Submitted for presentation at 2005 meeting of the Society for Neuroscience

Error profiles and working memory representations of sequences during imitation

Y. Agam1, D. Bullock2, R. Sekuler1,2
1Volen Center for Complex Systems, Brandeis University, Waltham, MA, USA
2Cognitive & Neural Systems Department, Boston University, Boston, MA, USA

To probe the neural mechanisms by which multiple actions are represented and sequenced for production, we varied an imitation task in which human subjects viewed unfamiliar, quasi-random, 2D motions of a disc and then used a stylus to reproduce each novel trajectory from working memory (WM). The disc motions were composed by end-to-end concatenation of equal-length line segments, whose orientations differed enough from one another that it was possible to distinguish successive segments in the resulting imitation.

A psychophysics-based algorithm decomposed each imitation into segments whose orientations were compared to corresponding stimulus segment orientations. As the number of segments in the stimulus grew from 3 to 7, imitated segment orientations became less accurate, and segments were more likely to be omitted. When segment-wise imitation error was plotted as a function of serial position, results showed a strong primacy effect (error rose with serial position) and a modest recency effect (somewhat less error at final position). Also, segment lengths were longest for final segments. Comprehensive transposition analysis showed that most exchange errors arose from swaps of adjacent segments. Repetition of a segment orientation within a model reduced imitation quality.

These results --finite WM; primacy, recency and repetition effects; preponderance of adjacent-item transpositions; longer final segments-- are consistent with expectations of competitive queuing models. Because the results replicate trends seen for immediate serial recall of words, this multi-segment imitation paradigm may be useful for probing a common primate substrate for sequence representation. However, results for longer sequences suggest that the segment-by-segment imitation generator may be supplemented in the final phases of imitation by an error-driven overlay to enable a late-course correction. This points to configural effects that may have no precise equivalent in the verbal domain.

Supported in part by AFOSR, NSF and NIH.