Relation words are of great significance for analyzing the semantic relations between clauses in multiple corn pound sentences. But in the process of automatic identification of relation words based on rule, we found that not all the relation marks in multiple compound sentence are relation words, and identifying the relation marks serving as relation words is the key point. An algorithm was proposed to identify typical relation words which are adjacent. By combining the relationship marked corpus and the extraction of relation markers, this algorithm uses the feature to identify these markers...