本稿は、自由発話に対する意味理解と、これを音声認識と統合する手法を述べる。まず、概念を理解単位(概念素)とし、発話を意図の伴った概念素列とみなす意味理解の枠組を提案する。次に、概念素仮説の言語尤度を評価しながら、文節ラティスから概念素を島駆動探索する概念素探索法を述べる。さらに、文意仮説の効率的な探索のために、概念素の脱落を許容して初期文意仮説を高速に得、これより想起される概念素予測知識を利用して脱落区間を再探索し、初期文意仮説を修復する手法を提案する。ホテル予約の自由な質問を対象とした音声理解実験の結果、第一位で92%の良好な理解率を得た。また、高速化手法により理解性能の低下なしに数倍?十倍の効率向上を達成した。This paper describes integration of speech recognition and semantic interpretation for spontaneous speech understanding. First, we propose a framework of semantic interpretation where a concept is a unit of semantic understanding and an utterance is regarded as a sequence of concepts with an intention. Secondly, we describe a basic search method which detects concepts from a phrase lattice by island-driven search evaluating linguistic likelihood of concept hypotheses. Moreover, an improved method to efficiently search for meaning hypotheses is proposed. This method quickly generates initial meaning hypotheses allowing deletion of concepts. Then, the initial meaning hypotheses are repaired by re-searching for missing concepts using prediction knowledge associated with the initial meaning hypotheses. Experimental results show that 92% of understanding rate has been acheived, and that the improved method has realized efficient search without reducing its performance.