Professional Documents
Culture Documents
Song A, Cola M, Plate S, Et Al. (2021) Natural Language Markers of Social Phenotype in Girls With Autism
Song A, Cola M, Plate S, Et Al. (2021) Natural Language Markers of Social Phenotype in Girls With Autism
Song A, Cola M, Plate S, Et Al. (2021) Natural Language Markers of Social Phenotype in Girls With Autism
Background: Girls with autism spectrum condition (ASC) are chronically underdiagnosed compared to boys, which
may be due to poorly understood sex differences in a variety of domains, including social interest and motivation. In
this study, we use natural language processing to identify objective markers of social phenotype that are easily
obtained from a brief conversation with a nonexpert. Methods: 87 school-aged children and adolescents with ASC
(17 girls, 33 boys) or typical development (TD; 15 girls, 22 boys) were matched on age (mean = 11.35 years), IQ
estimates (mean = 107), and – for ASC participants – level of social impairment. Participants engaged in an informal
5-min ‘get to know you’ conversation with a nonexpert conversation partner. To measure attention to social groups,
we analyzed first-person plural pronoun variants (e.g., ‘we’ and ‘us’) and third-person plural pronoun variants (e.g.,
‘they’ and ‘them’). Results: Consistent with prior research suggesting greater social motivation in autistic girls,
autistic girls talked more about social groups than did ASC boys. Compared to TD girls, autistic girls demonstrated
atypically heightened discussion of groups they were not a part of (‘they’, ‘them’), indicating potential awareness of
social exclusion. Pronoun use predicted individual differences in the social phenotypes of autistic girls. Conclu-
sions: Relatively heightened but atypical social group focus is evident in autistic girls during spontaneous
conversation, which contrasts with patterns observed in autistic boys and TD girls. Quantifying subtle linguistic
differences in verbally fluent autistic girls is an important step toward improved identification and support for this
understudied sector of the autism spectrum. Keywords: Autism spectrum condition; language; social phenotype;
sex differences; pronouns.
currently unknown whether differences in social depression use more first-person singular pronouns
motivation can be detected in natural language, in written essays than students who were never
examining how much children talk about other depressed (Rude, Gortner, & Pennebaker, 2004). In a
people (as indexed by personal pronouns) could study of narcissistic personality, participants spoke
provide initial insights. for 5 min about any topic they chose, and then
completed a narcissism personality inventory.
Greater narcissism was associated with the presence
Personal pronouns
of more first-person plural pronouns (we, us) in
Traditionally, when analyzing language, researchers participant monologues (Raskin & Shaw, 1988).
have focused on semantically rich content words like
nouns and verbs (Crandall, McDaniel, Watson, &
Personal pronouns in ASC
Yoder, 2019; McDuffie, Yoder, & Stone, 2006).
Function words, including pronouns, articles, prepo- A significant body of research suggests that individ-
sitions, conjunctions, and auxiliary verbs, are often uals with ASC use personal pronouns differently
overlooked or discarded from lexical analyses. How- than matched control participants without ASC
ever, more than half of all spoken words are function (Baltaxe & D’Angiola, 1996; Charney, 1980; Hauser
words (Rochon, Saffran, Berndt, & Schwartz, 2000), et al., 2019; Kanner, 1943; Naigles et al., 2016) or
and these ‘throwaway’ words contain meaningful with other conditions (Friedman, Lorang, & Sterling,
individual variation that predict a variety of com- 2019), particularly at younger ages (Arnold, Ben-
pelling outcomes (Chung & Pennebaker, 2007; Gor- netto, & Diehl, 2009). For example, children with
man et al., 2016; Irvine, Eigsti, & Fein, 2016; ASC have been reported to produce fewer first- and
Pennebaker, 2011; Pennebaker, Boyd, Jordan, & second-person pronouns (‘me’ and ‘you’) than intel-
Blackburn, 2015; Pennebaker, Mehl, & Niederhoffer, lectually disabled and typically developing compar-
2003). Personal pronouns, in particular, typically ison groups (Jordan, 1989). In 1994, Lee and
refer to individuals or groups of people (Kitagawa & colleagues showed that children with ASC produced
Lehrer, 1990). Variants include first-person singular the pronouns ‘you’ and ‘me’ at atypical rates, despite
(‘I’), first-person plural (‘we’), third-person singular having no difficulties with pronoun comprehension.
(‘he/she’), third-person plural (‘they’), and second- Rather than using pronouns, children were instead
person (‘you’) forms, and have been studied in more likely than nonautistic control group subjects
relation to a wide variety of social and cognitive to use proper nouns to refer to the experimenters
topics (Badr et al., 2016; Davis & Brock, 1975; and themselves (Lee, Hobson, & Chiat, 1994). The
Neysari et al., 2016; Nook, Schleider, & Somerville, gap between intact comprehension on one hand, and
2017; Simmons, Gordon, & Chambless, 2005; atypical production on the other, has also been
Twenge, Campbell, & Gentile, 2012). For example, reported in third-person pronoun production (e.g.,
leadership status and power relationships are ‘him’, ‘her’; (Hobson, Lee, & Hobson, 2010)).
reflected in personal pronoun use, such that leaders Studies hinting at a connection between social
with greater power use more first-person plural impairment and pronoun use suggest that differ-
pronouns (e.g., we, us) than those with less power ences in personal pronoun use could be an objective
(Kacewicz, Pennebaker, Davis, Jeon, & Graesser, metric for assessing social phenotypes in children
2014). Patterns of personal pronoun use also vary with ASC (Loveland & Landry, 1986). In the same
according to audience characteristics. In private study that examined third-person pronoun use by
accounts of relationship breakups (journal entries), autistic children (Hobson et al., 2010), researchers
people use more first-person singular (I, me) and examined the relationship between pronoun produc-
third-person plural (they, them) pronouns than in tion and communicative engagement. A rater, blind
public accounts (blog posts) of these same breakups, to children’s diagnostic status, reviewed videotapes
where they use more first-person plural pronouns of the pronoun production tasks, and rated partic-
(we, us) (Blackburn, Brody, & LeFebvre, 2014). ipants’ eye gaze direction during the tasks (i.e.,
Together, these results suggest that personal pro- orienting toward a third person or not), as well as
nouns may be sensitive indicators of interpersonal interpersonal connectedness with their conversa-
dynamics that vary according to contextual factors tional partner. They found that subjects with autism
such as target audience. and limited third-person pronoun production looked
Personal pronouns also reveal information about less at experimenters, and autistic participants who
people’s mental states and personalities, suggesting said ‘we’ more often were rated higher on interper-
that pronoun use might be informative for under- sonal connectedness (Hobson et al., 2010). These
standing individuals with psychiatric and neurode- results suggest a potential link between pronoun use
velopmental differences (Boals & Klein, 2005; and social behavior in ASC, such that subtle verbal
Brockmeyer et al., 2015; Kleim et al., 2018; Lyons, communication patterns – captured by first-person
Aksayli, & Brewer, 2018; Watson et al., 2012). For and third-person pronouns – might serve as objec-
example, college students with current or past tive markers of social phenotype.
Sex ratio 17f, 33m (66% male) 15f, 22m (59% male) v2 = 0.16, p = .69
Race Black/African American: 3 Black/African American: 8 v2 = 6.52, p = .16
Asian or Pacific Islander: 1 Asian or Pacific Islander: 1
White/Caucasian: 40 White/Caucasian: 23
Amber Song et al.
Multiracial: 4 Multiracial: 5
Other: 1 Other: 0
Maternal education ≤12 years: 3 <=12 years: 2 v2 = 1.52, p = .47
(in years) 13–16 years: 24 13–16 years: 23
17+ years: 22 17+ years: 12
Unknown: 1
Female Male Female Male Sex Dx Sex in ASC
Age (years) 11.48 (2.38) 11.79 (2.55) 11.38 (2.44) 10.58 (2.44) p = .74 p = .15 p = .68
8.90–16.35 8.25–16.66 8.22–15.51 8.16–16.64 d = .07 d = .32 d = .13
Full-Scale IQ 109.76 (11.99) 105.27 (13.03) 105 (13.07) 109.68 (13.70) p = .88 p = .74 p = .24
79–130 79–131 86–129 86–133 d = .03 d = .07 d = .35
Verbal IQ 108.35 (11.66) 104.76 (13.17) 105.87 (14.34) 109.27 (14.78) p = .87 p = .53 p = .35
85–130 70–130 80–128 86–131 d = .04 d = .14 d = .28
Nonverbal IQ 108.24 (14.48) 104.55 (12.43) 103.07 (12.93) 107.68 (11.49) p = .997 p = .997 p = .35
80–130 80–130 81–122 89–130 d=0 d=0 d = .28
ADOS-2 CSS 6.59 (2.69) 6.97 (2.01) 1.07 (0.26) 1.27 (0.46) p = .43 p < 2e 16 p = .57
Total 1–10 2–10 1–2 1–2 d = .09 d = 3.27 d = .17
ADOS-2 SA CSS 6.47 (2.62) 7.36 (1.76) 1.40 (0.74) 1.73 (0.83) p = .09 p < 2e–16 p = .16
1–10 3–10 1–3 1–3 d = .20 d = 3.22 d = .43
ADOS-2 RRB CSS 7.35 (1.66) 6.30 (2.39) 1 (0) 2.09 (2.11) p = .82 p < 2e 16 p = .11
5–10 1–10 1–1 1–7 d = .03 d = 2.50 d = .48
SCQ Totala 17.94 (7.11) 18.34 (6.79) 2.73 (2.52) 2.68 (3.27) p = .87 p < 2e 16 p = .85
6–31 5–33 0–8 0–14 d = .02 d = 2.81 d = .06
VABS-2a Expressive Voc 13.94 (3.51) 12.45 (3.15) 17.00 (1.35) 17.18 (1.68) p = .24 p<.001 p = .14
7–20 7–18 14–19 13–19 d = .27 d = 1.52 d = .54
Chi-square tests were used to compare distributions of sex, race, and parent education between diagnostic groups, and t-tests compared mean values for continuous variables. The first
subcolumn in the table (‘Sex’) refers to overall comparisons between males and females collapsed across diagnostic group, the second (‘Dx’) to overall comparisons between ASC and TD
groups collapsed across sex, and the third (‘Sex in ASC’) to a comparison of males and females in the ASC subgroup alone.
a
1 Male participant from the ASC group was missing an SCQ score; 2 female participants from the TD group were missing VABS-2 Expressive Vocabulary scaled scores.
severity scores (CSS) and SCQ scores. Parents or primary the samples and averaged 97.22% (range: 95.70%–98.53%).
caregivers were given an informed consent packet to review, After segmentation, speaker labeling, and transcription, words
and completed written informed consent upon arrival. Partic- produced by participant and confederate were separated using
ipants and parents were compensated for their time. All an in-house R script. Files were fed into LIWC software
participants in this sample consented to their data being used (Pennebaker et al., 2015), which calculated the overall number
for future studies; this study was overseen by the Children’s of words produced, as well as the number of first- and third-
Hospital of Philadelphia Institutional Review Board. person plural pronouns and ‘social’ category words produced
by each speaker (see Dependent variables, below).
Study procedure
Statistical approach
Linguistic data were drawn from a 5-min conversation between
the participant and study personnel (confederate), adminis- Data were analyzed using generalized linear mixed effects
tered as part of a larger study. The conversation was com- regression (GLMER) models in R (‘lme4’ package; R Core Team
pletely unstructured with no specific topics provided to either and contributors worldwide) with age and IQ (mean centered)
speaker. Twenty-three young adult confederates (21 females; as covariates, a random effect of confederate ID for separate
undergraduate students and research assistants) were ‘we’, ‘they’ and ‘social’ variant analyses, and random effects of
assigned to each participant based on availability (there were both participant ID and confederate ID for the primary
no significant differences in confederate sex distribution by omnibus test comparing ‘we’ versus ‘they’ production (since
participant sex or diagnostic group). Confederates were not participants produced both types of pronouns during every
informed of participant diagnosis, and their only instructions conversation). Estimated effects, standard errors (SE), z-
were to act natural and avoid dominating the conversation. To values, and p-values are provided. Variables were coded as
account for potential individual differences in confederate follows: TD = 0, ASC = 1; female = 0, and male = 1. Dependent
behavior, a random effect of confederate ID was included in variables were positive, interval, and non-normally distributed
all analyses. At the start of the conversation, the research (Shapiro–Wilk test ps < .001), so these data were modeled
assistant in charge of the visit said a variation of the phrase, using a Poisson distribution with a log link. Significance values
‘You two just chat and get to know each other. I’m going to for planned pairwise tests of GLMER estimated marginal
finish getting a few things set up’. Conversations were audio-/ means were corrected for multiple comparisons using the
video-recorded using a device with two HD video cameras and Tukey method. Effect sizes for GLMER are reported as
mics facing opposite directions, placed on a table between unstandardized effects (estimates; Pek & Flora, 2018), while
participants and confederates for simultaneous capture of Cohen’s d is reported for group mean differences on clinical
both speakers (Parish-Morris et al., 2018). and demographic variables (Table 1). Following Cohen (1988),
d = .2 is considered a ‘small’ effect, d = .5 a ‘medium’ effect,
and d = .8 a ‘large’ effect (Cohen, 1988). GLMER was used to
Measures assess relationships between pronoun production and clinical
phenotype (ADOS-2 Social Affect and Repetitive Behaviors
Participants in the study were administered a battery of domain scores).
diagnostic and cognitive tests including the Autism Diagnostic
Observation Schedule Second Edition (ADOS-2; (Lord, DiLa-
vore, et al., 2012). Either ADOS-2 module 3 or 4 was Dependent variables
administered to participants in this study by a research-
reliable clinician. ADOS-2 scores index two domains, Social Preliminary analyses controlling for age and IQ (mean cen-
Affect (SA) and Restricted and Repetitive Behaviors (RRBs), tered), and confederate ID revealed that participant groups
with higher scores indicating greater symptoms (Hus, Gotham, produced significantly different numbers of words during the 5-
& Lord, 2014). The Social Communication Questionnaire min conversations (Table S1). Thus, subsequent analyses were
(SCQ) – Lifetime version – is a yes/no parent response conducted on the number of first-person plural pronouns (‘we’,
questionnaire that asks parents or primary caregivers to ‘us’ and variants), third-person plural pronouns (‘they’, ‘them’
assess their child’s social behaviors associated with ASC and variants), and ‘social’ category words as calculated by LIWC
(Rutter, Bailey, & Lord, 2003), and was given to a parent/ (see Appendix S2), normalized per 1,000 words to account for
primary caregiver to fill out prior to clinical assessment. varying word production across conversations and to facilitate
Participant IQ estimates were calculated using one of the interpretation. We decided to normalize word use per 1,000
following tests, which were administered based on the protocol words based on the average range of words produced by
requirements of individual studies that were pooled across a participants in our study, and to illustrate relative frequency
large research center to generate the current sample: 57 without reporting percentages that could be misinterpreted
participants (35 TD, 24 ASC) received the Wechsler Abbrevi- when participants produced fewer overall words. We further
ated Scale of Intelligence, Second Edition (Wechsler, 2011), 3 avoided the use of proportions because they tend to violate the
participants (3 ASC) received the Wechsler Intelligence Scale underlying assumptions of common statistical tests, can be
for Children, Fifth Edition (Wechsler, 2014), 21 participants misleading when the number of words produced varies widely
(21 ASC) received the Stanford-Binet Intelligence Scales (Roid, (as in this study and most studies of productive language in
2003), 4 participants received the Differential Ability Scales, ASC), and do not generally adhere to the way words are counted
Second Edition (DAS-II, DAS-II School version (2 TD, 1 ASC)), (usually full words are counted as words and thus are better
or DAS-II Early version (1 ASC) used depending on age; Elliot, represented as count data than as decimals). Clinical pheno-
2007). Scores were standardized by a licensed neuropsychol- type was measured using ADOS-2 CSS scores.
ogist (J. Pandey) to create an overall cognitive estimate, a
verbal estimate, and a nonverbal estimate.
Results
Data processing A generalized linear mixed effects regression predict-
ing pronoun use revealed a significant three-way
Audio recordings of each conversation were segmented, labeled
by speaker, and orthographically transcribed using XTrans
interaction between sex, diagnosis, and pronoun
(Linguistic Data Consortium, 2018) (see Appendix S1). Average type (estimate: .94, SE = .18, z = 5.13, p < .001;
word-level transcription reliability was calculated for 20% of controlling for age and IQ (centered), with participant
Figure 1 Estimated marginal means (EMM) and standard errors (error bars) of ‘we’ variant production and ‘they’ variant production per
1,000 words in girls and boys with and without ASC, in two separate models accounting for age, full-scale IQ, and confederate ID.
and confederate IDs as random effects to account for diagnosis revealed a significant interactive effect of
repeated measures). Conditional main effects of sex and diagnosis on ‘they’ variants produced per
diagnosis (estimate: .71, SE: .24, z = 2.91, 1,000 words (estimate: .88, SE: .18, z = 4.92,
p = .004), sex (estimate: .43, SE: .23, z = 1.91, p < .001). Tukey-corrected pairwise comparisons of
p = .056), and pronoun type (‘they’ estimate: .94, estimated marginal means revealed that the interac-
SE: .10, z = 9.53, p < .001) on overall pronoun use tion was driven by significantly increased relative
emerged. To clarify the nature of the interaction, ‘they’ production in autistic girls (Figure 1;
subsequent analyses were conducted within each Table S1), such that autistic girls used significantly
pronoun type separately. more ‘they’ than autistic boys (estimate: .90, SE: .12,
z = 7.39, p < .0001; Table S1), TD girls (estimate:
.34, SE: .13, z = 2.65, p = .04), and TD boys
First-person plural (‘we’ variants)
(estimate: .36, SE: .11, z = 2.98, p = .02). ‘They’
A GLMER including confederate ID, age, IQ, sex, production by TD girls and boys did not differ from
diagnosis, and the interaction between sex and one another (estimate: .02, SE: .13, z = .13, p = .99;
diagnosis revealed a nonsignificant interaction ASC girls > TD girls = TD boys > ASC boys; Fig-
effect, so this factor was removed from the model. ure 1; Table S1).
The final model with sex and diagnosis predicting
‘we’ production per 1,000 words revealed a signifi-
‘Social’ category words
cant main effect of diagnosis (estimate: .31, SE:
.07, z = 4.25, p < .001; Table S1), such that par- To understand the broader context of differences in
ticipants with ASC produced less ‘we’ than TD pronoun use, we explored the amount of social talk
participants. A significant effect of sex also emerged; produced by each participant by analyzing ‘social’
boys were less likely to produce ‘we’ variants than category words as measured by LIWC. A GLMER
girls (estimate: .28, SE: .07, z = 4.28, p < .001). including age, IQ, sex, diagnosis, confederate ID,
Tukey-corrected pairwise comparisons of estimated and the interaction between sex and diagnosis
marginal means revealed that ‘we’ production dif- revealed a significant interactive effect of sex and
fered significantly in all subgroups except ASC girls diagnosis on ‘social’ words produced per 1,000
and TD boys, who did not differ significantly from words (estimate: .25, SE: .06, z = 4.52,
one another (TD girls > TD boys = ASC girls > ASC p < .001). Tukey-corrected pairwise comparisons of
boys; Figure 1; Table S1). estimated marginal means revealed that the interac-
tion was driven by reduced social word production in
ASC boys (mean = 71.47) compared to ASC girls
Third-person plural (‘they’ variants)
(mean = 92.45, z = 6.77, p < .0001), TD boys
A GLMER including confederate ID, age, IQ, sex, (mean = 94.62, z = 7.48, p < .0001), and TD girls
diagnosis, and the interaction between sex and (mean = 95.26, z = 7.58, p < .0001). Rates of social
word use by ASC girls, TD girls, and TD boys did not important – and previously unreported – nuances
differ significantly from one another (all ps > .88). in personal pronoun use that varied by sex in ASC.
Thus, we found that ASC boys produced distinctly Sex differences in word use emerged for both first-
lower levels of social talk, while ASC girls, TD girls, and third-person plural pronouns in ASC, with girls
and TD boys produced patterns of social talk that did consistently producing more social group-related
not differ significantly from one another. This anal- talk compared to boys. One explanation for this
ysis provides convergent evidence for differential finding is that autistic girls are hyperaware of groups
social language phenotypes in boys and girls with they are (and are not) not included in, while ASC
and without ASC, and suggests a connection boys – who produced diminished social group talk
between differential rates of pronoun production across the board – are comparatively less likely to
and social language more broadly. talk about social topics, regardless of group mem-
bership status. This conclusion is consistent with
research showing that girls with ASC tend to hover
Predicting clinical phenotype
around social groups on the playground and be
To determine whether pronoun production was neglected socially – in contrast to ASC boys who tend
associated with clinical phenotype in ASC, pronouns to be alone and rejected (Dean et al., 2014, 2017).
were used to predict ADOS-2 scores in the ASC Our pattern of findings could also be interpreted as
group alone. After controlling for age and IQ (cen- evidence of social camouflage or compensation,
tered), with random effects of participant ID and which may be more common in females than males
confederate ID, a GLMER revealed that ‘we’ variants with ASC (Allely, 2018) and may or may not be
and ‘they’ variants significantly predicted ADOS-2 consciously deployed. From that perspective, ASC
social affect (SA) CSS scores in the ASC group (‘we’ girls with heightened ‘they’ production might have
estimate: .01, SE = .006, z = 2.38, p = .02; ‘they’ learned to match TD levels of social talk as a way to
estimate: .01, SE: .007, z = 2.21, p = .03). A blend in linguistically with peers – thus partially
separate GLMER revealed that ‘we’ and ‘they’ vari- ‘normalizing’ natural speech (Parish-Morris et al.,
ants did not significantly predict ADOS-2 repetitive 2017).
behavior (RRB) scores (all ps>.19), suggesting that Sex differences in plural pronoun use by TD
pronoun use is specifically associated with social children and adolescents were also revealed, such
function and not repetitive behaviors. Exploratory that TD girls talked more about social groups they
subgroup analyses in ASC boys and girls separately were a part of (‘we’ variants) than did TD boys. Long-
showed that the relationship between ADOS-2 SA researched sex differences in friendship structure
scores in the overall ASC sample was driven by during childhood and adolescence may underlie the
significant linkages between pronoun use and greater amounts of ‘we’ talk observed in girls relative
ADOS-2 SA scores in autistic girls (‘we’ estimate: to boys in both diagnostic groups, which is also
.02, SE = .01, z = 1.87, p = .06; ‘they’ estimate: consistent with recent research showing that the
.03, SE: .01, z = 2.75, p = .006). Prediction in friendship structures of ASC girls and boys differ
autistic boys was nonsignificant (all ps > .21), along typical lines (Sedgewick, Hill, & Pellicano,
although these results should be viewed with cau- 2019; Sedgewick et al., 2016). In contrast to the
tion due to the small sample sizes in each subgroup. sex-differentiated patterns of ‘they’ production
observed in ASC, however, there was no effect of
sex on the amount of talk by TD boys and girls about
Discussion social groups they are not a part of (i.e., both sexes
This study is the first to compare plural pronoun use were equally likely to use ‘they’ variants). This
by girls and boys with and without ASC, and one of suggests that sex differences in ‘they’ production by
the few in the literature to analyze language pro- children and adolescents in the ASC group may be
duced by autistic children and adolescents during autism-specific.
brief conversations with nonexpert interlocutors. A Interestingly, autistic girls in our sample produced
number of notable findings emerged: First, we found significantly more third-person plural pronouns
that as a group, autistic children and adolescents (‘they’, ‘them’) than TD girls, which could indicate
used significantly fewer plural personal pronouns even greater-than-average social interest and moti-
than matched TD peers. This main effect of autism vation. However, autistic girls’ heightened focus on
on personal pronoun production is broadly consis- social groups (increased ‘they’) may be complicated
tent with reports of reduced social motivation and by their diminished membership in those groups
social attention in autism (Chevallier et al., 2012). (diminished ‘we’) relative to same-sex TD peers. This
Based on this finding and consistent with prior tension – between heightened awareness of social
research, reduced or atypical personal pronoun use groups on one hand, and reduced membership on
appears to be a good diagnostic marker for ASC, and another – could index social exclusion, which may
may – pending future research in younger children – contribute to elevated levels of depression and anx-
prove clinically useful for lowering the age of first iety in girls with ASC (Bargiela, Steward, & Mandy,
autism diagnosis. Our second finding revealed 2016). Recent research supports this interpretation,
as the friendship structures of girls and boys have prompting is conducted in approximately the same
been shown to vary systematically in ways that drive way across participants and would reduce the need to
sex-differentiated friendship experiences in ASC control for variable confederate behavior. However, a
(Sedgewick et al., 2019). Thus, when it comes to standardized interview administered by an expert is
identifying girls with ASC vs. typical girls, our results also less generalizable to everyday life experiences.
suggest that while the overall amount of talk about Second, it is still unknown whether these effects will
social groups is high in both populations, and might hold in older and younger samples, and in samples
therefore not be a reliable marker for whether or not with lower IQ estimates. A developmentally matched
a girl should be referred for an autism evaluation, comparison group (e.g., with Down syndrome) would
how girls talk about social groups might be a good allow us to extend our IQ range, and pilot research
indicator of social function that could be used to using this approach with ASC and TD children as
guide clinical decision-making. young as 5 years old is currently underway. Our
In line with recent research showing that autistic sample also lacked the power to determine whether
girls experience enhanced social motivation relative to medication use (e.g., SSRIs) may have an effect on
autistic boys (Sedgewick et al., 2016), we found that pronoun use, which is a promising avenue for future
autistic girls produced significantly more social words research. Third, the girls in this sample were identified
than autistic boys did – not just personal pronouns – as autistic during childhood despite a recognized
indicating generally heightened conversational atten- problem of late and missed diagnoses for females (Baio
tion to social topics. Sex differences in language et al., 2018; Loomes et al., 2017). Thus, the girls in this
markers of social motivation in autism are especially sample may differ systematically from autistic girls
important in light of the late (Baioetal., 2018)or missed who are diagnosed later, during adolescence in adult-
(Loomes et al., 2017) diagnoses that are common for hood, and it is unknown whether or not these findings
autistic girls and women. Specifically, heightened talk generalize to later-diagnosed or undiagnosed girls and
about social topics could complicate ASC referral and women on the spectrum. Fourth, it is important to
diagnosis when observers expect a male-centric ‘autis- further analyze what kind of group-focused talk chil-
tic’ behavioral pattern of reduced social motivation and dren produced. Although girls with ASC produced
attention to social groups, which ASC girls do not social category words at rates that did not differ from
necessarily exhibit. TD girls, it is nonetheless possible that girls with ASC
Finally, our study showed that pronoun use pre- spoke about social groups in ways that differed qual-
dicts social phenotype in autistic girls, such that itatively – if not quantitatively – from TD girls (e.g.,
greater attentional focus on social groups predicts differential reliance on proper nouns vs. pronouns (Lee
fewer (or milder) autism symptoms as rated by an et al., 1994), variable talk about friends vs. family, or
expert clinician. This relationship was not present in use of they as a gender neutral pronoun). Future
autistic boys, perhaps due to the restricted range of research will examine qualitative differences in the
pronoun production in that subgroup. Despite delib- social group-related talk in ASC and TD boys and girls.
erately matching boys and girls on overall autism Fifth, our study did not measure the real-world effects
symptoms, boys in our sample were significantly less of variable pronoun production, which may differen-
likely than girls to use both types of plural pronouns. tially impact the peer experiences of boys vs. girls;
Thus, while the social phenotype of girls with ASC future research is necessary to tease apart relation-
appears measurable via personal pronoun produc- ships between sex-differentiated peer social contexts
tion during brief conversations, it may be more and demands on the one hand, and language behavior
difficult to assess in boys (at least, using personal on the other. Sixth, our measure of social phenotype
pronouns as a metric). (ADOS-2 social affect total score) is not designed to be a
dimensional measure of social phenotype. A targeted
questionnaire about social interest and motivation, or
Limitations and future directions
a behavioral measure like attention to social stimuli
A number of notable strengths distinguish this study during eye tracking, might correlate more strongly
from prior research, including sufficient numbers of with pronoun use. Finally, given the lack of validated
ASC and TD girls and boys to analyze diagnosis- and social camouflaging or compensation measures for
sex-based differences in pronoun use. However, cer- participants in our age range, we do not have self-
tain limitations should be considered when interpret- report of this behavior; this limits our ability to inter-
ing our results. First, although our sample included a pret word-level differences as being due to this phe-
greater-than-average percentage of autistic girls (33% nomenon.
of the ASC group), this study bears replication with
larger samples. Future research in our laboratory is
planned, which will expand this approach to include Conclusion
more participants in semistructured question-and- Our study addresses multiple gaps in the literature
answer formats, such as the ADOS-2 social interview by exploring personal pronouns as linguistic mark-
section. This approach – with standardized social ers of social phenotype in brief spontaneous conver-
prompts – would increase the likelihood that social sations of autistic girls and boys, as compared to an
Key points
The communication patterns of girls with autism are just beginning to be understood.
In this study, we used computational linguistics to show that school-aged girls and boys with autism talk
differently about social groups during spontaneous conversations.
It is important to understand the natural communication patterns of girls with autism so they can receive
appropriate supports.
Boorse, J., Cola, M., Plate, S., Yankowitz, L., Pandey, J., Friedman, L., Sterling, A., DaWalt, L.S., & Mailick, M.R.
Schultz, R.T., & Parish-Morris, J. (2019). Linguistic markers (2019). Conversational language is a predictor of vocational
of autism in girls: Evidence of a “blended phenotype” during independence and friendships in adults with ASD. Journal of
storytelling. Molecular Autism, 10, 14. Autism and Developmental Disorders, 49, 4294–4305.
Brewer, M.B., & Gardner, W. (1996). Who is this “We”? Levels Goddard, L., Dritschel, B., & Howlin, P. (2014). A preliminary
of collective identity and self representations. Journal of study of gender differences in autobiographical memory in
Personality and Social Psychology, 71, 83–93. children with an autism spectrum disorder. Journal of
Brockmeyer, T., Zimmermann, J., Kulessa, D., Hautzinger, M., Autism and Developmental Disorders, 44, 2087–2095.
Bents, H., Friederich, H.-C., . . . & Backenstrass, M. (2015). Gorman, K., Olson, L., Hill, A.P., Lunsford, R., Heeman, P.A., &
Me, myself, and I: Self-referent word use as an indicator of van Santen, J.P.H. (2016). “Uh” and “um” in children with
self-focused attention in relation to depression and anxiety. autism spectrum disorders or language impairment. Autism
Frontiers in Psychology, 6, 1564. Research, 9, 854–865.
Brown, L. (2011). Identity-First Language. Autistic Self Advo- Grzadzinski, R., Dick, C., Lord, C., & Bishop, S. (2016). Parent-
cacy Network. http://autisticadvocacy.org/about-asan/ide reported and clinician-observed autism spectrum disorder
ntity-first-language/ (ASD) symptoms in children with attention deficit/hyperac-
Charney, R. (1980). Pronoun errors in autistic children: tivity disorder (ADHD): Implications for practice under DSM-
Support for a social explanation. The British Journal of 5. Molecular Autism, 7(1), 7. https://doi.org/10.1186/
Disorders of Communication, 15, 39–43. s13229-016-0072-1.
Chevallier, C., Kohls, G., Troiani, V., Brodkin, E.S., & Schultz, Harrop, C., Jones, D.R., Sasson, N.J., Zheng, S., Nowell, S.W.,
R.T. (2012). The social motivation theory of autism. Trends & Parish-Morris, J. (2020). Social and object attention is
in Cognitive Sciences, 16, 231–239. influenced by biological sex and toy gender-congruence in
Chung, C., & Pennebaker, J.W. (2007). The psychological children with and without autism. Autism Research, 13,
functions of function words. In K. Fielder (Ed.), Social 763–776.
communication (pp. 343–359). New York, NY: Psychology Harrop, C., Jones, D., Zheng, S., Nowell, S., Schultz, R., &
Press. Parish-Morris, J. (2019). Visual attention to faces in children
Clements, C.C., Zoltowski, A.R., Yankowitz, L.D., Yerys, B.E., with autism spectrum disorder: Are there sex differences?
Schultz, R.T., & Herrington, J.D. (2018). Evaluation of the Molecular Autism, 10, 28.
social motivation hypothesis of autism: A systematic review Hauser, M., Sariyanidi, E., Tunc, B., Zampella, C., Brodkin, E.,
and meta-analysis. JAMA Psychiatry, 75, 797. Schultz, R., & Parish-Morris, J. (2019). Using natural
Cohen, J. (1988). Statistical power analysis for the behavioral conversations to classify autism with limited data: Age
sciences (2nd edn). New York, NY: Academic Press. matters. In Proceedings of the Sixth Workshop on Computa-
Cola, M.L., Plate, S., Yankowitz, L., Petrulla, V., Bateman, L., tional Linguistics and Clinical Psychology (pp. 45–54).
Zampella, C.J., . . . & Parish-Morris, J. (2020). Sex differ- Hobson, R.P., Lee, A., & Hobson, J.A. (2010). Personal
ences in the first impressions made by girls and boys with pronouns and communicative engagement in autism. Jour-
autism. Molecular Autism, 11, 49. nal of Autism and Developmental Disorders, 40, 653–664.
Cramer, M.R., & Schuman, H. (1975). We and they: Pronouns Hus, V., Gotham, K., & Lord, C. (2014). Standardizing ADOS
as measures of political identification and estrangement. Domain Scores: Separating severity of social affect and
Social Science Research, 4, 231–240. restricted and repetitive behaviors. Journal of Autism and
Crandall, M.C., McDaniel, J., Watson, L.R., & Yoder, P.J. Developmental Disorders, 44, 2400–2412.
(2019). The relation between early parent verb input and Irvine, C.A., Eigsti, I.-M., & Fein, D.A. (2016). Uh, um, and
later expressive verb vocabulary in children with autism autism: Filler disfluencies as pragmatic markers in adoles-
spectrum disorder. Journal of Speech, Language, and Hear- cents with optimal outcomes from autism spectrum disor-
ing Research, 62, 1787–1797. der. Journal of Autism and Developmental Disorders, 46,
Davis, D., & Brock, T.C. (1975). Use of first person pronouns as 1061–1070.
a function of increased objective self-awareness and perfor- Jordan, R.R. (1989). An experimental comparison of the
mance feedback. Journal of Experimental Social Psychology, understanding and use of speaker-addressee personal pro-
11, 381–388. nouns in autistic children. International Journal of Language
Dean, M., Harwood, R., & Kasari, C. (2017). The art of & Communication Disorders, 24, 169–179.
camouflage: Gender differences in the social behaviors of Kacewicz, E., Pennebaker, J.W., Davis, M., Jeon, M., &
girls and boys with autism spectrum disorder. Autism, 21, Graesser, A.C. (2014). Pronoun use reflects standings in
678–689. social hierarchies. Journal of Language and Social Psychol-
Dean, M., Kasari, C., Shih, W., Frankel, F., Whitney, R., Landa, ogy, 33, 125–143.
R., . . . & Harwood, R. (2014). The peer relationships of girls Kanner, L. (1943). Autistic disturbances of affective contact.
with ASD at school: Comparison to boys and girls with and Nervous Child, 2, 217–250.
without ASD. Journal of Child Psychology and Psychiatry, Kauschke, C., van der Beek, B., & Kamp-Becker, I. (2016).
55, 1218–1225. Narratives of girls and boys with autism spectrum disorders:
Dunn, D.S., & Andrews, E.E. (2015). Person-first and identity- Gender differences in narrative competence and internal
first language: Developing psychologists’ cultural compe- state language. Journal of Autism and Developmental Disor-
tence using disability language. American Psychologist, 70, ders, 46, 840–852.
255–264. Kenny, L., Hattersley, C., Molins, B., Buckley, C., Povey, C., &
Dworzynski, K., Ronald, A., Bolton, P., & Happ e, F. (2012). Pellicano, E. (2016). Which terms should be used to describe
How different are girls and boys above and below the autism? Perspectives from the UK autism community.
diagnostic threshold for autism spectrum disorders? Journal Autism, 20, 442–462.
of the American Academy of Child & Adolescent Psychiatry, Kitagawa, C., & Lehrer, A. (1990). Impersonal uses of personal
51, 788–797. pronouns. Journal of Pragmatics, 14, 739–759.
Elliot, C.D. (2007). Differential ability Scalesâ-II - DAS-II. Kleim, B., Kleim, B., Horn, A.B., Kraehenmann, R., Mehl, M.R.,
Harcourt Assessment. & Ehlers, A. (2018). Early linguistic markers of trauma-
Friedman, L., Lorang, E., & Sterling, A. (2019). The use of specific processing indicate vulnerability for later chronic
demonstratives and personal pronouns in fragile X syn- posttraumatic stress disorder. Frontiers in Psychiatry, 9,
drome and autism spectrum disorder. Clinical Linguistics & 405.
Phonetics, 33, 420–436.
Klin, A. (2000). Attributing social meaning to ambiguous visual Pennebaker, J.W. (2011). The Secret Life of Pronouns: What Our
stimuli in higher-functioning autism and Asperger syn- Words Say About Us. New York, NY: Bloomsbury Publishing
drome: The Social Attribution Task. Journal of Child Psy- USA.
chology and Psychiatry, and Allied Disciplines, 41, 831–846. Pennebaker, J.W., Boyd, R.L., Jordan, K., & Blackburn, K.
Lee, A., Hobson, R.P., & Chiat, S. (1994). I, you, me, and (2015). The development and psychometric properties of
autism: An experimental study. Journal of Autism and LIWC2015. Available from: https://repositories.lib.utexa
Developmental Disorders, 24, 155–176. s.edu/handle/2152/31333
Linguistic Data Consortium (2018). XTrans. https://www.ldc. Pennebaker, J.W., Mehl, M.R., & Niederhoffer, K.G. (2003).
upenn.edu/language-resources/tools/xtrans Psychological aspects of natural language use: Our words.
Loomes, R., Hull, L., & Mandy, W.P.L. (2017). What is the Our Selves. Annual Review of Psychology, 54, 547–577.
male-to-female ratio in autism spectrum disorder? A sys- Raskin, R., & Shaw, R. (1988). Narcissism and the use of
tematic review and meta-analysis. Journal of the American personal pronouns. Journal of Personality, 56, 393–404.
Academy of Child & Adolescent Psychiatry, 56, 466–474. Ratto, A.B., Kenworthy, L., Yerys, B.E., Bascom, J., Wieck-
Lord, C., DiLavore, P.C., Gotham, K., Guthrie, W., Luyster, owski, A.T., White, S.W., . . . & Anthony, L.G. (2017). What
R.J., Risi, S., & Rutter, M. (2012). Autism diagnostic obser- about the girls? Sex-based differences in autistic traits and
vation schedule: ADOS-2. Torrance, CA: Western Psycholog- adaptive skills. Journal of Autism and Developmental Disor-
ical Services. ders, 48, 1698–1711.
Lord, C., Petkova, E., Hus, V., Gan, W., Lu, F., Martin, D.M., . . . Rochon, E., Saffran, E.M., Berndt, R.S., & Schwartz, M.F.
& Risi, S. (2012). A multi-site study of the clinical diagnosis (2000). Quantitative analysis of aphasic sentence produc-
of different autism spectrum disorders. Archives of General tion: Further development and new data. Brain and Lan-
Psychiatry, 69, 306–313. guage, 72, 193–218.
Lord, C., Risi, S., & Bishop, S.L. (2012). Autism diagnostic Roid, G.H. (2003). Stanford-Binet Intelligence Scales (5th edn).
observation schedule, second edition (ADOS-2). Torrance, Western Psychological Services.
CA: Western Psychological Services. Rose, V., Trembath, D., Keen, D., & Paynter, J. (2016). The
Loveland, K.A., & Landry, S.H. (1986). Joint attention and proportion of minimally verbal children with autism spec-
language in autism and developmental language delay. trum disorder in a community-based early intervention
Journal of Autism and Developmental Disorders, 16, 335– programme. Journal of Intellectual Disability Research: JIDR,
349. 60, 464–477.
Lyons, M., Aksayli, N.D., & Brewer, G. (2018). Mental distress Rotheram-Fuller, E., Kasari, C., Chamberlain, B., & Locke, J.
and language use: Linguistic analysis of discussion forum (2010). Social involvement of children with autism spectrum
posts. Computers in Human Behavior, 87, 207–211. disorders in elementary school classrooms. Journal of Child
Maenner, M.J., Shaw, K.A., Baio, J., Washington, A., Patrick, Psychology and Psychiatry, 51, 1227–1234.
M., DiRienzo, M., . . . & Dietz, P.M. (2020). Prevalence of Rude, S., Gortner, E.-M., & Pennebaker, J. (2004). Language
autism spectrum disorder among children aged 8 years— use of depressed and depression-vulnerable college stu-
autism and developmental disabilities monitoring network, dents. Cognition and Emotion, 18, 1121–1133.
11 Sites, United States, 2016. MMWR Surveillance Sum- Rutter, M., Bailey, A., & Lord, C. (2003). Social communication
maries, 69(4), 1–12. https://doi.org/10.15585/mmwr. questionnaire (SCQ). Western Psychological Services.
ss6904a1 Rynkiewicz, A., Schuller, B., Marchi, E., Piana, S., Camurri, A.,
McDuffie, A.S., Yoder, P.J., & Stone, W.L. (2006). Fast-map- Lassalle, A., & Baron-Cohen, S. (2016). An investigation of
ping in young children with autism spectrum disorders. First the ‘female camouflage effect’ in autism using a computer-
Language, 26, 421–438. ized ADOS-2 and a test of sex/gender differences. Molecular
M€uller, E., Schuler, A., & Yates, G.B. (2008). Social challenges Autism, 7, 10.
and supports from the perspective of individuals with Sedgewick, F., Hill, V., & Pellicano, E. (2019). ‘It’s different for
Asperger syndrome and other autism spectrum disabilities. girls’: Gender differences in the friendships and conflict of
Autism, 12, 173–190. autistic and neurotypical adolescents. Autism, 23, 1119–
Naigles, L.R., Cheng, M., Xu Rattanasone, N., Tek, S., Khetra- 1132.
pal, N., Fein, D., & Demuth, K. (2016). “You’re telling me!” Sedgewick, F., Hill, V., Yates, R., Pickering, L., & Pellicano, E.
The prevalence and predictors of pronoun reversals in (2016). Gender differences in the social motivation and
children with autism spectrum disorders and typical devel- friendship experiences of autistic and non-autistic adoles-
opment. Research in Autism Spectrum Disorders, 27, 11–20. cents. Journal of Autism and Developmental Disorders, 46,
Neysari, M., Bodenmann, G., Mehl, M.R., Bernecker, K., 1297–1306.
Nussbeck, F.W., Backes, S., . . . & Horn, A.B. (2016). Simmons, R.A., Gordon, P.C., & Chambless, D.L. (2005).
Monitoring pronouns in conflicts. GeroPsych, 29, 201–213. Pronouns in marital interaction: What do “You” and “I” say
Nook, E.C., Schleider, J.L., & Somerville, L.H. (2017). A about marital health? Psychological Science, 16, 932–936.
linguistic signature of psychological distancing in emotion Sturrock, A., Yau, N., Freed, J., & Adams, C. (2019). Speaking
regulation. Journal of Experimental Psychology: General, the same language? A preliminary investigation, comparing
146, 337. the language and communication skills of females and
Parish-Morris, J., Liberman, M.Y., Cieri, C., Herrington, J.D., males with high-functioning autism. Journal of Autism and
Yerys, B.E., Bateman, L., . . . & Schultz, R.T. (2017). Developmental Disorders, 50, 1639–1656.
Linguistic camouflage in girls with autism spectrum disor- Tager-Flusberg, H., & Kasari, C. (2013). Minimally verbal
der. Molecular Autism, 8, 6. school-aged children with autism spectrum disorder: The
Parish-Morris, J., Sariyanidi, E., Zampella, C., Bartley, G.K., neglected end of the spectrum: minimally verbal children
Ferguson, E., Pallathra, A.A., . . . & Tunc, B. (2018). Oral- with ASD. Autism Research, 6, 468–478.
motor and lexical diversity during naturalistic conversations Twenge, J.M., Campbell, W.K., & Gentile, B. (2012). Male and
in adults with autism spectrum disorder. In Proceedings of female pronoun use in U.S. Books reflects women’s status,
the Fifth Workshop on Computational Linguistics and Clinical 1900–2008. Sex Roles, 67, 488–493.
Psychology: From Keyboard to Clinic (pp. 147–157). Watson, A.R., Defteralı, C ß ., Bak, T.H., Sorace, A., McIntosh,
Pek, J., & Flora, D.B. (2018). Reporting effect sizes in original A.M., Owens, D.G.C., . . . & Lawrie, S.M. (2012). Use of
psychological research: A discussion and tutorial. Psycho- second-person pronouns and schizophrenia. The British
logical Methods, 23, 208. Journal of Psychiatry, 200, 342–343.
Wechsler, D. (2011). Wechsler Abbreviated Scale of Intelli- World Health Organization (2015). Fact sheet on gender: Key
genceâ—Second Edition (WASIâ- II). Pearson Clinical. facts, impact on health, gender equality in health and WHO
Wechsler, D. (2014). Wechsler Intelligence Scale for Childrenâ- response. World Health Organization. http://www.who.int/
Fifth Edition. Pearson Clinical. news-room/fact-sheets/detail/gender
Wood-Downie, H., Wong, B., Kovshoff, H., Mandy, W., Hull, L.,
& Hadwin, J.A. (2020). Sex/gender differences in camou- Accepted for publication: 29 September 2020
flaging in children and adolescents with autism. Journal of
Autism and Developmental Disorders. https://doi.org/10.
1007/s10803-020-04615-z [Epub ahead of print].