Abstract: With new words appearing continuously, many unknown words must be dealt with when we compute the real text. The author holds that, based on the known words in the dictionary, the attributes of every wordformation can be construed, such as the position in the word building, the wordbuilding mode of productivity, the syntactic and semantic requirements of wordbuilding components in different positions, and the meaning understanding mode of the combination. In this way, we can assist the compute to discriminate and understand unknown word correctly.

Key words: unknown word, principle of compositionality, double bytes