Text to design (TTOD)

Abstract

The processing, understanding and manipulation of natural language text has been a difficult milestone for years, and it is time that natural language becomes a medium by which users can design programs. Through a background and literature review, we investigate the linguistic principles and theories that govern natural language processing and information extraction. Using a myriad of technologies, unrestricted English text is converted into a form which can be processed for information. Ambiguity problems are tackled using a multi-pronged approach including part-of speech tagging, dependency parsing and machine learning algorithms such as an artificial neural network and a decision tree. Information is extracted by cleverly traversing grammatical dependencies. Furthermore, a set of rules is used to define how the system handles different inputs. The system ultimately delivers a UML diagram depicting the information written in the text. The results indicate that the employed methodology is effective for limiting the impact of ambiguity. In this regard, neural networks prove to be more accurate than decision trees. Moreover, a substantial amount of information could be extracted from text. However, its lack of training was a major flaw. It was concluded that while it had adequately proved that the concept was possible, the resultant system was too immature for real-world use, and while promisingly effective, it required more training data to be implemented in the future.