In summary, this paper has demonstrated that the method adopted by Ellsworth et al. in their frame-based contrastive text analysis, which builds on the existing FrameNet convention to analyze only the semantics of frame-evoking predicates, is not suffi cient to describe complex interactions between lexicon and grammar. Such interactions should be recorded and
accounted for, in order for us to understand how languages encode the same scene differently and still allow us to come up with comparable construals of the scene, no matter in which language we read. This paper also proposed a way to represent interactions between lexical units and grammatical constructions, as an extension of the current FrameNet annotation methodologies. By addressing the necessity for representing interactions between lexicon and grammar in English and Japanese FrameNets from the viewpoints of contrastive text analysis, it is hoped that this paper will contribute to the development of "Constructicon" proposed by Fillmore 2006.