Modeling Multidimensional Language Matrices to Learn Predictive Text

Anonymous

Abstract: The predictive text in the tray bed of the Chinese typewriter presents “Radiating style” and other important patterns, which reflect the main properties of the Chinese language. For a robot to understand these patterns like human’s “once glanced, never forgotten”, we construct multidimensional language matrices (MLM) to present the characters and/or words of predictive text for Chinese Natural Language Processing (NLP). Using 2D LM, our approach identified the core character as the prefix of radiating outward words, and as the suffix of radiating inward words to show the best distribution of the characters in a nine-grid. Using 3D LM, our approach, for robots doing as human, recognized the meaning and location of the words in a nine-grid by “Once learning mechanism”. Even though these approaches are proposed for the Chinese language, their methods are extendable to other languages.