)

-, °o-i

Q)

~

l) +

.J o

~

~

~

bI.o-i ~

'o-i +

.J ru

.,. 'o-i

~

~

Q)

Q)

- °o-i

°o-i +

.J +

.J '""i

ti) e

Q)'tJ

~

><

~

ru'tJ

ti) .o-i

Q)

Qj

~

~

~~

Q

) ru

+.J

ti) °o-i

o o

1\1(\)+.J

~p.

\\C!J

V~

-. C:'

'""il.il.ie'""i tI)~

- -

- l)~Q)01\1'tJQ)1\1

+.J~

I.i~~

I.iQ)

a.. N

N

I.i

~'o-il\1l.ie

~

.:.,;, O

Q)ru

~

0"-~

+.Jru

'""il)-

Q)

E.j

1\1 +

.J +

.J 1\1

~.:.

l) Q

ru

><

"o-i

Q)

ru

~

QQ

)'tJ'""i(\)(\).

+.J

Q)

1\1 e

+.J

C"\I'

e l)

IIIIIII III,III

1,111,. ~

~

'e tI)'tJ

'5 e

s Q

)Q)~

tI)O'

~tl)1.i1\11.i1.i'""i::J'l1\1

S;I-;O

.lIW3S

".~

hl\1~I\1l)1.i

V1

V1

+.J

°o-i ~

~

O

'o-i ti)

1\1 +

.J Q

)

~

~

+.J

°o-i ;3

I.i

ti) l)

~

1\1 ~

O

I'o-i .

Q)

~

1\1 +

.J 'tJ

" o-i Q

) Q

)

~

°o-i Q

) 'o-i

I.i ~

'""i lo.~

,e~

+.J

1\1'tJ ruo

~Q

)01\1li)1.il)'tJ+

.J +

.J ~

°o-i

~

::J'Il)+.J

ru Q

) E

rul)

'o-i -G

'tJ

I.i ru

~

~~

OI

~+

.J+.J-"",

Q)

Q)

"o-i Q

) ~

1\1

'o-i e..

~

'tJ I.i

ti) ""i

'tJ I.i

~

Ol

~~

1\1

(\)Q)Q

)ru 'tJbI'tJ

1.i1.i~'""i~

~0~

r... ru

+.J

l)" o-i

ru '""i

ru

I,I'\,

. !rJ;~Ct~-


lO"CClO .. ...æ 2 ..'!

Æ II : .. :;: V :

2: \1 I I I I 1 1 I I I I I I I 1 I 1 I I I I

'" '" '"

{I :;: .. :;: // :;:

I- r:- 1/ NZ N '"

:::I\S

L()

QJI,j(/ :;: ..:;: Y :;: "~

!=9~ r:- .Q ~~ .. N L()

O- QJ +J-'" I,j

~ ~~ ~

" QJ

'I~"" I'~""II II~III.I ro. ~'-!

QJ~1 :;: .* :;: il :;: ~

U N N QJI:Q '" '" rn-, vA

~ II: .*:;: \l:Z \ N N

I I I 1 1 I I 1-' I I I 1 1 I I I I I IIO - ~ - - - - O

S3NOJ.I~3S

-..

226 THORSEN

10cor:

~ " ~c( e '+-

~ ~ c:>V M J M 1 M

~ :;: re ~

~ ~ ~

I I I I I I I I 'I I I I I I III I I IIn '" '"

.., , .., M]J '" )l '" )l '" r-.., c:> ~ - . '" '" - '" .

f Z :zt\j

lf\.Q).., .., ~ ~I; '" JJj'" JI M :3 ~ 'ooi I:Q c:> c:> ~ t..

t!:> '" '" Z :z:z ~ E

Q) ~~ t:

~ I I~ I II I II ~II II ~ ~>1 -"ooi Q)

~ '"'i

Q)M P t:) ~ ~ :5 '" '" M U Q) ~ ~ :;: :;:: ~

.., t:) ..;. - -

.., ~ t:)Y '" ) .., 1 M & ~ t:) :;: z '" '"

M ~ t:).;. - -

I I I .. I I I I I I I I I I I I II I I Io (; o

S~J.IV;3S

.

..


At least this difference is greater, and corresponds better tothe difference observed in the clause initial unstressed syl-lables in figure 5b than does the slight or even reverse dif-ferences in the anset of lower lines in 2-2 and 3-3. Thiswould mean that clause initial unstressed syllables in textmedial position are aoverned tonally by the offset of the pre-ce 1ng c ause, or - 1n ot er wor s - suc sy a es JO1n upprosodically with the last stress group in thepreceding clause.They are nelther isolated tonally,-nor do they attach them-

} selves to the clause.to which they belong syntactically. Thisinterpretation gets a certain substantiation from figure 1(and from the corresponding longer texts not reproduced here):tKere are a majority of instances (exceptionless with GB andnearly so with NT) where clause initial unstressed syllablesperform a smooth and continuous fall together with the post-tonic in the preceding clause; together with the last stressedsyllable in the preceding clause the ensemble of unstressedsyllables thus perform the low plus high-falling Fø patterncharacteristic of prosodic stress groups in Standard Danish(cf. Thorsen 1980b, 1984). Only with JBC are there no suchcases of prosodic association across clause boundaries, maybebecause he (as the only ane) paused between the coordinateclauses. (I do not think that a pause necessarily disrupts atonal pattern (apart from the introduction of a silent inter-val),.and clause boundaries per se apparently do not, but thecombination: pause and clause boundary may be more than an Føpattern can survive.)

14. Initial unstressed syllables throu h a multi-sentencetext ecrease s 19 t t roug a str1ng o term1na sen-

tences: t e var1at10n lS sma an non-un1 orm W1t ,nearly nil with JBC, apparent only in NT's moY'- and faY'-sen-tences. With GB the decr.ease is considerable in the Amdnda-and moY'-sentences, cf. figure 5c. Again, this is analogousto the onset of lower lines, i.e. to the behaviour of thefirst stressed syllable in each clause in the texts, cf. (10)above. In coordinate main clause texts the decrease is greateraQd__~~~llymoreuni.form: all speakers. Again, this is morein line withthe way upper line offsets develop through a tex tthan with lower line onsets (cf. figure 4), i.e. it seems tobe caused by the tonal association of the clause initial un-stressed syllables with the preceding clause final stressgroup, cf. above.

What points (12)-(14) demonstrate is merely a tonal dependencyof initial unstressed syllables in the sentence or clause upon

. other parameters: In tex t initial position, initial unstres-sed syllables are constant (for a given speaker), as is thefirst stressed syllable, across different text lengths anddifferent boundary conditions. Af ter a text medial sentenceboundary, initial unstressed syllables likewise assoc;atewith the succeeding stressed syllable, i.e. differences infrequency location across different text positions are analo-gous with and can be derived from differences in the firststressed syllable in the sentence. Af ter a clause boundary(unaccompanied by any pause) cl-ause initial unstressed syl-

.

-,

.228 THORSEN

lables (often but not invariably) associate tonally with thepreceding prosodi c stress group, i.e. differences across dif-ferent text positions are analogous with the differences inthe last tonal pattern in the preceding clause. In otherwords, sentence and clause initial unstressed syllables intext medial position may be extrametrical phonologically (andrhythmically as well, af ter asentence boundary), but tonallythey are not isolated entities; they vary in a principledfashion in association either with a preceding or a succeedingprosodic stress group.

C, DURATION

I have measured the duration of each sentence and clause, asthe time interval between the first and last Fø measuringpoint, excluding the conjunction in the coordinate clauses.Average values for each speaker, and speakers' mean are givenin Table I.

The differences in duration of a given sentence across tex tpositions - in ane or the other boundary condition - rangebetween 16 cs (NRP's coordinate fdr-clauses) and 3 cs (NRp.sterminal fdr-sentences), i.e. 13.7 and 2.5%, respectively,of the shortest sentence/clause. The most consistent trendin the durations is that an initial sentence or clause isshorter than the succeeding,medialor final one (true in 29out o lnstances. eGon y, a flna sentence or clausetends to be shorter than a medialone in three sentence texts(true in f6out of 24 instances). Isolated sentences are notsystematically different from sentences inc6ntext:thoughwlth JBC the lsolated sentence is clearly longerthan initialsentences (6 and 7 cs), it is only 2 or 3 cs longer than medialand final sentences. With GB the isolated sentence is 7 cslonger than the final sentence of three but only 1 cs longerthan initial sentences. Lehiste (1975) found that isolatedsentences were longer than in three sentence texts. To judgefrom the present data, such a finding may not generalize toall speakers. Furthermore, coordinate main clauses tend un-expectedly to be longer (by around 6 cs) than terminal sen-tences (which is not due to the conjunction og which was ex-cluded from the measurement). The opposite relation wouldperhaps have been less surprising: A terminal sentence de-velops freely, a non-terminal clause is shortened by thesucceeding clause. I have no explanation to offer at themoment for the longer coordinate clauses.

A more thorough treatment of durational differences warrantsa separate investigation. The only conclusion I wish to drawat present is that sentences or clauses are shorter text ini-tially than in later positions but - somewhat surprisinglymaybe - isolated and text final sentences or clauses are notsystematically longer than sentences in other positions inthe text.

I

.,


Table I

Average durations in centiseconds of three different sentencesin various contexts: isolation ("1-1"), initially ("1- "),

medially ("2- "), and finally ("3- ") in texts consisting oftwo (" -2") or three (" -3") sentences, and in two differentboundary conditions, as indicated to the left. See furtherthe text.

NRP JBC GB NT mean

Amandaterminalsentences

1-1 128 135 119 137 129.81-2 128 128 118 131 126.32-2 130 132 117 136 128.81-3 131 129 118 136 128.52-3 132 133 118 137 130.03-3 133 133 112 143 130.3

coordinateclauses

1-2 131 136 114 136 129.32-2 139 146 122 141 137.0 ~1-3 129 136 120 134 129.8 :?

2-3 134 141 125 141 135.33-3 136 135 120 137 132.0

Hendes m6r ...terminalsentences1-2 148 157 140 145 147.51-3 148 153 139 146 146.52-3 156 157 144 149 151.53-3 152 157 145 148 150.5

coordinateclauses

1-2 151 154 141 150 149.01-3 152 153 140 150 148.82-3 159 160 145 156 155.03-3 159 160 141 153 153.3

Hendes far ...terminalsentences

2-2 121 138 111 124 123.51-3 121 128 118 127 123.52-3 120 132 116 131 124.83-3 118 129 109 126 120.5

coordinateclauses

2-2 122 144 118 128 128.01-3 117 133 114 130 123.52-3 133 135 124 134 131.53-3 123 130 121 131 126.3

.

-..

230 THORSEN

D. SUMMARY OF THE RESULTS ON INTONATION

A short text, composed of two or three simple declarative sen-tences or corresponding clauses, of which none is excessivelylang, will be associated with a superordinate Fø declinationwhich supports individually slanting sentence/clause components.The most interesting and consistent further findings, i.e.where the inter subject agreement is best, are points (2,3)and (8 - 11): Upper and lower lines - which characterize in-dividual sentences or clauses - are steeper and with greateramounts of resetting between them in a succession of declara-tive terminal sentences than in a corresponding succession ofcoordinate main clauses. Concomitantly, the anset and offsetFø values of these upper and lower lines decrease across atext, more so through coordinate main clauses, because indi-vidual lines are less slanted relative to the same super-ordinate decline. The decrease is greater in upper than inlower lines, ceteris paribus, because upper lines - individual-ly and as a whole - span a greater Fø range than do lower lines.These facts are depicted in the stylized graph in figure 6,which shows the course of upper and lower lines in two differ-ent three-sentence texts, ane with declarative terminal sen-tences (full lines), and ane with corresponding coordinatemain clauses (dashed lines). (Differences in duration andpause length are obliterated.) The dot ted lines reflect theoverall downdrift which is identical in the two boundary con-ditions. They connect the (upper and lower) anset values in

se~T(HS

,~:-.::-::-.-: '-:-.',.. ..

. . .::::;;:;;;""'-'""'~ ~~ ~ : .:::

I 1 1 I 1 I I 1 .. I I 1 I 1 I 11 I 1 I I I I I I I I I I I I I I I I I I I II Io ro ~ m 4OOlcol

TIME

Figure 6

Stylized model of the course of upper and lower lines in textswith three declarative terminal sentences (full lines) and threecoordinate main clauses (dashed lipes). The dotted lines re-flect the overall downdrift through the text, which is identicalin the two boundary (sentence or clause) conditions. A text withtwo components is obtained by leaving out the middle section inthe figure and moving together the initial and fipal componentsacross the gap. An isolated sentence results if the dottedlines are suitably compressed in time.

.

-.,


the first sentence/clause with the (upper and lower) offsetvalues of the last ane. These values can be considered con-stant for texts of different length (and with differentinternal (sentence or clause) boundaries), which is anotherimportant aspect of the results. An isolated sentence resultsif these dotted lines are suitably compressed in time.' A textwith two sentences or clauses is obtained by leaving out themiddle section of figure 6 and moving together the initial andfinal components across the gap. Since the frequency span is(assumed to be) constant, an isolated declarative sentence willdecline more steeply than the ensemble of two or more sentencesin a text: The results actually do show that isolated slopesare at least as steep as the steepest (initial) sentence inmulti-sentence texts. Text initial unstressed syllables arelikewise constant across different texts, whereas text medial,sentence or clause initial, unstressed syllables vary in aprincipled fashion in association with a succeeding and pre-ceding, respectively, prosodic stress group.

It should be evident from the account of the results that thedescription is true of the overall, average picture exhibitedby the tour speakers in this investigation. If either ane ofthe tour speakers had been the only ane in the experiment, theconclusions would have been slightly different on same points,because a certain amount of inter and intra individual varia-tion or I'noise" exists. However, this variation is randomti.e. ane speaker may differ in ane direction from the maintrend in ane aspectt and in another direction in anotheraspect; or the same speaker may deviate on a certain pointfrom the main trend in only ane of the three different sen-tences recorded here. Therefore, the presentation of a grandmean over tour speakers is possible, on the ane hand, whileat the same time we can assert that no individual is as regularin his or her behaviour as the resulting model (figure 6)would predict. Speech production is a more plastic processthan such a model can capturet also where the intonationalstructuring of multi-sentence/clause texts are concerned (cf.Thorsen 1984, where I reach a similar conclusion about the pro-duction of Fø patterns in prosodic stress groups).

The less steeply declining upper and lower line slopes in co-ordinate clauses vis-a-vis terminal sentencest ceteris paribus,corresponds well with the observations in Thorsen (1978) thatisolated terminal declarative sentences have steeper sentenceintonation contours than corresponding non-terminal (declarativeand interrogative) clauses.

The fact that different text internal boundaries (clause bound-ary or sentence boundary) induce differences in intonationalmanifestation is not a reflection of a general, tight syntactic/prosodic interplaYt cf. Thorsen (1980a): which treats wordboundaries and Fø patterning - and Thorsen (1983a) - where langsimple declarative isolated sentences are analysed and therelation between intonation contour resettings and sentenceinternal syntactic boundaries are discussed.

.

-

.232 THORSEN

E. COMPARISON WITH OTHER LANGUAGES

First of all, the present results - like Lehiste's (1975) andBruce's (1982) - clearly contradict Nakatani's conclusion thatspeech features [including prosodic anes, NT] can be reasonablygeneralized to sentences in a coherent context. The upper andlower line(s) of a terminal declarative sentence span a greaterfrequency range in isolation than in a larger text context.Secondly, Standard Danish agrees well with the gross text in-tonation features of English and Swedish. All three languages(and Japanese as well, cf. Uyeno et al., 1979) exhibit a super-ordinate text intonation structure upon which are superposedindividual sentence or clause components. There are a numberof (minor) differences, however. I cannot assert, as didLehiste (1975) that an isolated sentence is longer than thesame sentence in context. Lehiste further found that Fø inthe beginning of a sentence is lower in an isolated ane thanin paragraph initial position. This is not true here if weconsider the very first unstressed syllables, which have aconstant Fø value independently of the length of the succeedingtext. If we consider the Fø value of the first stressed syl-lable and especially the first succeeding Fø peak (in the firstposttonic syllable), then Lehiste's finding is true of anespeaker but the isolated sentence is only slightly lower (lessthan ane semitone) than paragraph initial anes with threespeakers. lehiste's observation that Fø in the beginning ofa sentence is higher in paragraph initial than in medial andfinal position holds for Standard Danish as well, no matterwhat Fø point we consider to be the beginning, and this differ-ence is not negligible.

Bruce's (1982) observations from his larger material, thatthere are initial and final Fø values which are constant acrosstexts of different length, hold true of Standard Danish aswell, though these constant points pertain to different unitsin the chain of syllables in the two languages. In SouthernSwedish the very first and very last syllable lie at the floorof the range of normal Fø variation. But the first local Fømaximum and minimum (in the first stressed plus unstressedsyllable) in Swedish is lower in an isolated sentence than incontext (corresponding to lehiste's (1975) observation), andit is higher in a longer than in a shorter text. The lastlocal Fø maximum and minimum is constant across texts of dif-ferent length. In the present material, not only extrametricalsentence initial unstressed syllables but the frequency loca-tion of the whole of the first prosodic stress group is almostinsensitive to text length. Thus, the total range of varia-tion varies proportionately (upwards) with tex t length inSouthern Swedish, whereas it is constant in Danish. Con-comitantly, superordinate text declination will vary more inDanish than in Swedish with the duration of the text.

Before the text internal pauses in Bruce's recordings Fø dropsto the floor of the range (where also absolute initial andfinal syllables are constantly located). Such a marking oftext medial boundaries is not replicated here. This may have

.

INTONATION ANO TEXT 233

to do with the tonal relation between stressed and unstressedsyllables in Standard Oanish: the first post-tonic syllablerises above the preceding stressed ane, succeeding post-tonicsdescribe a more or less steep fall. In the present sentencesthere was only ane sentence/clause final post-tonic, and adrop to the floor of the Fø range is therefore not possible.The high sentence final fall in Bruce's material may, however,also be due to a sentence accent, which evokes a particularlyelaborate Fø movement in the posttonic syllables, and Oanishlacks such a nuclear accent (cf. Thorsen 198Gb). According toa personal communication by Gosta Bruce, Southern Swedish mayalso be produced without a final focal accent, but it is notclear to me whether this was in fact the case in his (1982)recordings.

In Bruce's texts, the Fø minimum in the first accent af ter atext medial sentence boundary continues at the same Fø valuethat the Fø minimum had in the last accent in the precedingsentence. He sees that as a tonal coupling or assimilationof the succeeding constituent sentence to the preceding ane,see further below, section III. lam inclined to think, though,that this relation is an artifact of his material, due to theparticular length of his sentences (two stressed words in each).With longer sentences he might have found the anset of a suc-ceeding sentence to be higher than the offset of the precedingane, as it is in the present Standard Oanish material, cf.figure 6. Incidentally, the Oanish data confirm Cooper andSorensen's (1981) observation, that the resetting betweenclauses in a complex sentence is independent of any accompany-ing pause.

Uyeno et al. (1979) found no difference between coordinate mainclause constructions and successions of simple terminal sen-tences in Japanese. In Standard Oanish, the Gloser relationbetween coordinate main clauses is accompanied by a more inte-grated and coherent intonational structure than a series ofterminal sentences.

IV. ON THE ABSTRACT REPRESENTATIONOF INTONATION

By "abstract representation" I mean the level where significantand distinctive choices are made and from which the phoneticimplementation can be deri ved by rules. It contains no re-dundant specifications. This representation may correspond toa stage (early) in the actual speech production process. Thus,a representation of intonational data does not necessarilybecome more abstract by mere transformation of frequency scales,or by transcription of Fø curves in terms of high and lowpoints or contour shapes. The statement may seem trivial, butsuch a concept of "abstract" is not a matter of course, atleast not in same of the present day phonological schools.

-,

~

234 THORSEN

In this section I shall briefly present two different, currentmodels of intonation, the hierarchial and the tonal sequencerepresentation; discuss two particular claims about the short-comings of the hierarchical theory; and discuss the presentdata in the light of both representations.

A. THE HIERARCHICAL AND THE LINEARINTONATION THEORIES

On a number of previous occasions, most recently in Thorsen(1983a), I have presented data and arguments in favour of amodel of Standard Oanish intonation in terms of a hierarchical,layered system of simultaneous, non-categorial intonationalcomponents of varying temporal scope, from sentences throughprosodic phrases and prosodic stress groups to segments. Ishall not repeat the reasoning here, but note that to thehierarchy of intonational components we may apparently add atextual contour. Similar views of the composition of funda-mental frequency courses - with such modifications as languagespecific differences impose - are expressed about Japanese inFujisaki et al. (1979), about French in Vaissiere (1983), aboutOutch in 't Hart and Collier (1979), and about Swedish inGårding (1983) and Bruce (1977).1 However, recently Bruce(1982) seems to be leaning more towards the basic concept ofthe linear theory of intonation as formulated by Pierrehumbert( 1980) .Pierrehumbert (1980) analyses English intonation in terms ofa linear sequence of categorially different, non-interactingpitch accents, which are associated with the stressed syllablesof the utterance. They consist of either a high tone (H), alow tone (L), or bitonal combinations of the two. The inven-tory of pitch accents includes a (final) phrase accent and aboundary tone. The resulting description is very elaborateand very elegant, if occasionally somewhat complicated.Overall slope - if it is tao steep to be accounted for byother factors - is handled by a downstep rule which lowers aH tone by a constant factor relative to the preceding H, incertain environments. Everything else being equal (such asthe prominence relations between the stressed syllables ofthe utterance), this rule will create asymptotically decliningslopes. Liberman and Pierrehumbert (1983) modi fy and expandthe theory to include variations induced by changes in pitchrange, as well as a lowering of the final pitch accent.Once the Fø value of the initial pitch accent is determined,such a theory requires no pre-planning, no look-ahead, on thepart of the speaker for the execution of intonational phenome-na, and "In general, we see no evidence that the Fo implemen-tation for an entire phrase is necessarily laid out befarespeaking begins, even when the phrase is known in advance andfluently produced. All of Dur measurements can be modeledquite well on a left-to-right, plan-as-you-go basis.11 (Liber-man and Pierrehumbert, 1983. The quotation is from the manu-script; I do not have the final version at the time of writing).This is also made explicit by Pierrehumbert (1983, p.. 141):

.


"Superficia11y global trends arise from iterative app1icationof these 10ca1 rules." and "We a1so deny that intonation isbui1t up in 1ayerst by superposing 10ca1 movements on a globalcomponent."

Pierrehumbert and Liberman see the data on Standard Danish in-tonation (specifica11y the fact that the slope of the dec1ina-tion varies with utterance type and functiont cf. e.g. Thorsen1978 and 1980c) as a cha11enge to the idea that intonation canbe genera ted by 10ca1 ru1est i.e. ru1es with a narrow domainor temporal scope (extending backwards). They meet the cha1-1enge with a proposa1 for a downstep ru1e (whose specific for-mulation is different from the ane they posit for Eng1ish)where either the downstep factor or the reference line (whichis part of the formu1a for target va1ue ca1cu1ation) is a110wedto vary with sentence type. (Thorsen 1983a+b address thi6specific issue.)

The essentia1 difference between the two types of intonationmodel t a hierarchica1 versus a linear anet is not mere1y formalor notationa1 but conceptua1 or ideological t in a hackneyedterm. One theory c1aims - or imp1ies tacit1y - that certaintgross aspects of the intonation of an utterance are anticipatedand planned fort or 1aid outt at the moment the utterance isinitiated. The other theory denies the existence of such pre-p1anning and 100k-aheadt except that variation in total 1engthmay be staked out in the va1ue of the very first pitch accent.One sees the fundamental frequency contours of utterances as acomposite of contributions from, among others, an intonationa1and an accentua1 component. In the linear theory intonationis made up af, consists af, a sequence of pitch accents.These aspectst the 1ayered versus simp1y sequentia1 and theglobal versus local planning of tonal events, are what Pierre-humbert (1980) and Liberman and Pierrehumbert (1983) are mostconcerned with when they opt for the linear, tonal sequencemodel. They do not concern themselves so much with, e.g.,descriptive adequacy. This and other problems are treatedextensively by Ladd (1983a and b).

B. LADD'S CRITICISM OF THE HIERAR-CHICAL REPRESENTATION

Ladd (1983a and b) modifies Pierrehumbert's (1980) theorywhi1e staying within its general framework of linearity andlocality. He restricts the inventory of pitch accents andreplaces the downstep rule with a downstep feature. Ladd addsa couple of other features and claims (I think rightly) there-by to achieve a less comp1ex description of intonational phe-nomena. Ladd (1983a) specifically treats the generation ofoverall declining intonation contours without recourse to pre-planned grid-lines (cf. Gårding, 1983)t or baselines andplateau's (cf. Vaissiere, 1983) or other such abstract contourshapes, among which he counts the sentence intonation componentwhich I have posited for Standard Danish (e.g. Thorsent 1978and 1983a). In his discussion of the relative merits of the

.

-....

236 THORSEN

two types of intonation models, he makes a couple of state-ments about the hierarchical model which I would like to ad-dress, because they are likely to obscure the issue.

Ladd (1983a, p. 40) sees a difference in the two theories intheir weighting of phonetic explicitness against functionalrelevance. Functional generalizations are conveniently ex-pressed in descriptions of overall shape or slope of sentencecontours in the hierarchical model, but phonetic detail isironed out or last, he says. Once such detail is admitted,Ladd claims, the number of different sentence intonation typesmultiplies rapidly, and he sees no basis for extracting further,explicit functional generalizations. Ladd refers here to theresults on sentence intonation in lang sentences in Danish(Thorsen 1983a): At and above five prosodic stress groups,sentence intonation contours are not straight lines but becomebumpy, due to partial resetting of the declining slope. Thetheme is picked up again later (p. 50), where Ladd assertsthat not all hierarchical models assume that the lines con-necting accents should be straight, since "...Thol"sen..admitsthe existenae of irregular intonation lines, though she is ata loss to aaaount for them or fit them into her overalt vie1Uof sentenae intonation." This conclusion is Laddls, not mine;moreover, it is incorrect. The fact that lang intonation con-tours are bumpy does not exclude them from being globally pre-planned, and it does not prohibit the extraction of explicitfunctional generalizations (about sentence type and the like).But the intonational realization of the same function, sayterminal declarative, may Game out in different shapes, orvariants, depending - inter alia - on the length of the utter-ance. A complication arises from the fact that the locationof the bumps (the resettings) in an intonation contour is de-termined in an intricate and - so far - nowhere near fullyinvestigated interplay with the syntax and semantics of theutterance. This is (probably) what Ladd alludes to when heclaims that Ilirregular" contours cannot be accounted for.However, the problems of prosodic, syntactic and semanticinterplay are not specific to the hierarchical model of in-tonation, and they are problems that await a massive amountof empirical studies for their solution. Thus, I would stilllike to claim that the variation in sentence intonation con-tour shape is principled (allowing, of course, for inter aswell as intra speaker variation or "noise"); i.e. it is con-text determined, bound variation, even though the governingprinciples are not all known to us yet.

C. PRE-PLANNING TEXTUAL CONTOURS

Thorsen (1983a+b) contain a discussion of same specific draw-backs that I think the tonal sequence theory faces in the de-scription of Danish intonation. Suffice it here to repeat theconclusion: If the downstep rule or feature is to capture theamount af variability and complexity in intonation contours insimple sentences of varying length and syntactic and semanticmake-up, it must be a scalar and continuous feature which will

.

-..

INTONATION AND TEX T 237

aften have to take different values within ane and the samephrase or sentence. I.e. it must have at least same of theproperties it presumably was intended to rid intonation ana-lysis af. - lam of course fully aware that the hierarchicalmodel faces just as serious problems where the formalizationand description of fundamental frequency data are concerned,but I repeat that the real difference between the two theoriesis neither formal nor notational but lies in their assumptionof hierarchical, globally preplanned versus sequential, locallydetermined tonal events.

Before I proceed, let me stress the fact that the texts ac-counted for in section IIIwere composed solely of declarativesentences and clauses. I do not know how a series of questions,strung together into a text, would have behaved, or alternatingdeclarative and interrogative sentences. Nor can I make anyclaims about longer texts or texts composed of shorter orlonger constituent components. And last, but not least, thepresent analysis pertains to highly planned (read) speech.The production of an ensemble of utterances, constituting acoherent text, may be much less regular in free, spontaneous,non-monitored speech.

The discussion to follow can be limited to the behaviour of thelower lines, since the upper lines are closely correlated withand can be derived from them, cf. sectionIII.B.1 above. In it-self this fact may be taken as evidence for the layered compo-sition of fundamental frequency: the lower line is the connec-tion of the stressed syllables of an utterance, the intonationcontour proper in my framework. The upper line connects thefirst post-tonic syllable in the prosodic stress groups. Astress group may or may not have post-tonics, i.e. there mayor may not be centres of suspension for an upper line. How-ever, the lower line is oblivious to this vacillation. Thestressed syllables are frequency scaled in relation to eachother, without regard to the presence or not of any "highs"in the surroundings. The lower line is a constant, ceterisparibus, upon which high-falling patterns are superposedwhose range and extent and detailed shape are determined bythe context (earlier o: later, on a mor~ or less slopin9.con-tour) and by the duratlon of the prosodlc stress group (lnterms of number of syllables). See further Thorsen (198Gband 1984).

First of all, I find it difficult to accept that patterns suchas those in figure 6 could be obtained without a fair amountof pre-planning and look-ahead on the part of the speaker.More specifically, the different arrangement of the individualsloping (lower) lines according to the strength of the boundarybetween otherwise identical sentence's in a tex t can only bethe result of look-ahead and pre-planning, it seems to me.The step down between the first and second, and the second andthird stressed syllable, i.e. the slope of the line connectingthem, is different when the unit is bounded by asentenceboundary versus a clause boundary, in a manne r which makes threesuccessive coordinate clauses less differentiated and more co-

.

-;,

...

238 THORSEN

herent relative to the overall declination exhibited by thewhole text than three terminal sentences in succession, cf.figure 6.

The other - as I see it - crucial point ties up directly witha very explicit hypothesis about the domain af phonetic imple-mentation rules in Liberman and Pierrehumbert (1983 - onceagain the quotation is from their manuscript):

"NoUJ aomes the aonjeatural part: UJe insist that the aomputa-tion of any parameter or objeat Y[i]. ean only depend on the"aaaessiblett prope2"ties of Y[i] and Y{i-l], UJhere Y[i-l1 meansthe immediately previous objeat of the same type Cif anyJ.Thus, pitah aaae.nt ac:;zn look back to previous pitaha(Jaent,phrase to previous phrase, eta.

... A restriation of this type has a aertain funatio~al value,for both speakers and hearers; speakers (Jan ~et on UJith thetask at han d UJithout knoUJing all the dett;zils of UJhat follows,UJhiZe hearers ean in prinaiple aomplete the phonetiq processingof UJhat they have heard up to any given point in the streamofspeeah.

Many apparentinstanaes of antiaipatory effects are k~01.l)n.Unless our aonjeature is UJrong, all suah cases must tu;pn outto be explained either by feature siJreading at the phonologiaalleve!, or e!se by aomputation ofsome parameter of a higherleve t aons tituen t. tt

I will not argue the functional value af this conjecture. But,functional ar not, I think it is clearly contradicted by thefollowing fact (cf. III.B.1. and B.2. above): the slope af asentence ar clause contour, occupying a given slot in the left-to-right arder af components, takes different values accordingas it is succeeded by ane ar more components, ar not. Thus,an isolated sentence has a steeper slope, the downstep isgreater between successive stressed syllables, than when thesame sentence is succeeded by ane or more (that is immaterial)sentences in the same text. And sentence or clause no. 2from the left is steeper when it is simultaneously text finalthan when it is tex t medial. In fact, at least three positionsare distinguished in longer texts: initial, medial, and final.Note specifically that the variation with text length (iso-lated sentence versus multi and two versus three sentencetext) is not primarily expressed in a lower or higher begin-ning of the whole cohtour. (Such a phenomenon could be accommo-dated in Liberman and Pierrehumbert's representation by mani-pulating the overall range and the deri ved reference lineagainst which the initial pitch accent target value iS' com-puted.) On the contrary, beginning and end points are essenti-ally constant across texts af different length. The overalltextual contour slope varies, and so do the slopes of itsindividual components, according not only to preceding, butalso to upcoming events.

.

-.


This last observation differs from Brucels (1982) conclusion.As mentioned in sectionIII.E above he found that an Fø minimumin the first accent af ter a text medial sentence boundarycontinues at the same value that the corresponding Fø minimumhad in the last accent befare the boundary. He says (p. 285):"The observed tonal adaptation of the earlier part of a sue-eeeding sentenee to the later part of a preeeding one ~ithinthe same textual unit ean be taken as an indieation that thereis a partial rather than a total preplanning of the global Pøeourse in relation to the length of the aetual text unit."If - as I speculated above - this particular intonational con-figuration of successive sentence components is an artifactof the material, and if it would not be replicated with longerindividual sentence components, then Brucels hypothesis ofpartial rather than total preplanning is no longer so easilysustained.

Similar contradictory evidence against exclusively backwardsextending tentacles was found in intonation contours in de-clarative terminal sentences of varying length (Thorsen 1983a).The second stressed syllable is lower, relative to the firststressed syllables when it is simultaneously sentence finalthan when another stressed syllable follows, and it is lowersucceeded by ane than by two stressed syllables; likewise forthe third stressed syllable. The phonetic implementation isdemonstrably not independent of or insensitive to succeedingevents.

If data of the kind presented in this paper is to be accommo-dated within the theory of locally determined tonal sequences,say Liberman and Pierrehumbert in the last paragraph of thequotation above, it must be explained either as featurespreading at the phonological level, or else by computationof same parameter of a higher level constituent. I cannotsee that feature spreading at the phonological level is anythingbut the linguist's way of formally accounting for the preplan-ning of articulatory events; it does not "explain" these events.And I cannot see that referring anticipatory effects to ahigher level constituent is anything but admitting the existenceof hierarchically organized tonal events.

Experiments like the present ane are the most favourable wecan imagine if we want to know what speakers can achieve byway of precision and constancy in speech production tasks.Not only is the speech monitored and planned for them, butthey get to repeat the same items over and over (though not instraight succession). Even so, the signals we pick up are"noisy". This intra speaker variation (as well as the more con-siderable variation among speakers) ,is perhaps less disturbing,less alien, to the representation of intonation in terms ofgross, global events which are filled in with more detailedcontributions (i.e. of lesser temporal scope). Precisely be-cause the actual execution of tonal events is the result ofinteracting larger and smaller scale components (because theyare expressed in the same medium), we should not expect theseevents to be performed with mathematical precision. It seems

-..

,.

240 THORSEN

to me that the tonal sequence theory, and the concept thatgave birth to it, focus tao heavily on just such a mathemati-cal precision and thus ignores or at least underrates theplasticity involved in all aspects of speech production. Withthis last paragraph lam also implying that free, non-monitoredtexts will hardly be as regular as the anes investigated here.That is not to say that it would not be an interesting andworthy object of study - quite the contrary.

V. NOTE

1. lam aware that Ladd (1983a and b) sees It Hart and Col-lier's description of Dutch as belonging within the tonal

sequence theory of intonation, and GårdingIs and Bruce'sSwedish model as a compromise model between the hierarchicaland linear theories. I disagree with Laddls interpretation.This controversy is, however, not central to my present pur-pose, and will not be dealt with any further here.

ACKNOWLEDGEMENT

My thanks are due to Jørgen Rischel for valuable comments.Jeanette Holtse competently performed the measuring and graphicwork.

REFERENCES

Bruce, G. 1977: "Swedish word accents in sentence perspective",TPavaux de l'Institut de Linguistique de Lund 12, p. 1-155

Bruce, G. 1982: "Textual aspects of prosody in Swedish",Phonetiaa 39, p. 274-287

Cooper, W.E. and Sorensen, J.M. 1981: Fundamental Frequenayin Sentenae Produation, (New York: Springer Verlag)

Fujisaki, H., Hirose, K. and Ohta, K. 1979: "Acoustic featuresof the fundamental frequency contours of declarative sen-tences in Japanese", Ann. Bull. Res. Inst. Logopediasand Phoniatrias 13, p. 163-173

Gårding, E. 1983: "A generative model for intonation", in:A. Cutler and D.R. Ladd (eds.) Prosody - ModeZs andMeasUI'ements, (Berlin: Springer-Verlag), p. 11-25

't Hart, J. and Collier, R. 1979: "On the interaction'of ac-centuation and intonation in Dutch", in: E. Fischer-Jørgensen, J. Rischel and N. Thorsen (eds.) Proa. 9thInt. Congo Phonetia Saienaes, vol. II, (Copenhagen:Institute of Phonetics, Copenhagen University), p. 395-402

I ~~~-~~

r . -


Hombert, J.-M. 1977: "Oevelopment of tones from vowelheight?", J. Phonetics 5, p. 9-16

Jeel, V. 1975: l'An investigation of the fundamental frequencyof vowels af ter various consonants, in particular stopconsonants", Ann. Rep. Inst. Phon. Univ. Copenhagen 9,p. 191-211

Ladd, O.R. 1983a: "Peak features and overall slope", in:A. Cutler and O.R. Ladd (eds.) prosody - Models and

Measurements, (Berlin: Springer-Verlag), p. 39-52

Ladd, O.R. 1983b: "Phonological features of intonationalpeaks" , Language 59, p. 721-759

Lehiste, l. 1975: "The phonetic structure of paragraphs", in:A. Cohen and S.G. Nooteboom (eds.) Structure and Processin Speech Perception, (Berlin: Springer-Verlag) , p. 195-206

Lehiste, I. 1979: "Perception of sentence and paragraph bound-aries", in: B. Lindblom and S. Ohman (eds.) Frontiers ofSpeech Communiaation Research, (London: Academic Press) ,p.191-201

Lehiste, I. 1980: "Phonetic characteristics of discourse" ,Ac. Soc. Japan, Transactions of the committee on speechresearch, April 198O, p. 25-38

Lehiste, I. and Wang, W. S.-Y. 1977: "Perception of sentenceboundaries with and without semantic information", in:W.U. Dressler and O.E. Pfeiffer (eds.) Phonologica 1976,Innsbruaker Beitrage zur SprachlJissenschaft (Institutfur Sprachwissenschaft der Universitat lnnsbruck), p. 277-283

Liberman, A.M., Cooper, F.S., Shankweiler, D.S., and Studdert-Kennedy, M. 1967: "Perception of the speech code",psyahological RevielJ 74, p. 431-461

Liberman, M. and Pierrehumbert, J. 1983: 'IIntonational in-variance under changes in pitch range and length", in:M. Aronoff and R. Oehrle (eds.) Language Sound Structure,(Cambridge: MIT Press)

Nakatani, L. 1975: "Sentence prosody in' isolated and in storycontexts'l, J. Acoust. Boa. Am., Supplement 1, p. 120

Petersen, N. Reinholt 1974: "The influence of vowel height onthe perception of vowel duration in Oanish", Ann. Rep.Inst. Phon. Univ. G'openhagen 8, p. 1-10

Pierrehumbert, J.B. 1980: The Phonology and Phonetics ofEnglish Intonation, M.I.T. Doctoral dissertation

.. - -.'..

/~~

242 THORSEN

Pierrehumbert, J.B. 1983: "linguistic units for Fo synthesis",AbstY'acts of th~ 10th ~nt. ,ContJ. Phonetic Sciences,(Dordrecht: Forls Publlcatlons) , p. 137-144

Thorsen, N. 1978: "An acoustical investigation of Danish in-tonation", J. Phonetics 6, p. 151-175

Thorsen, N. 1979: "Interpreting raw fundamental frequencytracings of Danish", Phonetica 56, p. 57-78

Thorsen, N. 1980a: "Word boundaries and Fo patterns in Ad-vanced Standard Copenhagen Danish", Phonetica 5?, p. 121-133

Thorsen, N. 1980b: "Neutral stress, emphatic stress, and sen-tence intonation in Advanced Standard Copenhagen Danish" ,Ann. Rep. Inst. Phon. Univ. Copenhagen 14, p. 121-205

Thorsen, N. 1980c:"A study of the perception of sentence in-tonation - Evidence from Danish", J. Acoust. Soc. Am. 6?,p. 1014-1030

Thorsen, N. 1983a: "Standard Danish sentence intonation -Phonetic data and their representation", Folia Linguistical?, p. 187-220

Thorsen, N. 1983b: "Two issues in the prosody of StandardDanish", in: A. Cutler and D.R. ladd (eds.) PY'osody -Models and MeasuY'ements, (Berlin: Springer-Verlag),p. 27-38

Thorsen, N. 1984: "Variability and invariance in Danishstress group patterns", Ann. Rep. Inst. Phon. Univ.Copenhagen 18, p. 13-43. To appear in Phonetiaa 41

Uyeno, T., Hayashibe, H. and Imai, K. 1979: "On pitch con-tours of declarative, complex sentences in Japanese",Ann. Bull. Res. Inst. Logopedics and PhoniatY'ias 13,p. 175-187

Uyeno, T., Hayashibe, H. and Imai, K. 1981: "Syntactic struc-tures and prosody in Japanese: A study on pitch contoursand the pauses at phrase boundaries", Ann. Bull. Res.Inst. Logopedias and PhoniatY'ics 15, p. 91-108

Vaissiere, J. 1983: "language-independent prosodic features",in: A. Cutler and D.R. ladd (eds.) PY'osody - Models andMeaBuY'ements, (Berlin: Springer-Verlag), p. 53-b6.

cg

l

danpass.hum.ku.dk › ng › papers › aripuc18_1984_185-242.pdf · . aripuc 18, 1984, p. 185-242...

Documents