Developing Intelligence

Although grammar is usually considered the “uniquely human” aspect of language, and the capacity to use primitive symbols is thought to be common among primates, high-level cognition is nonetheless strongly impacted by the use of symbols. For example, symbols can help in putative inhibition tasks; labeling a stimulus-response relationship can result in increased proactive interference when the correct response changes; and providing infants with unique object labels allows them to demonstrate knowledge they wouldn’t otherwise demonstrate for several months.

One hypothesis about the role of symbols – and in particular, verbal labels – is that they allow a complex neural representation to become more durable or stable. For example, the word “fun” allows us to refer to a huge variety of experiences and emotions, and manipulate that concept as though it were a single entity.

This hypothesis was directly investigated by Lupyan in a recent article. In this study, subjects were told they were visiting an alien planet, populated by “approach” and “escape” aliens. Their job was to learn which kind of aliens to approach, and which kind to avoid; the “aliens” themselves were YUFOs – akin to “greebles,” YUFOs are a set of life-like stimuli that have been extensively used to test perceptual discrimination.

In the task, a subject might see one of these “aliens” appear at the bottom of the screen. The subject would then press “up” (if the subject thought it was an escape alien) or “down” (if the subject thought it was a approach alien). After pressing a button, auditory feedback was given (positive or negative as appropriate). Finally, for subjects in the label condition, a word appeared then on the screen: for example, the “approach” aliens might be followed with the word “leebish,” while “escape” aliens might be followed with the word “grecious.”

After 144 training trials, subjects participated in testing trials which involved them categorizing each alien as approach or escape, without feedback (and without the presentation of labels for those in the label condition). In addition, extra alien exemplars from each family of YUFOs were included in the test trials.

The results showed that subjects who had experienced written labels during training did better overall, and improved more quickly than those who had not. This result is particularly surprising considering that the labels were not technically providing any additional information: the labels were redundant with auditory feedback tones which categorized each alien as approach or escape. In addition, those who had learned with the help of “labels” were better at sustaining their performance during the test phase than those who had not learned with labels (although this effect could be due to lower performance of those who did not learn with labels, rather than a steeper forgetting curve).

The author concluded that labels allow a more discrete representation (“leebish vs. grecious”) to structure the difference between the two kinds of stimuli, whereas without labels the distinction is a matter of degree: “slightly more rounded & larger vs. less rounded & smaller.” In mathmatical terms, labels may deepen the attractor basins for the neural representations of each alien, thereby stabilizing or discretizing patterns of neural activity that otherwise might be less durable.

But there are simpler explanations. Although this effect may result from “labels” per se, it may actually result from lower task demands in the label condition: those subjects didn’t need to remember what their response had been in order to categorize that alien. This alternative theory is supported by additional work from the same author, which indicates that pictorial, nonverbal “labels” are statistically no less effective in this task than written verbal labels. This hypothesis could be tested with a condition in which feedback was provided visually, and “labels” were presented as high and low auditory tones.

There are even more interesting alternatives. First, consider that YUFOs were originally constructed to engage the same perceptual mechanisms involved in face processsing – which is known to engage the right hemisphere more than the left. In contrast, linguistic stimuli typically engage the left hemisphere more than the right. Putting these two facts together, the apparent efficacy of labels (generally, left > right hemisphere) with YUFOs (generally, right > left hemisphere) might result from improved inter-hemispheric communication. This hypothesis could be tested by presenting high & low tones exclusively to the right ear, or determining whether nonverbal pictorial labels show more benefit when presented to the right than the left hemifield.

Finally, even if changes in hemispheric interaction or task demands are not to blame, labels may help because they can be rehearsed in a temporary phonological store (if this concept is unfamiliar, see the Baddeley Working Memory model). If this is the case, a simultaneous articulatory suppression task (i.e., “repeat the words ‘coca-cola’ throughout the alien task”) should remove the benefit of labels.

Related Posts:
Generalization and Symbolic Processing in Neural Networks
Word Learning in Feature Space
Verbal Labeling and Proactive Interference
Labels As an Accelerator of Ontological Development