pitch accent on discourse marker and discourse construction kiwako ito & ross metusalem
TRANSCRIPT
Discourse Markers
• Discourse Markers (DMs):
words or short phrases that set relations between prior discourse and current utterancese.g., now, well, anyway, next, however, by
the way, in any case, etc.
“Cue phrases … directly signal the structure of discourse” (Hirschberg & Litman, 1993)
Intonational variation for DMsProsody differentiates the uses of multi-
functional DMs
F: It was at one time all: almost all Jewish.Now it’s I would say si-
J: sixty Jewish, forty Italian.
(Schiffrin, 1987: p231)
Intonational variation for DMsa. They aren’t brought up the same way.b. Now Italian people are very outgoing.c. They are very generous.d. When they put a meal on the table it’s a
meal.e. Now these boys were Irish.f. They lived different.
(Schiffrin, 1987: p231)
Intonation and Meaning
• Intonational contours used to indicate pragmatic meaning (Pierrehumbert & Hirschberg, 1990)
– e.g., H* is ‘new’, L-H% is ‘continuation rise’
• L+H* is commonly used to ‘mark a correction or contrast’ (p. 296)
– e.g., ‘Don’t hand me the blue pen. Hand me the RED one.’
Intonation and Online Discourse Processing
• Intonation affects comprehension– L+H* used felicitously aids sentence
comprehension (Bock & Mazella, 1983)
e.g., ARNOLD/Arnold didn’t FIX/fix the radio. DORIS fixed the radio’ (p. 66)
• L+H* can lead to anticipatory eye movements (Ito and Speer, in press)
e.g., First, hang the green ball. Now, hang the BLUE ball’ (p. 11)
L+H* and DMs
Ito & Speer cont’d• Hang the blue ball.
And THEN/then, hang the GREEN ball.
– No anticipatory eye movements, but faster decline from target when L+H* on DM
– L+H* on DM may lead to anticipation of contrast
Research Question
Does L+H* on a DM lead to expectation of contrast between preceding utterance and upcoming utterance?
If yes, is the effect global or local?
Is accentual property of a DM interpreted a/c prosodic structure of prior context?
The Experiment• Discourse completion task
– subjects listen to short stories and provide an appropriate continuation
– accentual pattern of stimuli varied to test effect of L+H* in prior discourse and DM
• Hypothesis: L+H* on DM evokes a contrast between preceding utterance and upcoming utterance– Prosodic and informational structure of responses
should be more predictable from preceding sentence when L+H* on DM versus H* on DM
MaterialsThree-part stimulus
– Context: two people in naturalistic situation; ‘Collaborative’ and ‘Companionship’ contexts
‘This spring, Mary and Adam finally started gardening.’
– Prompt: one person from Context engages in an action; strict SVO structure
‘Early on, Mary planted basil.’
– DM: 4 temporal DMS prompt sequential event; test effect of DM accentual pattern
‘And then,’ ‘And next,’ ‘After that,’ ‘Following that’
Materials (cont’d)
3 prosodic variations of Prompt1. ‘Mary planted basil.’ H* !H* L-L%
2. ‘MARY planted basil.’ L+H* L-L% H* L-L%
3. ‘Mary planted BASIL.’ H* L+H* L-L%
Prompt 1
MaryplantedbasilH* !H*!H*L–L%
75
300
100150200250
Time (s)1 2.4171MaryplantedbasilL+H*L–L%H* L–L%
75
300
100150200250
Time (s)1 2.29982
MaryplantedbasilH* L+H*L–L%
75
300
100150200250
Time (s)1 2.3483and thenH* H* L–H%
75
300
100150200250
Time (s)0 0.881633
and thenH* L+H*L–H%
75
300
100150200250
Time (s)0 0.907302
PS1 PS2
PS3
DM1
DM2
Prompt 2
MaryplantedbasilH* !H*!H*L–L%
75
300
100150200250
Time (s)1 2.4171MaryplantedbasilL+H*L–L%H* L–L%
75
300
100150200250
Time (s)1 2.29982
MaryplantedbasilH* L+H*L–L%
75
300
100150200250
Time (s)1 2.3483and thenH* H* L–H%
75
300
100150200250
Time (s)0 0.881633
and thenH* L+H*L–H%
75
300
100150200250
Time (s)0 0.907302
PS1 PS2
PS3
DM1
DM2
Prompt 3
MaryplantedbasilH* !H*!H*L–L%
75
300
100150200250
Time (s)1 2.4171MaryplantedbasilL+H*L–L%H* L–L%
75
300
100150200250
Time (s)1 2.29982
MaryplantedbasilH* L+H*L–L%
75
300
100150200250
Time (s)1 2.3483and thenH* H* L–H%
75
300
100150200250
Time (s)0 0.881633
and thenH* L+H*L–H%
75
300
100150200250
Time (s)0 0.907302
PS1 PS2
PS3
DM1
DM2
Average Duration and f0 Value of Prompts
Subject Object Prompt # & Accentual
Type dur
(ms) f0 (Hz) dur
(ms) f0 (Hz)
Pro mpt 1 H* !H*
332 201 537 162
Pro mpt 2 L+H*L-L% H*
352 219 543 173
Pro mpt 3 H* L+H*
319 203 607 198
DM 1
MaryplantedbasilH* !H*!H*L–L%
75
300
100150200250
Time (s)1 2.4171MaryplantedbasilL+H*L–L%H* L–L%
75
300
100150200250
Time (s)1 2.29982
MaryplantedbasilH* L+H*L–L%
75
300
100150200250
Time (s)1 2.3483and thenH* H* L–H%
75
300
100150200250
Time (s)0 0.881633
and thenH* L+H*L–H%
75
300
100150200250
Time (s)0 0.907302
PS1 PS2
PS3
DM1
DM2
DM 2
MaryplantedbasilH* !H*!H*L–L%
75
300
100150200250
Time (s)1 2.4171MaryplantedbasilL+H*L–L%H* L–L%
75
300
100150200250
Time (s)1 2.29982
MaryplantedbasilH* L+H*L–L%
75
300
100150200250
Time (s)1 2.3483and thenH* H* L–H%
75
300
100150200250
Time (s)0 0.881633
and thenH* L+H*L–H%
75
300
100150200250
Time (s)0 0.907302
PS1 PS2
PS3
DM1
DM2
Average Duration and f0 Value of DMs
1st word (e.g., ‘and’)
2nd word (e.g., ‘t hen’) DM
Accent dur (ms) f0 (Hz) dur (ms) f0 (Hz)
L+H* 381 193 593 206 H* 395 209 499 176
Experimental Conditions DM1 (L+H*) DM2 (H*)
Prompt1 (H* !H*) C1 C2 Prompt 2 (L+H*L-L% H*) C3 C4 Prompt 3 (H* L+H*) C5 C6
Condition Prompt DM
C1 ‘Early on, Mary planted basil.’ ‘And NEXT…’
C2 ‘Early on, Mary planted basil.’ ‘And next…’
C3 ‘Early on, MARY planted basil.’ ‘And NEXT…’
C4 ‘Early on, MARY planted basil.’ ‘And next…’
C5 ‘Early on, Mary planted BASIL.’ ‘And NEXT…’
C6 ‘Early on, Mary planted BASIL.’ ‘And next…’
Conditions 1-6
Examples of Conditions 1-6
Experimental Setup
• 48 target trials (8 per condition)
• 48 filler trials– intransitives and datives exhibiting L+H* on
subject, verb, or direct/indirect object– included ‘however’ as DM
• 6 lists, 3 blocks per list
Procedure
• Participants (25 total) seated in soundproof booth and presented stimuli through Eprime
• Continuations recorded in Praat V4.5.15
Data Analysis
• Each continuation coded for status as a ‘parallel’ or ‘non-parallel’ continuation
– ‘parallel’ involves syntactic/thematic structures and discourse purpose
– coding gives view of direct contrast within discourse context
Parallel Continuation
• Syntactic structure – SVO
• Thematic structure– agent - transitive verb - patient
• Discourse purpose– Contributes to topic/goal of discourse in a
way parallel to the Prompt
Parallel Continuation (cont’d)This spring, Mary and Adam finally started gardening.
Early on, Mary planted basil. And then…
she planted oregano. Adam planted tomatoes.
Before heading into the movie theater, Jenna and Wally stopped at the concessions stand.
Considering many options, Wally chose popcorn. And then…
Jenna bought Sour Patch Kids.
Parallel Continuation (cont’d)Before choosing their new home, Drew and Nora toured many
houses. In the first house, Drew explored the kitchen. After that…
Nora checked out the bathroom.
Non-Parallel Continuation• Syntactic/Thematic violation
After setting up their tent, Gary and Laurie started the BBQ. Before anything else, Laurie seasoned the meat. And next…
she put the meat on the grill.Gary turned the barbeque on.
To renovate the kitchen, both Arnold and Molly spent a lot of money.
Initially, Molly replaced the cabinets. Following that…
Arnold put the new tile on the floor.
Non-Parallel Continuation (cont’d)
• Discourse Purpose violationThis spring, Mary and Adam finally started gardening.
Early on, Mary planted basil. And then…
Adam uprooted the basil.
Before hanging the new curtains, Lara and Brian decided to clean the living room.
First, Lara opened the window. After that…
Brian threw the lamp out.
Ambiguous Cases
• Some continuations could not be classified as parallel or non-parallel
When the power went down, Julie and Ben were cooking dinner.
Unable to see, Julie dropped a plate.
And next…
Ben lit a candle. discourse purpose?
Ambiguous Cases
For the first time in their lives, Donna and Bill stayed in a five-star resort hotel.
Soon after lunch, Donna visited spa.
Following that…
Bill got a massage.
Information StatusSubject Verb Post -verba l Argument
RetN : lexica lly ident ica l to Pro mpt sub ject
RetV: lexica lly ident ica l to Pro mpt ve rb
Ret A: lexica lly ident ica l to Pro mpt arg ument
RetP : pron oun of ParN ( i.e., ‘he’ or ‘she ’)
RetR : synony m to Pro mpt Verb (e.g ., picked ~ chose)
RetP : pron oun of ParA (i.e., ‘it’)
ContN: lexica lly ident ica l to pers on from Conte xt not ment ioned in Prompt
ContV: co ntrast ive act ion to Pro mpt verb
ContA: inferab le argu ment con trast ive to Pro mpt argu ment
ContP: prono un of ContN InfV: inferable but not contrast ive to Pro mpt verb
InfA: inferab le but no t contrast ive to Pro mpt argu ment
DP: pronoun of bo th peop le in Context (i.e., ‘they’)
RepV : act ion repea ted from Co ntext
Rep A/P: noun /pron oun of item from Conte xt
BN: brand new sub ject not ment ioned in discourse
NewV : non -inferab le act ion ne w to discour se
New A: non -inferab le concept ne w to disco urse
ShiftN: Pr ompt’s ob ject beco mes subject
ShiftA/P: no un/pronoun of Pro mpt sub ject
InfN: an y item in Conte xt other t han ParN or ContN
• Tags motivated by Ch. 2 ‘A theory of discourse coherence’ in Coherence, Reference and the Theory of Grammar by Andrew Kehler (2002)
Example Transcription
she planted tomatoes
PN V N
subj main V DO
ParP ParV ContA
L+H*
0.52
–0.61
5000
0
0 1.67
200
75
FundamentalFrequency
(Hz)
SpectralFrequency
(Hz)
Time (s)
Amplitude(Pa)
Predictions
Comparison Pred ictions C1 versus C2 (no L+H*)
• More subject and argument contrast in 1C than 2C • Parallel subj /ectargument pairs with contrastive argum /entsubject more often in C1 than 2C
3 4C versus C (subject + *)L H
• More contrastive subjects in C3 thanC4 • More parallel arguments in 3C than 4C
5 6C versus C (object + *)L H
• More contrastive arguments in 5C than 6C • More parallel subjects in 5C than 6C
• Informational focus should be more predictable from Prompt when L+H* on DM than H*on DM
DM Accent and Parallel Continuation
• 573 continuations analyzed– 203 parallel (35%)
• L+H* on DM did not induce parallel continuation more than H*
DM# of Parallel
Continuations
L+H*
And NEXT104
H*
And next99
TOTAL 203
Experimental Conditions and Parallel Continuation
• L+H* on DM effect hinted at only when object of Prompt had L+H* (C5 & 6)
Accent Pattern Prompt ‡ DM
# of Parallel Continuations
C1 Mary planted basil. And THEN… 30
C2 Mary planted basil. And then… 30
C3 MARY planted basil. And THEN… 33
C4 MARY planted basil. And then…
39
C5 Mary planted BASIL. And THEN… 41
C6 Mary planted BASIL. And then… 30
TOTAL 203
Experimental Conditions and Parallel Continuation
• L+H* on DM effect hinted at only when object of Prompt had L+H* (C5 & 6)
• Prompt 1 (C1&C2 = 60)• Prompt 2 (C3&C4 = 72)• Prompt 3 (C5&C6 = 71)
Parallel continuations appeared more often when Prompt had L+H*
Accent Pattern Prompt ‡ DM
# of Parallel Continuations
C1 Mary planted basil. And THEN… 30
C2 Mary planted basil. And then… 30
C3 MARY planted basil. And THEN… 33
C4 MARY planted basil. And then…
39
C5 Mary planted BASIL. And THEN… 41
C6 Mary planted BASIL. And then… 30
TOTAL 203
Information Structure of Continuation Types
• Parallel continuations exhibit mainly contrastive subjects, parallel verbs, and contrastive arguments
Continuation Type
Contrastive Subject
Retained Subject
Contrastive Verb
Retained Verb
Contrastive Argument
Retained Argument
Parallel (203 total)
125 (62%)
65 (32%)
25 (12%)
160 (79%)
171 (84%)
24 (12%)
Non-parallel (333 total)
94 (28%)
117 (35%)
35 (11%)
15 (5%)
16 (5%)
35 (11%)
Ambiguous (37 total)
20 (54%)
16 (43%)
14 (38%)
7 (19%)
13 (35%)
9 (24%)
Information Status Distribution in Parallel Continuations
• No clear effect of DM accent in Prompt 1 (no L+H*)
• Patterns emerge for other Prompt types
Condition Contrast ive Sub ject
Reta ine d Sub ject
Contrast ive Verb
Reta ine d Verb
Contrast ive Argument
Reta ined Argumen t
C1 Mary plan ted
bas il. And THEN…
47% 37% 17% 80% 87% 13%
C2 Mary plan ted
bas il. And then …
70% 23% 17% 83% 87% 10%
C3 MARY
planted bas il. And THEN…
82% 18% 12% 82% 76% 18%
C4 MARY
planted bas il. And then …
56% 31% 15% 82% 87% 13%
C5 Mary plan ted BAS IL.
And THEN…
51% 46% 7% 80% 76% 15%
C6 Mary plan ted BAS IL.
And then …
67% 33% 7% 77% 97% 3%
Subject Prominence (C3 & 4)
• Contrastive subject more often when DM has L+H*
Condition Contrast ive Sub ject
Reta ine d Sub ject
Contrast ive Verb
Reta ine d Verb
Contrast ive Argument
Reta ined Argumen t
C1 Mary plan ted
bas il. And THEN…
47% 37% 17% 80% 87% 13%
C2 Mary plan ted
bas il. And then …
70% 23% 17% 83% 87% 10%
C3 MARY
planted bas il. And THEN…
82% 18% 12% 82% 76% 18%
C4 MARY
planted bas il. And then …
56% 31% 15% 82% 87% 13%
C5 Mary plan ted BAS IL.
And THEN…
51% 46% 7% 80% 76% 15%
C6 Mary plan ted BAS IL.
And then …
67% 33% 7% 77% 97% 3%
Subject Prominence (C3 & 4)
• Contrastive subject more often when
DM had L+H*
• Subject retained more often when DM did not have L+H*
Condition Contrast ive Sub ject
Reta ine d Sub ject
Contrast ive Verb
Reta ine d Verb
Contrast ive Argument
Reta ined Argumen t
C1 Mary plan ted
bas il. And THEN…
47% 37% 17% 80% 87% 13%
C2 Mary plan ted
bas il. And then…
70% 23% 17% 83% 87% 10%
C3 MARY
planted bas il. And THEN…
82% 18% 12% 82% 76% 18%
C4 MARY
planted bas il. And then…
56% 31% 15% 82% 87% 13%
C5 Mary plan ted BAS IL.
And THEN…
51% 46% 7% 80% 76% 15%
C6 Mary plan ted BAS IL.
And then…
67% 33% 7% 77% 97% 3%
Object Prominence (C5 & 6)
• More contrastive arguments when DM did not have L+H*
Condition Contrast ive Sub ject
Reta ine d Sub ject
Contrast ive Verb
Reta ine d Verb
Contrast ive Argument
Reta ined Argumen t
C1 Mary plan ted
bas il. And THEN…
47% 37% 17% 80% 87% 13%
C2 Mary plan ted
bas il. And then…
70% 23% 17% 83% 87% 10%
C3 MARY
planted bas il. And THEN…
82% 18% 12% 82% 76% 18%
C4 MARY
planted bas il. And then…
56% 31% 15% 82% 87% 13%
C5 Mary plan ted BAS IL.
And THEN…
51% 46% 7% 80% 76% 15%
C6 Mary plan ted BAS IL.
And then…
67% 33% 7% 77% 97% 3%
Object Prominence (C5 & 6)
• More contrastive arguments when DM did not have L+H*
• More retained arguments when DM had L+H*
Condition Contrast ive Sub ject
Reta ine d Sub ject
Contrast ive Verb
Reta ine d Verb
Contrast ive Argument
Reta ined Argumen t
C1 Mary plan ted
bas il. And THEN…
47% 37% 17% 80% 87% 13%
C2 Mary plan ted
bas il. And then…
70% 23% 17% 83% 87% 10%
C3 MARY
planted bas il. And THEN…
82% 18% 12% 82% 76% 18%
C4 MARY
planted bas il. And then…
56% 31% 15% 82% 87% 13%
C5 Mary plan ted BAS IL.
And THEN…
51% 46% 7% 80% 76% 15%
C6 Mary plan ted BAS IL.
And then…
67% 33% 7% 77% 97% 3%
Interesting Findings
• Prominent Subject (C3 & 4)– L+H* on DM led to more contrastive
subjects and less retained subjects
aligns with predictions
• Prominent Object (C5 & 6)– L+H* on DM led to less contrastive
arguments and more retained arguments opposite of predictions
Three Possibilities
(1) L+H* on DM reinforces contrast in subject position but blocks contrast in object position
Subject L+H*
Object L+H*
Subject
Object
Three Possibilities(2) L+H* in Prompt lead to different
expectations a/c the accent location
DM reinforces the appropriate expectation
Subject
Object
Subject
Object L+H*
Three Possibilities(3) L+H* in Prompt leads to different
expectations due to FOCUS PROJECTION
DM highlights the optional broader focus?
Subject
Object
Subject
Object L+H*
VP (Object)L+H*
Contrast Frequency: Subject vs. Object
• Overall, data exhibited more subject contrast than object contrast
– Prosodically highlighted subject evokes alternative agent from Context salient
– Prosodically highlighted object evokes set of possible alternatives less salient
Cross-Subject Variability
• Continuation strategies varied widely between subjects– Parallel continuations: 12 to 29– Contrastive subjects: 10 to 41– Parallel verbs: 8 to 24– Contrastive arguments: 11 to 26
Stimuli Problems• Some items exhibit bias for contrast due to
salience of contrastive entities
Following dinner, Al and Gail stopped at the ice cream shop.
After waiting in line, Al ordered vanilla. After that…
– many salient contrasts with ‘vanilla’– semantically biased to parallel continuation
Stimuli Problems (cont’d)
• Some items exhibit bias against contrast
With the tornado siren sounding, Rose and Greg prepared to take cover.
In a hurry, Greg entered the basement. And then…
– few, if any, salient contrasts with ‘basement’
– Parallel continuation mainly limited to ‘Rose entered the basement.’
Stimuli Problems (cont’d)
• Some items did exhibit appropriate salience of contrastive entities
Before selling their old Civic, Dewey and Anna took a whole day to clean it.
When they were nearly finished, Anna wiped the dashboard. And then…
– several salient contrasts with ‘dashboard’– not biased toward parallel continuation
Future Directions
• More subjects will be analyzed to confirm patterns presented here
• ToBI transcription and f0 analysis• Study to be conducted again with more
carefully controlled stimuli– Perhaps present both alternative subjects and
objects mentioned in Context
• Eye tracking to test effect of L+H* in prior discourse and on DM