廣義知網是以「個體–關係」為框架的知識表達模型，我們希望藉著這個表達模型，使詞彙與詞彙間的自動語意合成機制得以建立，從而幫助計算機由了解詞彙進一步發展到了解文意。為達成此目標，本文以比較詞與比較結構為例，討論比較詞的分類、定義以及比較句的合成等等問題。比較詞本身並沒有核心語意，而須在一個結構中才能顯現它的意義，因此我們參考框架網(FrameNet)所提出的比較事件框架，以其事件框架元素(frame element)作為定義比較詞的特徵。然後提出一個對應表，使比較句成分能一一對應到比較詞的語意特徵中，亦即使句子的表面結構轉換為一種深層的語意結構，達成計算機理解語言的目標, In this paper, we propose an approach for studying the semantic representations of comparison words and comparative constructions based on the framework of E-HowNet. Our objective is to determine how the semantic composition mechanism works. The proposed method establishes a mapping between grammatical structures and fine-grained event structures for comparative constructions. We encode the event structures of comparative constructions as part of the representations of comparison words. The fine-grained semantic roles are adopted from FrameNet. A semantic composition mechanism is then developed to unify word sense representations under syntactic constraints. Our ultimate goal is to achieve fully automatic semantic composition.