Based on HNC theory, a new method of doing machine translation is introduced in this thesis. It is: To judge how much a computer understands a literal text by ASC criterion, and take the semantic chunk as the basic unit of translation. Two kinds of alignment are applied in order to ensure the validity of the translation, and they are alignments from sentence category to sentence category and from semantic chunk to semantic chunk.
Semantic chunks are divided into Eigenchunk and Generalized Objectchunk, from its composing, a semantic chunk can be distinguished from whether it has sentence-degeneration composing or not. Different means are applied to process different semantic chunks and different composing. The particularity of Eigenchunk makes it become a determinative facter in translating other chunks. The correlation among semantic chunks make every chunk gets restriction from other chunks when it is translated.
Correlation among concept is the theoretical base of search°™test method, search°™test method is applied throughout word-selection.
As for sentence-degeneration composing, the composing of element sentence-degeneration, the distributing of its formats, and its translation are maily discassed.
Key words: Natural Language Processing, HNC theory, Machine Translation, Sentence Generation, Word-selection