This animated GIF shows an agent-based model animation in two 2-D DCT F2 planes

Shown are two panels: left the agent's memory distribution in the DCT-0 (height) vs. 
DCT-1 (slope) plane of the second formant trajectories for the vowel /i:/, /ju:/ and /u:/;
right the same for DCT-0 (height) vs. DCT-2 (curvature).
Memory data of young speakers are shown on top, the data of older speakers below
Each dot represents a memory token of an agent (i.e. the memorized F2 trajectory); the 
black dots are the memory tokens at the begin of the simulation; the red dots are updated
memory tokens.
The simulation runs over 80000 random interaction between 11 young and 11 older agents; 
each agent carries a constant set of 6 word types containing the vowels /i:/, /ju:/and /u:/ 
with 10 tokens each.
Plots are updated after 1000 random interactions; after 80000 interaction the animation starts 
anew.

Watch the red dots in relation to the black. What you see here:
The trajectories of /i:/ do not differ very much in the initial state (black) and stay stable 
for both speaker groups (red, left panels).
The trajectories of /ju:/ for the young speakers remain quite stable (upper mid), while for the 
older speakers the distribution moves into the area of the younger speakers' group (lower mid).
The trajectories of the /u:/ spreads out a little for the younger speaker but stays in the same 
location; the trajectories of the older speakers move into the area of the young speakers
(right panels).