The form and function of
extra-sentential elements
Lluïsa Astruc-Aguilera
Francis Nolan
University of Cambridge
Extra-sentential elements:

vocatives


appositions


‘“Your meal is ready”, I said to Anna’
dislocations


‘He’s won the lottery, the lucky bastard’
direct speech markers


‘This is John, my brother’
epithets


‘Your meal is ready, Mary’
‘They’re crazy, those Romans’
non-restrictive relative sentences

‘Anna’s friends, who were loyal, supported her’
among others …
Intonational phrasing

‘At present, the principles governing intonational
phrasing are not well understood. Certain syntactic
constructions – vocatives, appositions,
parentheticals, preposed clauses, nonrestrictive
relative clauses – are necessarily set off in separate
IPs’. (Selkirk 1995: 567)
(Pierrehumbert 1980, Bonet 1984, Nespor & Vogel 1986)
Research questions

Extra-sentential elements (ESEs)

What are their characteristics?

Do they constitute a homogeneous group?

Do they always form independent tonal units?

Are they accented or not?
Experimental design: 3 areas

General comparative study of the main type of ESEs
in the two languages

Quantitative studies to find evidence that some
ESEs are completely deaccented

Quantitative study to test the hypothesis that English
and Catalan use different estrategies for marking
ESEs prosodically
Objective


Estudying ESEs in English and Catalan

Languages with different degrees of syntactic and
intonational flexibility
 Catalan is more flexible syntactically
 English is more flexible intonationally

If ESEs in English and Catalan show intonational
differences
 There would no support for the view that the prosodic
form of ESEs follows from their syntactic form
5 production experiments
Experiment 1

General comparative study of the main types of
ESEs in the two languages

Methodology: ESEs in different positions in the
sentence, in sentences with identical meaning in
English and Catalan:



‘L’Anna va guanyar-la, Manu’
‘Anna won it, Manu’
Corpus: 605 phrases, read by 12 people
Apposition: ‘John’

Anna and her brother meet Alma and Anna says
‘This is my brother, John’
20
reduplication
12
6
-2
This is
my
bro
ther,
0
John
1.02531
Time (s)
12
Epithet: ‘the lucky bastard’
6
-2
wrong

again,
the but
cher
0
1.05633
‘He’s won
the lottery, the lucky
bastard’
Time (s)
20
H*
12
H*
deaccenting
6
-2
He's won the lo
tte
ry
the luc
0
ky
bas
tard
1.65878
Time (s)
Results
20
20

Reduplication
 at lower level
Appositions
12
12
6
6
Non-restrictive relatives
-2
-2
This is
wrong
0
again,
my
0

Deaccenting
Dislocations
Quotation markers
(English) vocatives
Sentential adverbs
ther,
John
cher
1.02531
Time 1.05633
(s)
Time (s)
(English) epithets
bro
the but
20
12
6
-2
He's won the lo tte
ry
the luc
0
ky
bas
tard
1.65878
Time (s)
Experiments 2, 3, 4

Quantitative study

Objective: finding empirical support for the
perceptual impression that most types of ESEs
do not receive pitch accents

Methodology:
One language
 One category of ESEs (right-dislocated phrases)
 2 different methods for uncovering potential pitch
accents



Lombard speech
Changes in metrical prominence
Lombard speech

Reported 1911 by audiologist Etienne Lombard


Speakers increase their voice level when the
ambient noise increases


Lombard discovered that applying an intense noise
to a patient’s ear would cause the patient to speak
louder
In noise, people speak louder
Auditory masking can be used to induce an
increase in voice level

Expansion of pitch range
Right-dislocated phrases

Research question: are such constructions
deaccented or do they receive very reduced pitch
accents?

Masking noise:

60 minutes of speech-shaped noise generated by computer
 Covers the range of frequencies of human speech
 More effective at lower volumes


Less dangerous
Task:

Reading aloud text (short dialogues) while hearing the
masking noise
Right-dislocated phrases

Material:
 7 phrases with identical form but different meanings
depending on intonation




‘She saw Anna, the bride’
Participants: 5 Central Catalan speakers
Method:
 Noise through open-ear headphones
 4 levels of noise, from 0 to 4
 Increased at controlled intervals (70dB, 75dB,
80dB)
Corpus:
 112 phrases
Results
0.04568
0
-0.05539
0
6.07238
30
Time (s)
1
2
3
4
pitch level
-12
0
6.07238
Time (s)
‘Vol la vela, la vella’ (‘She wants the sail, the old lady’)
Dislocation
30
vol
la
ve
la
la ve
lla
Apposition
-12
0
1.111
Time (s)
30
vol la
ve
la
la ve lla
-12
0
1.45837
Time (s)
Results


Deaccentuation: 79% of the cases
Independent phrasing: 70%

Deaccentuation appears to be a more robust cue
 At least with pre-planned, read speech

Observation: those speakers that read fast rely less
on phrasing and more on deaccentuation

Interpreted as indication that phrasing and
deaccentuation are strategies that complement each
other
Lombard speech

Caused an increase in voice level rather than
expansion of pitch range

Lombard speech is not entirely physiological



4 speakers showed the Lombard Effect
1 speaker, a teacher, said she wouldn’t speak loud
because of a sore throat
Lombard speech is a communicative strategy


It only appears when an audience is present or evoked
It is a speaking style
 Rephrasing: less pauses but longer, and on different
locations
 Changes in accentuation
Lombard speech: further work

Measuring pitch range of the rest of speakers

Further experiments

Monitoring loudness of response


Gradient or categorical?
Using masking noise for different frequencies
Experiment 3

Highly controlled material regarding pragmatic context and
experimental conditions
 Clear-cut results

Corpus: 324 phrases, 18 phrases mixed with ‘distractors’, read by 6
Central Catalan speakers (3 males and 3 females)

Methodology: words with 3 degrees of prosodic prominence
 Stress 0: Vilà
 Stress 1: Vila
 Stress 2: Vilamalla

Hypothesis: if syllables with a higher degree of prosodic prominence
(stress 1) receive a higher fundamental frequency than those with
secondary stress and no stress (stress 2 and 0), we can say that
the former get a pitch accent
Ja
li a
gra
da
Results
la lli mo na
da
0
0
1.94738
Time (s)

Right-dislocated phrases
form independent
intonational phrases
40
40
Jaenl’hi
a
li amen
Ja
ja
gra da
de lli
la lli
ma
ma
-12
-12 0
0
1.64106
Time (s)
1.83438
Time (s)

They are nearly always
deaccented

40
No statistically significant
difference (ANOVA) between
prominence levels
l’hiaa
Ja li
gra da
la lli mo na da
-12
0
1.94738
Time (s)
400
Experiment 4:
sentential adverbs
Some disagreement about the syntactic characteristics of the other
categories (especially vocatives and parenthesis)
 But most grammar textbooks agree about sentential adverbs
 Function: modifying the whole clause
verb phrase-modifiers
clause-modifiers
‘The car broke down, unfortunately’
‘The car broke down suddenly’
semantic scope
 Agreement about syntactic function
 Agreement about semantic function
Sentential advs





domain (‘politically’)
modality (‘possibly’)
evaluation (‘fortunately’)
speech act related (‘frankly’)
connective (‘moreover’)
‘The seminar was a bore, frankly’
Verb phrasal advs
 manner adv

degree
 frequency
 duration

…
‘She spoke quite frankly’
(The Cambridge Grammar of the English Language 2002, Hoye 1997, McCowley 1980)
Research questions

Independence of sentential adverbs from the main
phrase:



Rhythmically independent, i.e. separated by pauses or
lengthening?
Tonally independent, i.e. separated by tonal boundaries?
Accentuation:

Are they deaccented?
Methodology
Experimental material



Corpus: 760 phrases

22 sentential adverbs

Phrases with sentential and manner adverbs in
final position

Manner adverbs acting as control
Semantically equivalent in English and Catalan
Sounding natural


Oxford English Dictionary on-line as a searchable
corpus
Pragmatic conditions kept constant (broad focus: new
information, non-contrastive information)
The speakers
 Recorded 16 speakers
 Analyzed 10
5 Southern British English (Cambridge; 3 males, 2
females)
 5 Central Catalan (Reus; 3 males, 2 females)

Catalan and English
Prediction 1

‘Plastic’ languages (Dutch, English…) rely more on
prosody to mark information status

‘Non-plastic’ languages (Catalan, Spanish, Italian…)
rely more on word order
 Expected:
Less deaccenting of sentential adverbs in Catalan
than in English
(Vallduvi 1994, 1996; Steedman 2000; Swerts, Krahmer & Avesani 2002)
Sentential and manner adverbs
Prediction 2




‘I want to speak to you frankly’ (manner adv)
‘I don’t agree with you, frankly’ (sentential adv)
They have different syntactic function and different
semantic interpretation
Their semantic differences can be cued by intonation
 Expected:
Sentential and manner adverbs will receive different
intonation patterns
0
Intonation of sentential and
manner adverbs: English
-7
this
con
clu
ssion
fo
llows
i ne
vi
ta
bly
0
1.67161
Time (s)
manner adv
sentential
.2363
0.4342
0
0
.2679
-0.551
1.67161
0
20
0
Time (s)
1.58385
20
12
Time (s)
12
6
6
0
0
-7
-7
this
con
clu
sion
fo
llows
0
i
ne
vi
ta
bly
he
1.67161
Time (s)
‘inevitably’
en ded
up with no
mo
ney
0
i
ne
vi
ta
bly
1.58385
Time (s)
Intonation of sentential and
manner adverbs: Catalan
manner adverb
sentential
0.209
0.2816
0
0
0.3044
0
20
-0.5059
1.07931
0
20
Time (s)
12
12
6
6
0
0
-7
1.50806
Time (s)
-7
vi
uen
sim
0
ple
ment
i ai xòxo ésesel el que
qu
1.07931
Time (s)
‘They live simply’
va
pa
ssar
0
sim
ple
ment
1.50806
Time (s)
‘And this is what happened, simply’
Results:
Intonation
15%
English
50%

accented

> 1%
Catalan
accent
deaccented
19%


manner


sentential
Phrasing
English
phrasing
indep units
same unit


Catalan


manner


sentential
Results
Predictions 1 and 2

YES: sentential adverbs behave differently
from manner adverbs

YES: cross-linguistic differences

More deaccenting in English than in Catalan
Inter-speaker differences

Divergences in accentuation

Some speakers de-accent much more than the
rest
Reading styles: careful vs. casual
 Teachers


Homogeneity in phrasing across speakers,
both in Catalan and English
Conclusions
Prediction 1
yes
Prediction 2
yes
Less deaccenting of
sentential adverbs in
Catalan than in English
English (‘plastic’ lang) relies
more on prosody to mark
information status than
Catalan does (‘non-plastic’)
Sentential and manner
adverbs will receive
different intonation
patterns
Indirect relation of
syntax and prosody, via
semantics. Some
semantic diff are cued
by intonation.
Conclusions

ESEs are not a homogeneous group, neither syntactically nor
prosodically

Only sentential adverbs form independent intonational phrases
obligatorily

Sentential adverbs in Catalan and English show differences in
frequency of accentuation

Interaction between prosodic form and syntactic function of
ESEs to signal a specific communicative function

Existence of a level of information structure mediating between
syntax and phonology
The form and function of
extra-sentential elements
Many thanks!
Questions?
www.astruc.info
Descargar

Lombard Speech - Centres docents