DOUBLE BRANCH DECISION WOOD BUILDING

摘要

PURPOSE: To construct a binary branching decision tree by adding one word to a list at a time and deleting one word from the list at a time and continuing this processing without increasing entropy until any word cannot be added/ deleted. CONSTITUTION: A decision tree has a true or false (namely, binary) question in each node and has a probability distribution over each leaf. The processing is started in the root of the tree and answers the question encountered in each node and advances to a first or second branch in accordance with 'true' or 'false' of the answer and goes toward one leaf. The question in each node is described in relation to available data (for example, already spoken words) and is selected so as to ensure that the probability distribution of the leaf gives information related to the next word as much as possible. Thus, the binary branching decision tree which starts from the root to reach leaves and has the true or false question in each node is designed and constructed.