Abstract: The calculation of the sentence semantic distance acts a base role in many intelligent systems. Based on multi-scale analysis,a multi level semantic distance calculation framework was proposed. All sentence pairs were filtered by word-level semantic distance algorithm first, and then syntax parsing and semantic parsing were executed for the sentence pairs which semantic distances were below the threshold. After getting the standard semantic frameworks, the core conceptions in the frameworks were compared then. The final semantic distances of the sentence pairs passing the second level filter were obtained using isomorphism-based semantic distance algorithm,which could dynamically adjust its weights. Experiments show that the total precision of the method reaches 73.3%. For the cases, it has higher relevance, reaches 91.4%，similar to the semantic-level algorithm.

Key words: Semantic distance, Multiscale analysis, Word-level, Semantic level