Abstract  Machine translation services available on the Web  are becoming increasingly popular. However, a  pivot translation service is required to realize  translations between non-English languages by  cascading different translation services via English.  As a result, the meaning of words often drifts due to  the inconsistency, asymmetry and intransitivity of  word selections among translation services. In this  paper, we propose context-based coordination to  maintain the consistency of word meanings during  pivot translation services. First, we propose a  method to automatically generate multilingual  equivalent terms based on bilingual dictionaries and  use generated terms to propagate context among  combined translation services. Second, we show a  multiagent architecture as one way of implementation, wherein a coordinator agent gathers and  propagates context from/to a translation agent. We  generated trilingual equivalent noun terms and implemented a Japanese-to-German-and-back translation, cascading into four translation services. The  evaluation results showed that the generated terms  can cover over 58% of all nouns. The translation  quality was improved by 40% for all sentences, and  the quality rating for all sentences increased by an  average of 0.47 points on a five-point scale. These  results indicate that we can realize consistent pivot  translation services through context-based coordination based on existing services