Complex information needs are hidden in every query that is used to retrieve information requested by users. The non-availability of these complex information needs decreases the accuracy of information retrieval systems. Our proposed retrieval method uses non-relevancy information to clarify users' complex information needs. In our method, the non-relevancy information, which is represented by natural language sentences, is first structured using dependency analysis and heuristics. Next, the structured non-relevancy information is translated into a structured query and then used to score and rank the documents retrieved. Retrieval experiments using the NTCIR-2 test collection for information retrieval systems have shown that our method is equivalent to Rocchio's automatic relevance feedback method, when the initial query consists of only a few words.