Our Word-Morph Network

Do you remember the old word game in which you are asked to change the word 'cat' into 'dog' by exchanging one letter at a time, only using real English words? So, with cat and dog it is easier as it may seem: cat-cot-dot-dog. It is not always that easy: for ivy and dry the shortest path is: dry-dey-dee-dye-aye-ace-ice-icy-ivy. So, here, it is not possible to directly come 'closer' to the goal with each exchanged letter. The graph above shows all correct English three-letter words (extracted from the Oxford dictionary), where two words are connected by an edge if they differ in exactly one letter. So, this is the graph in which the word-morph game essentially takes place.

We asked ourselves how people learn to play this game: will they be slow at the beginning and get faster soon? Will they learn the shortest paths between all pairs or do something different? We found out that people learn 'landmark words' through which they tend to navigate. Thus, they only need to learn n many paths in a network with n nodes, instead of learning n*n different paths. Of course, the paths that navigate through a certain landmark word are a bit longer than the shortest paths (in this case by a factor of about 1.6). It thus seems that our human mind tries to optimize both: path length but also the effort to learn about the network. Check out our paper if you want to learn more about this research; we got a best paper award for it from CogSci 2011.

The graph was layouted in Gephi, and coloured by an automatic clustering algorithm. An obvious pattern emerged, consisting of five central groups of words (dark blue, green red, yellow, light blue). Can you guess what these five groups are?