Cross-Language Record Linkage Across Humanities Collections Using Metadata Similarities Among Languages

摘要

This paper proposes a method for cross-language record linkage across digital humanities collections by exploiting similarities between metadata values in different languages without using any translation method. Our method represents metadata values in Japanese and English as vectors by using monolingual word embeddings. Then, we calculate similarity between metadata value vectors by learning a mapping between vector spaces that represent Japanese and English. The proposed method could help users to acquire multilingual information of the objects in digital collections. We evaluate the effectiveness of our method on Japanese Ukiyo-e print databases in Japanese and English.