 Figure 1: Paper surveys consisting of the above photograph and the instructions: (1) "Please draw an "x" at the center of the subject of this image" and (2) "write down a few words to describe it." |
 Figure 2: Scatter plot of locations marked by human subjects normalized to photograph dimensions. |
 Figure 3: Histogram of the horizontal components of the positions chosen by human subjects plot of locations (normalized by photograph width).  Figure 4: Histogram of the vertical components of the positions chosen by human subjects plot of locations (normalized by photograph height). |
|
Our operational definition of a spatial taxon defines it as a object or object group centered at the position indicated. This definition extends the classic definition of figure - ie thing-like and bounded by a shape defining contour,to include multiple object foregrounds. Initially, we expected the ghost/woman stimuli to have three possible states: the black ghost on the top left; the woman on the top right; and the hand on the bottom left. However, subjects chose locations aligned along the lips and used gerunds such as ‘kissing’ to label them. Our cluster analysis was not sensitive enough to measure this. To capture this effect, clusters were segmented manually. The table and bar chart show that the cumulative term count for each spatial taxon are significantly grouped into relevant words. (df=2, x2 = 107.43 p <.01).  Figure 7: Each table shows the most 3 most frequent noun phrases (including synonyms), their corresponding work count and their corresponding count as a percentage sample size for that taxon. |
 Figure 6: Frequency of subject locations classified in a spatial taxon as a function of the rank of that spatial taxon plotted in log-log coordinates. A rank of one, refers to the most common spatial taxon. We chose a log-log coordinate system for easy comparison against the word frequency vs. rank plots shown below. If the visual system’s spatial taxon structure within scenes is information normative with respect to word usage within human natural language, one would expect these curves to be similar. | |
 Figure 8: As explained above, k-means cluster analysis separated the data into spatial taxons. To help the reader visualize how our operational definition of spatial taxon applies to the data, the visual taxons are shown as layers. Scatter plots of the location data for each spatial taxon are overlaid on top of the spatial taxon regions. The bottom layer shows the original image overlaid with a scatter plot of all the data. |
    Figure 9: As explained above, k-means cluster analysis separated the data into spatial taxons based on location. For each survey included in the cluster, the words collected by the second survey question were counted and ranked according to the procedure first introduced by George Zipf (Zipf, George Kingsley (1932): Selected Studies of the Principle of Relative Frequency in Language. Cambridge (Mass). The trend line equations and R-squared statistic are shown in the upper right corners. Though power-law functions provide a good fit for some of the clusters, they do not provide a good fit for others. Check back here for a follow-up analysis. | |
