Abstract: G-anaphora disambiguation between sentences has become an important issue in Chinese information processing currently.Based on semantics and pragmatics，the anaphora phenomenon in legislation texts is formalized in the multi-levels model，which can be used in Chinese sentences group G-automatic processing.Antecedent and reference expression can be linked by concept primitive，57 basic sentence patterns provide framework to identify the relations，and context unit contains the macro knowledge for them.The analysis of corpora indicates that the model is efficient in G-anaphora disambiguation between sentences.

Key words: anaphora disambiguation, formalization of context, semantics chunks share, natural language understanding

摘要： 句间回指消解是当前中文信息处理的一个重要研究课题，直接从语义和语用入手，以法律文本为语料来源，对句间回指进行形式化描述和消解，服务于计算机句群自动理解。概念基元是“显微镜”，看清指代语与先行语的微观语义联系；句类是“放大镜”，将指代语和先行语纳入57组基本句类中进行关联；语境单元则是“望远镜”，为指代语和先行语提供宏观的语境知识。语料考察结果表明，这一多层次的消解模型对实现句间回指消解是有效的。

关键词: 回指消解, 语境形式化, 语义块共享, 自然语言理解