Neural Associative Memory with Finite State Technology

Morphological learning approaches have been successfully applied to morphological tasks in computational linguistics including morphological analysis and generation. We take a new look at the fundamental properties of associative memory along with the power of turing machine and show how it can be adopted for natural language processing. The ability to store and recall stored patterns based on associations make these memories so potentially valuable for natural language processing. A neural associative memory recollects word pattern based on an input that resembles the input word pattern, which is sufficient to retrieve the complete pattern. Morphological parser and Turing lexicon are used to process regular inflectional morphology. The rules of unrestricted grammar are extracted from the morphological parser. Turing Lexicon processes input word using the unrestricted grammar. The associative learning algorithm is a classification method based on Euclidean minimum distance for stem or attribute. The stems and attributes are indexed in the associated memory for quick retrieval. The associative memory is used to recall correct and incorrect word patterns. A machine is designed to implement the morphological process as a special case by achieving both concatenative and non-concatenative phenomena of natural languages. The experimental results shows that the proposed approach has attained good performance in terms of speed and efficiency.