Acoustical Society of America
158th Meeting Lay Language Papers
[ Lay
Language Paper Index | Press Room
]
'Autistic'
Traits Predict Individual Differences in Speech Categorization
Dan Hufnagle - hunagle@cmu.edu
Department of Psychology
Carnegie Mellon University
Pittsburgh, PA 15213
Lori Holt
Department of Psychology
Carnegie Mellon University
Pittsburgh, PA 15213
Popular version of presentation 4pSC7
Presented Thursday afternoon, October 29, 2009
158th Meeting of the Acoustical Society of America
The human visual system processes the external world in ways that are essentially illusions. We use context to make assumptions about the properties of the things we perceive. For example, when comparing the brightness of two objects in shadow, we mentally compensate for the darkening caused by the shadow. Similarly, an object will seem to be smaller or larger depending on what we think we know about the sizes of objects around it.
The effects are not limited to vision -- context influences sound perception as well. A person who hears a sound that, on its own, is entirely ambiguous between the sound "g" and "k" will categorize it reliably as "g" if it occurs before "ift" and reliably as "k" if it occurs before "iss." Our perceptual system is biased to perfer words like gift and kiss over non-words like "kift" and "giss." Perceptual illusions like these help us to find structure and organization in the external world, even when sensory input on its own is not precise.
Our primary goals were to measure individual variability within the influences of context on perception and to relate this variability to autistic traits. In our experiment, people heard a sound and judged whether it was "da" or "ga." We established that context that is temporally closer to the speech target is weighted more heavily than context that is further away from the target. While this finding is novel, it is not entirely surprising. Analyses demonstrated that while most people weighed nearby context more than distant context, others did not. Moreover, a person's contextual weighting was related to his or her level of autistic traits, yielding clues about the nature of auditory categories and suggesting a way forward for research investigating perception in autism.
There are good reasons to suspect that individuals differ in how susceptible they are to these kinds of perceptual processes. Furthermore, differences in perceptual processing characterize some disorders, including autism. Autism is a spectrum disorder, and the range of severity means that some people with autism lead lives characterized by profound disability while others are very high functioning. Characteristic traits of autism extend, to a lesser degree, to people without autism as well. These characteristics can include difficulty with the ability to shift attention, a tendency to focus on small details, or problems with interacting in social situations.
We were interested in how these traits relate to a particular auditory illusion in which the pitch of a sound's context influences listeners' perceptions of what that target sound is. A sound that is ambiguous between "da" and "ga" will be reliably perceived as "da" when it follows lower-pitched tones and reliably as "ga" when it follows higher-pitched tones. Context tones between these high and low pitches lead to ambiguous perception. The process also works with speech: speech that falls within a certain pitch range will bias the perception of other sounds. The purpose of this work is to demonstrate variability in auditory perception in a non-clinical population and to link that variability to a specific set of traits that would provide us with insight into ways to examine perception in people with autism.
In the experiment, people heard a series of tones followed by a target speech sound that ranged from "da" to "ga". The sounds were manipulated so that some sounds were ambiguous (they sounded a bit like "da" and a bit like "ga") and other target sounds were unambiguous. People performed hundreds of trials so that we would have a detailed record of their speech categorization judgments in each ambiguity and context combination.
Figure 1. Spectrogram of a trial showing pitch and timing of sounds
We chose to investigate variability in speech categorization by looking at heterogeneous contexts in which some context elements use high pitch tones, some low pitch tones, and some middle pitch tones. By mixing these tone types, we can see whether listeners give processing preference to context that is closest to the speech target. That is, a context that uses high, middle and then low tones could be used in two ways. First, people may simply take an average of the entire context when using it to help process the speech target, with the resulting average middle context not influencing perception. On the other hand, people may give more weight to the context tones that are closest to the speech target, which would mean that different speech categorization results would depend on whether the low tones or the high tones were nearest to the speech target. Our results supported this second hypothesis: categorization responses to identical speech targets differed depending on whether contexts were high-medium-low tones or low-medium-high tones. Therefore, people generally weight temporally closer context more strongly than temporally distant context.
Figure 2. General effect of context pitch. All means are significantly different. Bias for temporally closer context is established by the difference between LMH and HML contexts.
The finding that the weight that people place on variable context weakens the further that context is from the target is novel, but it is not terribly surprising. However, our purpose was to understand individual differences in this use of context. We correlated two measures from each individual: the difference between categorization in high-medium-low and low-medium-high contexts (local context bias) and a score on the Autism Quotient questionnaire, which measures a person's autistic traits. For sounds that were intermediate between "da" and "ga", this correlation was a strong -0.53. This correlation indicates that the fewer autistic traits that an individual has, the more he or she will weight nearby context relative to distant context when categorizing an ambiguous sound. Conversely, the more autistic traits a person has, the less the difference in weighting near and distant context.
Figure 3. Scatterplot showing each individual's bias for temporally closer context and Autism Quotient score. The trend line shows a correlation of -0.53.
These findings offer some insight into how people use context to process sound information, but they yield as many questions as answers. One important question is whether these differences are due to differences in perceptual processes themselves or whether the source of these differences are due to qualitative differences in the auditory categories that we are testing. This research also suggests some lines of further research that investigate the influence of context of perceptual processing in autism.