This report describes the completion of a project to develop a functioning neural-network-based machine translation system. Our system architecture consists of a sequential neural network to produce a rough unordered translation, and a pair of RAAM encoding networks linked by a backpropagation mapping network to remap the rough translation into the correct sequence. Performance was very good on the training data (99% perfect overall), but was poor on the testing data (only 25% perfect overall). The sequential and mapping networks had poor generalization performance, while the RAAM networks performed well overall. Some possible methods for improving the performance of the sequential and mapping networks are discussed, and if these are implemented the system may become more practically useful.

PDF

Technical Report HR-93-01, Department of Computer Sciences, The University of Texas at Austin.

@techreport{bednar:ugthesis93, title={Multi-Level Neural Network Language Translator}, author={James A. Bednar}, number={HR-93-01}, school={Department of Computer Sciences, The University of Texas at Austin}, institution={Department of Computer Sciences, The University of Texas at Austin}, type={Undergraduate Honors Thesis}, url="http://www.cs.utexas.edu/users/ai-lab?bednar:ugthesis93", year={1993} }