| |
Abstract:
Several methods for componential analysis were applied to
high-density (128-channel) ERPs. Data were collected from 57
subjects using an N400 sentence paradigm with mid-sentence and
sentence-final semantic anomalies. A conventional (temporal) PCA
was initially applied to each of the four conditions, and promax
rotation was used to derive a simple structure for the non-noise
factors. Each of the rotated factors was plotted over channel
space, multiplying factor scores by channel loadings and averaging
across subjects. The resulting "component waveforms" were
visualized in waveplots and topographic maps. Results showed a
common set of factors across conditions, although several effects
(N1/P2 and "P1-reprise") were misallocated, suggesting the need for
separation of spatial components. The strongest factor across all
conditions was a centromedial P300, which was delayed to the
incongruous sentence-final word (peak latency difference of 40ms).
In addition, we found an N400 factor with a centroparietal
distribution in all but the congruous-final condition, suggesting
that the P300 factor alone was insufficient to explain the
difference between congruous and incongruous words. To address
misallocation of variance and P300 latency shifting, we are
currently comparing spatiotemporal PCA and trilinear decomposition.
In TLD, subjects are treated as an explicit variable, and
additional constraints permit estimation of a unique solution,
unlike traditional factor-analytic techniques.
|
