2/22/2014

An autistic language strength, and an example of bias in autism research (From the AAAS Meeting)

People often say we learn more about others through their nonverbal communication than through their words.  But our words also convey a great deal of content beyond the message they express.  From others' accents and other vocal characteristics, we can learn their age, gender, region of origin (e.g., Southern US), racial or ethnic background, whether they had a different native language, and sometimes what that native language might have been.  Neurotypical people can recognize all of these characteristics except for region given very little information--for example, given short samples or whispered speech.  (With regional accents, they require a longer speech sample that offers more opportunities for subtle dialect variation).

Autistic people are more likely than average to have difficulty processing spoken language, and may pick up less than neurotypicals from others' nonverbal communication.  So a team of researchers wanted to know: can autistic adults also learn about others' age and region of origin from their speech?

We don't just learn information about others through their accents, though: we also make judgments about them based on this information.  Neurotypical people perceive others' intelligence, success, friendliness, and reliability differently depending on what region of the country they come from.  In the United States, for example, they may view New Englanders as more successful, or Southerners as friendlier.  In general, neurotypical people perceive speakers from their region of origin as more friendly and stable than people from other regions.

The researchers wanted to know if autistic adults also have these biases based on regional stereotypes.  They predicted that even if autistic adults perceived speaker characteristics from accent, they would not use these to make assumptions about the speaker.

The Experiment
In the first part of the experiment, 32 neurotypical and 13 ASD adults, all monoliingual English speakers were asked to judge speaker age from an audio clip.  The audio clips were from 18 American male talkers, ranging from their early 20's to their 40's, without strong regional accents.

In the second part of the experiment, 30 neurotypical and 15 ASD adults were asked to determine speaker region of origin from audio clips.  The speakers were 5 male talkers from 4 US dialect regions: Northern, New England, Midland, and Southern.  Participants were from Columbus Ohio, which was in the Midland region.
Above: Map of regional U.S. accents, including those used in the experiment.
In the third part of the experiment, 28 neurotypical and 14 ASD adults listened to the audio clips from Part 1.  Thus, they were asked to judge speaker's friendliness, reliability, intelligence, and successfulness from age cues.

In the fourth part of the experiment, 31 neurotypical and 14 ASD adults listened to the audio clips from Part 2 and were asked to judge speaker friendliness, reliability, intelligence, and successfulness from regional accent cues.

Of note, participants were not matched for gender.  The first author, who presented the study, explained that her group was unable to recruit more than two or three ASD women.  The neurotypical participants--all college students taking introductory linguistics classes--were mostly women (as is perhaps representative of such classes?).  The experimenter did not explain why she did not recruit mostly male neurotypical participants in order to make the gender ratio comparable.  However, she said that the age and region of origin judgments made in the study did not differ by gender among neurotypicals in previous studies.


The Results
Parts 1 and 2.
Autistic and neurotypical adults turned out to be equally accurate at estimating talker age; if anything, ASD participants performed a little better, with slightly less individual variability. 

Parts 3 and 4.
The experimenters expected that ASD adults would make no consistent judgments based on the speaker's age.  The actual pattern was a bit more complex.

Neurotypical participants' judgments of friendliness, reliability, intelligence, and success were not affected by speaker age.  ASD participants showed a similar pattern for friendliness, reliability, and success.  However, they perceived older participants as more intelligent.

Interestingly, in the regional experiment, ASD adults rated all speakers as friendlier, more reliable, smarter, and more successful than did neurotypical adults.  The experimenters did not attempt to explain this effect, which could be interpreted as greater charitability for the ASD group.

The pattern of regional judgments differed in ASD and NT groups, but not in the way predicted.

Neurotypical participants viewed speakers of the local (Midlands) dialect and Southerners as more friendly than other speakers.  They also viewed speakers of the local dialect as more reliable than New Englanders.  They perceived local and Midlands speakers as smarter than Southerners, and local and Northern speakers as more successful than other speakers.

ASD participants' ratings of speaker friendliness, reliability, and success were not affected by dialect.  However, they perceived local speakers to be smarter than the rest, and Southerners as less intelligent.

In short, these ASD participants did not lack a general awareness of regional stereotypes.  Instead, they appeared to have a qualitatively different set. 

The Choice of Interpretation
The researchers started with the assumption that ASD individuals excel at "low-level," more perceptual judgments, while they would have difficulty with "higher-level" ones, especially those with social content.  They assumed that discriminating speaker age and region of origin were low-level and non-social processes, while age and regional stereotyping was a high-level, social process.

However, if you think about what it takes to determine a speaker's age and region of origin, it's a lot more high-level than even the most complex "high-level" judgments made in most autism auditory processing studies.  Different individuals produce the same sound (e.g., "d") slightly differently, and various consonants (such as "s") may be pronounced differently depending on what vowel follows.  Cues that autistic people are known to perceive extremely well--such as pitch--may not be the essential ones in these tasks.  Furthermore, the subset of participants who were in both study 1 and study 2 had to switch from attending to the cues that mark speaker age to the ones that mark speaker region--a sophisticated and arguably "high-level" switch.

Furthermore, as Morton Gernsbacher pointed out, age and region of origin judgments are, in a sense, social in nature.  They are learned with experience and exposure to speakers of a variety of ages and regions of origin (the latter judgment, in particular, may be learned later in life if the listener comes from a low-population area without a lot of speakers from other regions).  Listeners likely do rely on information from others, at least in part, to learn the associations between the relevant auditory cues and speaker characteristics.

To their credit, the researchers did not ignore the ASD listeners' enhanced performance at identifying speaker age and regional accent.  Nor did they explain this strength away as a deficit.

However, the researchers lacked an adequate explanation for either of the most striking results of the study: 1) ASD listeners' overall higher ratings of speaker characteristics, or 2) ASD listeners' qualitatively different emphasis on intelligence.

When asked about ASD participants' overall higher ratings of speaker characteristics, and whether this indicated the ASD group was more charitable, the first author's response was noncommittal.  She said the result was likely a characteristic of the particular small sample she used rather than ASD individuals in general and that further, her ASD group may have simply had a different bias regarding how to use the rating scales in the study.  To me, neither answer seems adequate.  First of all, if ASD participants' ratings weren't reliable on this matter, why would any of the other study results be meaningful given that the sample size and ratings scales were the same?  Second, if ASD adults really respond differently to a basic Likert rating scale, this is a serious problem because huge numbers of studies on a variety of different topics use these scales.  If ASD adults have a different response bias, this could jeopardize the validity of any self-report scale used with them.  If this is really true, it should have been mentioned at the outset. If it's even a possibility, rigorous pilot studies should have been done to determine how ASD adults respond to the rating scales.  If the authors thought of this problem before obtaining results and did not investigate it, this is a major methodological weakness.  More likely, it's an ad hoc explanation.

This research team had not predicted that ASD individuals made judgments about speakers' intelligence, much less ones that neurotypical adults did not themselves make.  Their explanation was roughly that, because of functioning labels, intelligence is highly salient in the life of an autistic person, whereas friendliness, stability, and success are not.  The first part seems possible, but the second part does not.  While an autistic person may have difficulty determining when a person is being friendly (rather than, say, teasing them), surely for a person vulnerable to bullying and abuse friendliness would be more salient than normal.  Similarly, an autistic adult struggling to maintain employment and be financially independent--where the odds are overwhelmingly stacked against disabled adults and particularly autistic ones--would scoff at the idea that success is not salient to them.  To be fair, I'm not sure myself what an adequate explanation would be.  Perhaps you have some ideas?

The speaker also simply assumed that failure to make stereotypes about the speakers based on age or region of origin meant an unawareness of the stereotypes, or an inability to make such judgments.  The experiment did not control for the possibility that ASD adults were aware of stereotypes but deliberately chose not to use them.  Yet one might expect the latter given that ASD adults are a minority subject to negative stereotypes, and thus likely unusually attuned to the dangers of stereotyping.  The speaker, who appeared not to expect such a question, replied by saying that ASD adults were enrolled in various social skills intervention programs and may have learned not to stereotype there.  An odd answer, given that social skills programs are more likely to teach how and when to stereotype than explore the why or the ethics behind it. Her answer also reveals an assumption that ASD adults can only display positive social behavior if taught in a formal program, which is unproven and insulting, to say the least.

In short, this talk revealed some interesting information about autism, but a lot more about the current state of autism research.
1) Autistic adults have an unanticipated strength in identifying speakers' age and regional accent, performing slightly better than neurotypical adults.
2) Researchers now appear more willing to report autistic strengths as such.
3) The Weak Central Coherence Theory has not fallen out of favor after all, and some researchers still have yet to hear of the Enhanced Perceptual Function theory.
4) Participant matching is still an issue, even for relatively uncomplicated matters such as gender.
5) Interpretations of studies still involve untested assumptions, biases, and ad hoc explanations for unexpected results.