PAT-trees with the deletion function as the learning device for linguistic patterns

Abstract

In this study, a learning device based on the PAT-tree data structures was developed. The original PAT-trees were enhanced with the deletion function to emulate human learning competence. The learning process worked as follows. The linguistic patterns from the text corpus are inserted into the PAT-tree one by one. Since the memory was limited, hopefully, the important and new patterns would be retained in the PAT-tree and the old and unimportant patterns would be released from the tree automatically. The proposed PAT-trees with the deletion function have the following advantages. 1) They are easy to construct and maintain. 2) Any prefix substring and its frequency count through PAT-tree can be searched very quickly. 3) The space requirement for a PAT-tree is linear with respect to the size of the input text. 4) The insertion of a new element can be carried out at any time without being blocked by the memory constraints because the free space is released through the deletion of unimportant elements. Experiments on learning high frequency bigrams were carried out under different memory size constraints. High recall rates were achieved. The results show that the proposed PAT-trees can be used as on-line learning devices.