The more I think about machine learning the more I realize the importance of finding similarities by using analogies as a way of learning.
If I want to categorize words into hierarchical tree this method would work I think, if not tell me why?
Find two sentences that contain same words but their order is different and some words can also be different.
Find another two such sentences.
Evaluate strength of analogy between these 4 sentences.
The stronger it is the more probability that words mean similar thing, that they belong to same category.
When you learn categories you can do analogies on abstract sentences that contain these found categories and in this manner you can build hierarchy of categories.
For example:
Deer eats grass. ...is to... Man eats deer.
as Cow eats grass is to ?
If we can find sentence Man eats cow. being used in texts that we analyze then we have confirmed that cow and deer belong to category Animals that eat grass and that human eats them.