Distributional Learning of Appearance
The left column illustrates an example identification task, the right an example naming task. In both tasks the aim is to choose a correct pairing of a word and an appearance over an incorrect one. The plots in the upper row show data for the correct pairings, those in the lower for the incorrect. Each scatter plot relates to a different pairing of word (W) and appearance (A). For example, the top-left plot relates to , the pairing of the word beans with the appearance of beans; while the bottom-right plot relates to , the word snake with the appearance of piano. x is a variable ranging over the words of known appearance. Each scatterplot has a point for each possible value of x. The horizontal coordinate of the points indicates , the word similarity; and the vertical coordinate indicates the appearance similarity. For both axes, values nearer the origin indicate greater similarity. The correlation of the points in a plot is visually indicated by the green covariance ellipses, and nearby them we give the correlation value. In both examples, the more positive correlation is in the upper row, so the correct pairing is identified.