Behavioral and anatomical consequences of early versus late symbol training in macaques

Whew, another exceptional article!

This time, researchers at Harvard trained juvenile and adult monkeys to tell the difference between all ten numerals (0, 1, 2, 3, 4, 5, 6, 7, 8 and 9) and 16 letters of the latin alphabet (X, Y, W, C, H, U, T, F, K, L, N, R, M, E, A and J).  Each symbol was associated with an extra drop of juice.  For example 0 meant you got no juice, 1 meant 1 drop of juice, 2 meant 2 drops, all the way up to 9.  Then X meant 10 drops of juice, Y meant 11, up to J which gave a full 25 drops of juice.  They would see two letters on the screen and have to pick the one that gave them the most juice.

The monkeys trained on this task every day for two years.  They also trained on a similar task where the amount of juice was represented by dots on the screen (more dots: more juice).  They did very well on both tasks, but it's interesting to see that they don't do as well when the difference between the number of dots was proportionately smaller (ie: they're less accurate on 24 vs 25 dots than 2 vs 3 dots) but they don't show the same pattern when the juice reward is represented by symbols.

While it's been shown before that monkeys can learn this type of task (which, lets remember, contains symbols that monkeys would never naturally encounter) a major find of the paper is that juvenile monkeys are way better than the adults at learning symbols.  And interestingly they're both just as good as each other at learning the dot task.

Not satisfied with just training these monkeys on a task for two years, the researchers then scanned them using fMRI.  They passively viewed the letters and numbers they'd learned, Greek letters they'd never seen before, and faces.  You can see in the figure below that there are face and symbol selective regions in both juvenile and adult monkeys, but only the juvenile monkeys showed a "learned symbol" region: an area that responded better to the letters they knew than those they had never seen.

Figure 4. Category selectivity maps of two adult symbol learners and one adult who was not trained in the symbol task, for learned symbols, untrained shapes, or faces. Color scale indicates t score. Voxels that responded significantly more to Faces than to Shapes (conjunction of Faces > Learned symbols AND Faces > Untrained shapes) are indicated in red; the three largest and most consistent face-selective regions are labeled f1, f2, and f3 and are the loci used to calculate the corresponding percent activations in Figure 7 and the time courses in Figure 8. Voxels that responded significantly more to Shapes than to Faces (conjunction of Learned symbols > Faces AND Untrained shapes > Faces) are indicated in green; the three largest and most consistent shape-selective regions are labeled s1, s2, and s3. No regions in any of the adults showed more activation to Learned symbols compared to Untrained shapes (blue).

Figure 5. Category selectivity maps of three juvenile symbol learners. Conventions as in Figure 4. All three juvenile learners showed regions that were activated more by Learned symbols compared to Untrained shapes (L > U AND L > F), in addition to the pattern observed in adults of alternating regions relatively selective for Faces or for Shapes.

The researchers conclude that juvenile brains are both better at learning a task, and that the mechanism through which this behavior emerges is a clustering of neurons during development which become tuned to the same stimuli.  They argue that "intensive early experience drives the generation, or segregation, of domain-specific modules".  They suggest that it is an innate property of the brain to be able to cluster similar neurons together, and that this clustering is beneficial to behavior through cortical inhibition: when one area is working hard, the neighboring regions are inhibited, which allows modules to form clear boundaries between the stimuli they do and do not respond to.

The paper represents a very exciting contribution to the mechanisms through which we learn and the ways in which they differ between children and adult.  There was a huge amount of work in this paper, and I'm suitably impressed :)

Behavioral and anatomical consequences of early versus late symbol training in macaques.

Srihasam K, Mandeville JB, Morocz IA, Sullivan KJ, Livingstone MS.

Neuron. 2012 Feb 9;73(3):608-19.

PMID: 22325210