A cross-linguistic investigation of information structure in
infant-directed speech
Kristine M. Yu (UMD College Park, UMASS Amherst), Sameer ud Dowla Khan (Brown University),
Megha Sundara (UCLA)
I
Hypothesis: prosodic modifications in IDS
consistent not only with engaging the infant’s
attention, but also with signaling information
structure relevant for IDS
I Cross-linguistic analysis in Bengali and English to tease
apart language-specific prosodic correlates of
information structure from global prosodic
modifications
I Focus on phonological intonational analysis to index
information structure
Speakers and speech materials
I
I
10 speakers for each language (5M/5F)
I Parents of young infants/children
I Speech materials: North Wind passage—Aesop’s fable
I Standard language sample with controlled pragmatic context
across languages
I Suitable for laboratory speech and IDS elicitation
I
L*
f01
200
Recording procedure
I
I
I
I
I
Annotation in Praat with phonemic transcription, syllable boundaries
ToBI annotation as a tool for analyzing intonational phonology
I
I
L*. . .Ha: default pattern of
successive rises (see fig. below)
L*+fH and fH* mark
focus/surprise
Boundary tones include default
L%, topicalized HL%, etc.
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Time (s)
1.6 1.8
2
2.2
2.4
2.6
2.8
3
I
I
I
H*: default marker of salience in
discourse
L+H* and L*+H markers of
focus/surprise
Boundary tone inventory less rich
than in Bengali
3.2
Ha
Fundamental frequency (Hz)
L*
80
I
60
15
Ha
L*
L*
L*
Ha
L* L*
40
150
100
rumu
nepaler
ranir
malider
namgulo
mone rakhte
Rumu
Nepal’s
queen’s
the gardeners’
the names
remember
pare
f10
20
50
15
15
f11
10
25
5
20
20
20
5
IP tone
10
0
0
0
0
0
H%
read ids
read ids
read ids
read ids
read ids
LH%
m01
m02
m03
m04
m05
HL%
30
40
25
15
15
15
10
10
5
5
5
5
0
0
0
0
10
5
0
0
read ids
read ids
m10
m4
m5
m6
12
5
0
0
read ids
ni
read ids
read ids
read ids
I
English pitch accent (PA) types:
H* (above), L+H* (below)
(Pierrehumbert & Steele 1990)
ETAP2, McGill University, 2011/09/23
40
30
20
20
10
10
30
10
0
0
0
20
50
0
read ids
read ids
m01
m02
m03
m04
m05
70
70
60
60
50
50
40
40
60
60
30
20
20
20
10
10
10
0
0
0
read ids
5
0
read ids
read ids
1.5±0.3 times more higher level
boundary tones and increased
variety of boundary types
disrupt regularity of rises in IDS.
10
10
10
10
10
0
0
0
0
0
read ids
read ids
read ids
read ids
m10
m4
m5
m6
40
40
40
30
30
20
10
0
0
read ids
20
20
10
0
read ids
read ids
10
10
0
0
read ids
read ids
Style
I
I
For most speakers, more pitch
accents—more words marked as
salient in discourse
More L+H* and L*+H: pitch
accents marking focus/surprise
IDS prosody showed language-specific
modifications that both engage the listener and
reflect the information structure relevant to IDS
Reduction in regularity of language-specific pitch variation in
IDS consistent with communicative goal of engaging
infant’s attention
I Increase in language-specific intonational phonological units
known to mark focus/surprise and topicalization
Acknowledgments
5
0
read ids
0
read ids
read ids
Style
I
20
20
read ids
30
20
20
read ids
30
20
30
40
40
20
30
40
30
30
40
50
30
read ids
80
40
40
30
100
30
50
Conclusions
I
couldn’t
Successive L*. . .Ha rises in
Bengali lab speech (Khan 2008)
40
m1
10
2
Style
I
50
f9
50
40
150
L+H*
10
4
0
0
50
f8
L*+H
6
5
10
5
15
10
10
read ids
15
14
20
10
10
10
8
15
10
50
L*
I
15
read ids
15
20
60
f11
40
60
H+!H*
m1
20
read ids
15
20
15
0
ambig
20
20
10
M%
25
30
20
HLH%
30
25
60
5
L%
f9
20
30
5
60
f10
H*<
25
10
20
70
f05
H*
IP
f8
25
10
10
30
Ha
L* L%
ip
40
40
Ha
Ha
f05
60
15
f04
More focus PAs: fH*
(surprise), L*+fH
(wh/corrective)
More HL%: boundary tones
marking topicalization
I
English
f04
20
50
200
I
20
f03
!H*
read ids
50
Boundary level
f02
25
300
250
Bengali
f01
f03
fH*
read ids
100
I
In both languages, parents reduced regularity of patterns of
pitch variation in IDS
English
L*+fH
Style
Frequency
I
I
*
Pitch accent
read ids
read ids
Results: prosodic modifications for engaging
listener’s attention
Frequency
Bengali
0
150
Information structure via intonational phonology
Reliance on previous work on Bengali (Khan 2008; to appear) and
English (e.g. Beckman and Ayers Elam 1993, Pierrehumbert and
Hirschberg 1990) ToBI in non-IDS speech linking language-specific
ToBI intonational categories and information structure
f02
20
50
Intonational analysis
I
L*+H
30
0
I
H*
40
100
Three repetitions for each speech style (lab speech, IDS)
Lab (read) speech as control: “Read at a comfortable pace”
Simulated IDS: “Read as if speaking to an infant (≈4-5 months)”
Subjects given stuffed toys; infants not in room to minimize audio
interruptions
English
Pitch accent
150
I
I
Parents also made language-specific prosodic modifications
consistent with increase in topicalization, focus in IDS
Bengali
Frequency
Cross-linguistically, infant-directed speech (IDS) (e.g. Blount &
Padgug 1977, Fernald & Simon 1984, Grieser & Kuhl 1988, Fernald et
al. 1989, Kitamura et al. 2002, Liu et al. 2007) shows prosodic
phonetic modifications such as an expanded pitch range
I
Results: changes in information structure in IDS
Frequency
Materials and Methods
Abstract
1.3±0.3 times more boundary
tones in IDS—more intermediate
(ip) and high level (IP) breaks.
I
I
Kristi Hendrickson, Amanda Ritchart, J’aime Roemer for help with
recruitment, recordings, and intonational labeling
Sun-Ah Jun for discussions
NSF graduate fellowship to first author and NSF BCS-0951639
Any opinions, findings, and conclusions or recommendations expressed in this material are
those of the authors and do not necessarily reflect the views of the National Science
Foundation.
[email protected], [email protected], [email protected]