Development of multilingual semantic networks (Semantic analysis of ambiguous words and evaluation of translation appropriateness)

Abstract

Words are communication media to share a concept in a community. A word involving ambiguity represents multiple concepts depending on a context. Such ambiguity causes misunderstanding between people having different contexts. On the other hands, a community uses words to obtain responses and/or evaluations from target population, such as customers and participants. The word ambiguity causes misunderstanding between a community and a target population due to different contexts. A community dealing with multiple languages (e.g. multinationals) has a difficulty in translation if there are no words in a second language, all meanings of which do not correspond to all meanings of a word one wishes to translate. To deal with above issues caused by word ambiguity, I propose a multilingual semantic networks(MLSN) framework in this paper. The MLSN is a graph where multiple languages words, as nodes, are semantically linked through concepts, as another type nodes. I implemented MLSN in a graph database with datasets of WordNet in three languages: English, Japanese, and French. With MLSN, I conducted two analysis. In the first analysis, I investigate the meanings of ambiguous words such as “design” and Japanese word “Kansei”, and their semantic relations with relevant words in other languages. I found that there are no words corresponding to all meanings of those words in second languages. For the word “Kansei”, I illustrate semantic relations with words such as “emotion”, “affect”, “feeling”, “impression”, and “intuition” which are often used to define “Kansei”. In the second analysis, I discuss how MLSN supports to select and translate a set of words used as evaluation descriptors. I analyze 10 positive emotion words from well-established Geneva Emotion Wheel and their translation in French and Japanese. I demonstrate how MLSN automatically find translation mismatches and semantic independence between emotion descriptors.

Keywords

• design
• kansei
• emotion
• semantic network
• multiple language
• translation