Deep cases representing the significant meaning of nouns in sentences play a crucial role in semantic analysis. However, a case tends to be manually identified because it requires understanding the meaning and relationships of words. To address this problem, we propose a method to predict deep cases by analyzing the relationship between nouns, verbs, and supplemental words, such as particles, in Japanese sentences. We also propose new deep cases based on a verb thesaurus and a deep case prediction method using a neural network. [For full proceedings, see ED571459.]