Categorical and gradient effects of information structure on nuclear - - PowerPoint PPT Presentation

Categorical and gradient effects of information structure on nuclear prominence in American English

Eleanor Chodroff, Alaina Arthurs, Priya Kurian, Jonah Pazol, and Jennifer Cole

Northwestern University, Department of Linguistics

f0 duration intensity semantics syntax pragmatics

PROSODY

emotion intention affect Pitch accent

pragmatics

PROSODY

phrase position

[JOHN went to an AWEsome PARty]IP

Nuclear accent Prenuclear accent

Information structure

NUCLEAR ACCENTS

Information structure: relation between information in a sentence and the knowledge state of the participants in the discourse

Chafe 1974, Büring 2007

Given L* / unaccented

Early work: relatively binary relation between information structure and pitch accent type

Pierrehumbert & Hirschberg 1990, Halliday 1967a, Brazil et al 1980

New / Focus H* / rising

Given L* / unaccented New / Focus H* / rising

Empirical studies have revealed a fairly probabilistic relation between information structure and pitch accent type

English: Terken & Hirschberg 1994, Bard & Aylett 1999, Ito et al. 2004 German: Röhr & Baumann 2010, de Ruiter 2015

Information structure

NUCLEAR ACCENTS

f0 duration intensity

Information structure may also be expressed in the phonetic realization of nuclear accents

American English: Effects of focus location and type on duration, intensity, and f0

Breen et al. 2010

Australian English: Lower f0 peaks on given information relative to new information

Calhoun 2012

German: Gradient effects of givenness on f0

Röhr & Baumann 2010

Mandarin: Effects of givenness on duration, intensity, and f0

Ouyang & Kaiser 2011

Information structure

NUCLEAR ACCENTS

Affect Pitch accent type Acoustic-phonetic prominence GIVEN ACCESSIBLE NEW CONTRASTIVE NEUTRAL LIVELY

duration | intensity | voice quality

AMERICAN ENGLISH

H* | L+H* | L*+H | L*

Information structure

Contrastive focus New Accessible Given

Pitch accent type

L+H* H* L*+H L* unaccented

Acoustic prominence

Duration Intensity Modal voice

1) INTRODUCTION 2) METHODS 3) MEASURES 4) RESULTS 5) DISCUSSION

32 native speakers of American English 23 female, 9 male 20 sets of three sentence mini-stories Manipulated information structure (IS) of final object noun in target sentence

METHODS

! ! ́ ! !́ ! !́!! det N V det N

Semantic neutrality throughout the story Third sentence = target sentence: Avoid voiceless obstruents Maintain consistent metrical and syntactic structure

METHODS

Given She knew it would take hours to make the marmalade. Accessible She especially enjoyed making homemade preserves. New She likes to make everything from scratch. Contrastive Our father loved the strawberry jam. Context sentence 2: Context sentence 1: Our sister Jamie spent all day Saturday in the kitchen. Target sentence: Our nana loved the marmalade.

Given She knew it would take hours to make the marmalade. Accessible She especially enjoyed making homemade preserves. New She likes to make everything from scratch. Contrastive Our father loved the strawberry jam. Context sentence 2: Context sentence 1: Our sister Jamie spent all day Saturday in the kitchen. Target sentence: Our nana loved the marmalade.

Given She knew it would take hours to make the marmalade. Accessible She especially enjoyed making homemade preserves. New She likes to make everything from scratch. Contrastive Our father loved the strawberry jam. Context sentence 2: Context sentence 1: Our sister Jamie spent all day Saturday in the kitchen. Target sentence: Our nana loved the marmalade.

Given She knew it would take hours to make the marmalade. Accessible She especially enjoyed making homemade preserves. New She likes to make everything from scratch. Contrastive Our father loved the strawberry jam. Context sentence 2: Context sentence 1: Our sister Jamie spent all day Saturday in the kitchen. Target sentence: Our nana loved the marmalade.

Given She knew it would take hours to make the marmalade. Accessible She especially enjoyed making homemade preserves. New She likes to make everything from scratch. Contrastive Our father loved the strawberry jam. Context sentence 2: Context sentence 1: Our sister Jamie spent all day Saturday in the kitchen. Target sentence: Our nana loved the marmalade.

20 unique story-IS pairings per participant, counterbalanced across participants (multiple of 4) Order of stories randomized 4 blocks Affect manipulation: neutral/casual vs. lively/expressive

METHODS

Comprehension question following each story Yes-No Q Targeted IS manipulation in second sentence Median accuracy: 97% Min accuracy: 85%

METHODS

1) INTRODUCTION 2) METHODS 3) MEASURES 4) RESULTS 5) DISCUSSION

Pitch accent category Duration of initial trochee (sec) Intensity of initial trochee (dB) Voice quality (modal/creaky voice sequence)

MEASURES

H* L+H* L*+H L* unaccented

MEASURES Pitch accent category

H* L*/unaccented H* L+H* L*+H L* unaccented

MEASURES Pitch accent category

MEASURES Duration and intensity

Force aligned speech (FAVE aligner) Manually corrected critical word boundaries Re-ran aligner on corrected final word Obtained duration and intensity of initial trochee

MEASURES Voice quality

Manually labelled intervals of modal and creaky voice throughout critical word Analysed simple sequence of voice quality type (none to one transition)

1) INTRODUCTION 2) METHODS 3) MEASURES 4) RESULTS 5) DISCUSSION

32 participants × 80 productions = 2560 tokens 218 productions were excluded Median of 6 errors per participant TOTAL: 2294 tokens

RESULTS

Pitch accent

Logistic mixed model: condition * affect

Duration & Intensity

Linear mixed model: condition * affect * pitch accent type

Voice quality

Multinomial logistic model: condition * affect * pitch accent type

RESULTS

47 14 103 29 121 35 153 45 250 266 187 247 172 243 143 239

H* L*/UA g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e 100 200

count affect

neutral lively

PITCH ACCENT

47 14 103 29 121 35 153 45 250 266 187 247 172 243 143 239

H* L*/UA g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e 100 200

count affect

neutral lively

PITCH ACCENT

H* L*/UA g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e 200 300 400 500 600

duration (ms)

DURATION

H* L*/UA g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e 200 300 400 500 600

duration (ms)

DURATION

H* L*/UA g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e 40 50 60 70 80

intensity (dB)

INTENSITY

H* L*/UA g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e 40 50 60 70 80

intensity (dB)

INTENSITY

H* L*/UA g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e g i v e n a c c e s s i b l e n e w c

• n

t r a s t i v e 40 50 60 70 80

intensity (dB)

INTENSITY

5 53 3 9 91 26 10 111 28 11 146 34 234 218 56 134 236 54 127 221 58 136 193 41

H* L*/UA given accessible new contrastive given accessible new contrastive 50 100 150 200

count label

fully modal modal then creaky fully creaky

VOICE QUALITY

Creak Sequence Female Male

Fully modal 13% 15% Modal then creaky 57% 57% Fully creaky 30% 28%

VOICE QUALITY

1) INTRODUCTION 2) METHODS 3) MEASURES 4) RESULTS 5) DISCUSSION

Highly probabilistic relation between information structure and prosodic realization of pitch accents Given

H*

Accessible New Contrastive

L* / UA

Highly probabilistic relation between information structure and correlates of phonetic prominence Given

Duration

Accessible New Contrastive

Intensity Voice quality

Given information has greater influence on prosodic realization relative to new information, particularly in phonetic realization

See also Schwarzschild 1999

Findings in line with idea that aspects of information structure license prosodic enhancement / reduction but do not require it

Schwarzschild 1999

Probabilistic relation in production has important implications for the perceptual processing of prosody-IS relations

e.g., Fowler & Housum 1987, Rump & Collier 1996, Breen et al 2010, Roettger & Cole 2019

• Separate nuclear position from edge of intonational phrase

My nana loved the marmalade she made.

• Investigate relation between information structure and prenuclear pitch

accents

• Examine IS-prosody relations in additional languages (e.g. German,

Spanish)

• Determine how production data relates to listeners’ perceptual

knowledge of these relations

FUTURE DIRECTIONS

Thank you!

Many thanks to:

Katherine Glew Stefan Baumann Jane Mertens Timo Roettger Dan Turner Maria Gavino

BANDAGES BARRIERS BOULEVARD GOVERNOR LABRADOR LAMINATE LAVENDER LEMONADE LIMERICK LULLABY MAGAZINES MARIGOLDS MARMALADE MELODY MEMORY MONITOR MONUMENT NEIGHBORHOOD NOVELTIES ROBBERY

Phonological and phonetic encoding of information structure in nuclear position. Pitch accent differences: relative to the average production, H* is … v Less likely on given information v More likely on new information v More likely on contrastive information v More likely when lively

Phonological and phonetic encoding of information structure in nuclear position. Phonetic differences: relative to the average production… v Given information was significantly shorter and less intense v Contrastive information was significantly more intense v Lively information was significantly longer and more intense