All plots for the trajectory starting from the flag word.
Prototypes were obtained from data with Gaussian synaptic noise (var 0.02) but without the accommodation mechanism.
A particular prototype for a given word is computed as the mean of the last 100 trajectory points (170 overall), all of them contained in the attractor basin of this word.
Dissimilarities are computed using cosine metrics (for details see metrics in pdist function).
Dendrogram
Three different methods were used (average, single, complete) for dendrogram generation. Methods differ from one another in how they measure the distance between clusters.
The first two methods (average, single) show clearly two main clusters, which correspond to abstract (top cluster) and concrete (bottom) sets of words.
Average - unweighted average distance (UPGMA)
Single - shortest distance
Complete - furthest distance
MDS
2-dimensional MDS for combined prototypes and attractors of the flag word.
The red line connects attractors in correct order (i.e. with respect to sequence of system transitions), it starts from the attractor basin of the flag word (which is the first attractor the system falls into).
MDS parameters (see mdscale function for details):
- dimensions: 2
- criterion: sammon
- start: cmdscale
- metric: cosine
Prototypes for concrete words only
All prototypes
