Date of Graduation
Since the advent of computers, scientists have tried to use the human languages for communication with computers. This pursuit has led several scientists, such as Chomsky to establish that there must be a specific part of the brain adapted to language learning and processing genetically, namely the "language organ". Humans are able to understand and produce sentences that they may have never heard. This leads to the understanding that language must follow some strict rules of processing which is referred to as "grammar of a language". The Artificial Neural Networks (NN) inspired by the neural networks of biological systems have inspired many to attempt to solve many of the pattern recognition problems which have been quite impossible or inefficient to solve with today's computers. Many of the features of NN have attracted a few to application of NN to process natural language. This research applies NN to natural language processing. In this research the ART-1 and backpropagation networks have been used to learn language. First, the intended system learns simple phrases. Using the simple phrases more complex phrases at a higher level are learned. The phrases are used in various combinations to form meaningful sentences. Thematic roles of the various constituents are learned at the sentence level. From the limited number of samples the system can successfully produce the intermediate form of a sentence (deep structure), which it has not encountered. Since the computer languages and commands follow the same structural rules as human language the same process is applied to the most popular Unix "mail" and Unix "vi" editor commands. The deep structure produced from natural language is used to translate into computer commands. Since humans are capable of understanding incomplete sentences, a system is devised using backpropagation NN to correct and produce complete sentences from partially incomplete sentences. The system fails to produce an output if the incomplete sentence is incomprehensible which corresponds to a meaningless sentence or a sentence that has a correct syntactic form but incorrect semantic form.
Moozoun, Mohammad Reza, "A neural network based inference engine generator for comprehension and translation of English and Spanish commands to Unix "e-mail" and "vi" editor." (1997). Graduate Theses, Dissertations, and Problem Reports. 9450.