Linguistic Duality

摘要

Currently, the approaches from the results of word or document vectorization significantly achieve their success in large-scaled language processing, such as high-quality neural network machine translation trained on a large-scale parallel corpus. The data driven approach with minimum concern in language knowledge efficiently captures its syntactic information. However, there are some deficiency occurred in capturing the semantic information, such as in the similarity measure when performs on both sides of the language interpretation. We propose an add-on process to correct the similarity measure of the concepts defined in WordNet. Similarity measure is the fundamental procedure for language processing highly depending on the context integration and its representation method.