The invention discloses a machine translation method and system. The method comprises the following steps: S1) loading dual-language or multi-language keyword data, reading each keyword and a translation thereof, identifying different language types by utilizing language identification, and converting each keyword into a natural language variable to form a natural language variable library; S2) carrying out word segmentation to a sentence to be translated to divide the sentence to be translated into a plurality of keyword groups; S3) according to the setting of a user, adding the language identification of the language type which needs to be translated and output after each keyword to form a keyword sequence; S4) replacing each element in the keyword sequence with a variable address to generate variable typesetting language; and S5) compiling or explaining the variable typesetting language to form the translation. The invention has the advantages of being practical, high in translation efficiency and favorable in translation stability.