Gender Typicality in Childrens Spoken Narratives: Acoustic and Perceptual Analyses Students: Heather

1
Gender Typicality in Childrens Spoken
NarrativesAcoustic and Perceptual
AnalysesStudents Heather Bauer, Alysse Zittnan,
Adviser Benjamin MunsonDepartment of
Speech-Language-Hearing Sciences, College of
Liberal Arts

• Gender Differences in Adults Speech
• Spoken language is highly variable. Variability
  affects all aspects of language sentence
  structures, discourse strategies, word choice,
  and pronunciation.
• Adult men and women pronounce speech sounds
  differently from one another. These differences
  are evident even in short stretches of speech,
  and allow a talkers sex to be ascertained at
  greater-than-chance levels even when other
  aspects of linguistic variation are controlled
• The differences are not solely the consequence of
  anatomic and physiologic differences between the
  sexes. We know this because
• 1. They are language- and culture-specific (Van
  Bejooijen, 1995)
• 2. There is only a weak correlation between
  variation in pronunciation and variation in the
  morphology of the speech-production mechanism
• 3. They are not across the board (Munson et
  al., 2006). They are more evident in some sound
  classes than in others. Moreover, there are
• Gender Differences in Childrens Speech
• Adults perceive boys and girls to sound different
  from one another well in advance of the sex
  differentiation in the speech-production
  mechanism that occurs at puberty (Perry, Ohde,
  Ashmead, 2001 inter alia).
• Gender differences in childrens pronunciation
  are not across the board. As with adults, boys
  and girls speech differs more in some sounds and
  sound classes than others.
• Ergo, sex differences are to represent learned
  behaviors that are the result of selective
  attention to and emulation of specific models in
  the ambient language.
• In our laboratory, we have studied gender
  typicality in childrens speech by comparing the
  productions of 5-13 year old boys with GID (i.e.,
  boys whose gender expression has been deemed not
  to meet cultural expectations) to children whose
  gender expression is deemed to meet cultural
  expectations (henceforth Expected Gender
  Development EGD). This provides us with an
  opportunity to study gender in speech without the
  confounding influence of biological sex.

• Results Acoustic Analyses
• When the average acoustic characteristics of
  vowels were considered, the only difference to
  approach statistical significance was one measure
  of formant frequencies of vowels. There were no
  differences for the consonants s or sh
• Regression analyses predicting perceived gender
  typicality from acoustic variables found that
  adult listeners paid particular attention to the
  way the vowel /æ/ was spoken. There was a weak
  association between the acoustic characteristics
  of s and sh. This is consistent with
  previous work on adults, in which it has been
  shown that adult attend to the characteristics of
  these sounds when judging the gender typicality
  of adults' voices. Together, 41 of variance in
  the ratings was accounted for.

• Speech Samples
• The recorded spoken narratives used in this
  experiment were elicited by asking each child to
  tell a story about a picture (see title banner).
  This was taken from a standardized test of
  childrens narrative abilities. Narratives were
  elicited as part of a larger data-collection
  protocol that included measures of single-word
  productions and sentence repetitions.
• Data from 28 children was used. Each speaker was
  categorized as either younger (average age 7
  years) or older (average age 10 years) and as
  either GID or EGD. The diagnoses of GID were
  made by psychometrists with expertise in
  childrens gender development. We used seven
  speakers from each category for a total of 28
  narratives, varying in length from about 30
  seconds to 1 minute.

• Results Perception Task
• Two, two-factor within-subjects ANOVAs examined
  the influence of gender typicality (GID vs. EGD)
  and age group on perceived gender and perceived
  age.
• For perceived gender, there was a significant
  main effect of gender group, F1,24 118.2, p lt
  0.001, partial h2 0.83. This interacted
  significantly with age, F1,24 4.6, p 0.042,
  partial h2 0.16. Figure 1 shows that the
  interaction occurred because the older GID boys
  sounded less boy-like than the younger GID boys,
  while the older EGD boys sounded more boy-like
  than the younger EGD boys.
• For perceived age, there was a significant main
  effect of both gender group (F1,24 19.2, p lt
  0.001, partial h2 0.44) and age group (F1,24
  178.4, p lt 0.001, partial h2 0.88). These
  interacted significantly, F1,24 24.0, p lt
  0.001, partial h2 0.50. Figure 2 shows that
  the interaction occurred because the older GID
  boys were rated to sound older than the older EGD
  boys, while there were no differences for the
  younger group.

• Acoustic Measures
• The acoustic analysis was conducted with the
  Praat signal-processing program.
• Our analysis focused on acoustic correlates of
  the articulation of vowels and consonants,
  including vowels fundamental frequency (the
  acoustic parameter that is perceived as pitch),
  vowels formant frequencies (the acoustic
  parameters that are perceived as vowel quality),
  and two consonants known to relate to the
  perception of gender, s and sh. These
  parameters were chosen based on previous studies
  that compared speech differences of adult men and
  women (e.g., Munson et al., 2006).
• Inter-rater reliability was assessed by having
  both students analyze randomly selected passages
  independently and compare their measures.

.
Figure 6. Perceived Gender based on Single
Words, Sentences, and Narratives
Figures 1 and 2. Spectrograms of the word sun
produced by a Boy with EGD (left) and
GID (right). The green circle is the s sound
and the red square is the first two formant
frequencies for the u sound.

• A comparison with previous acoustic and
  perceptual analyses of these childrens single
  word and sentence productions data showed that
  the ratings of the gender typicality of the
  narratives better differentiated between the two
  groups than the ratings gender from words or
  sentences
• However, the correlation between acoustic
  measures and the perceptual measures was weaker
  for the narratives than it was for the single
  words or the sentences.
• Listeners might attended to suprasegmental
  features (i.e., intonation, rate of speech) or
  narrative content more than pronunciation

• Discussion
• As expected, children with GID were rated as
  sounding less boy-like than their EGD peers.
  This tendency was stronger in the groups of older
  (10 year old) boys than in the younger (7 year
  old) boys. In fact, the older boys with GID were
  rated to sound less boy-like than the younger
  boys.
• When compared with previous perceptual data,
  stronger differentiation between groups was found
  for the narratives than for the single word
  productions or the sentences. This suggests that
  the less-structured task of narrative production
  gives children more opportunities to use
  permissible linguistic variation to convey gender
• At the same time, the fact that there was a
  smaller correlation between acoustic measures and
  perceptual measures for narratives than for
  single words and sentences suggests that some of
  the parameters that adults use to perceive gender
  in childrens speech are not related to
  pronunciation.
• Our ongoing research is examining the extent to
  which gender-typicality judgments are related to
  judgments of childrens overall language ability.
  Our perceived age findings suggests a positive
  relationship between how boy-like a child sounds
  and how developmentally advanced they are.

• Perception Task
• Twenty-five adult participants were asked to
  listen to each narrative and provide a rating of
  gender on a 6-point scale, where 6 indicated
  "definitely a boy", 1 indicated "definitely a
  girl", and 2 through 5 indicated various degrees
  of uncertainty (see Figure 3). They also
  provided an estimate of the child's age in years.
  The order of presentation was fully randomized.
  The task took place in a sound-treated booth.
  Responses were logged using the E-Prime software.
• Listeners were unaware that they were listening
  to children with GID and children with EGD. They
  were also not told that all of the children were
  boys.

• Purpose
• To examine whether the spoken narratives of
  children with GID differ from those of children
  with EGD perceptually (i.e., do adults rate the
  children as sounding different?) and acoustically
  (i.e., can we find evidence that children are
  producing speech differently?)
• Compare the acoustic and perceptual measures to
  previously made acoustic and perceptual measures
  of these childrens productions of single words
  and sentences, to examine the influence of
  linguistic complexity on the expression and
  perception of gender in childrens speech
• Significance
• Childrens gender expression is evidence of
  selective attention during language acquisition.
  Most theories of language posit that children
  treat the input uniformly. The results of this
  research have the potential to question this
  assumption, and suggests that models of language
  acquisition must posit a stronger role for
  selective attention during acquisition.

Figure 3. Perceived Gender Typicality
Figure 4. Perceived Age
Acknowledgements The research on the acoustic and
perceptual characteristics of the speech of
children with Gender Identity Disorder was
completed in collaboration initiated by J.
Michael Bailey of the Northwestern University
Department of Psychology. The funding for the
data collection phase of this project was
provided by a grant to Janet Pierrehumbert
(Department of Linguistics, Northwestern
University) and J. Michael Bailey from the
Northwestern University Office of the Vice
President for Research. The data-collection
protocol was designed by Janet Pierrehumbert,
Benjamin Munson, and Karla K. McGregor
(Department of Speech-Language Pathology and
Audiology, University of Iowa). Data collection
was supervised by Ken Zucker (Clark Institute for
Mental Heath, Toronto, Ontario). The acoustic
analysis protocol was developed by Benjamin
Munson and Janet Pierrehumbert. The perceptual
analysis protocol was developed by Benjamin
Munson. All of the acoustic and perceptual
analyses described herein were conducted at the
University of Minnesota. The authors of this
poster take sole responsibility for the
interpretations that are presented herein, and
acknowledges that the different collaborators on
this project might interpret these same findings
differently. References Munson, B., Jefferson,
S.V. McDonald, E.C. (2006). The influence of
perceived sexual orientation on fricative
perception. Journal of the Acoustical Society of
America, 119, 2427-2437 Perry, T.L., Ohde, R.,
Ashmead, D. (2001). The acoustic bases for
gender identification from children's voices.
Journal of the Acoustical Society of America,
109, 2988-2998. Van Bezooijen, R. (1995).
Sociocultural aspects of pitch differences
between Japanese and Dutch women. Language and
Speech, 38, 253-265.