The extraction of description is important in audio-visual document retrieval. Although various approaches were proposed, extracting semiotic descriptions that reflect the content knowledge remains a challenging task. This paper proposes an approach for describing external and a semiotic description of audio-visual document. The main objective is to associate a set of semiotic descriptions to each filmic segment in order to facilitate the interrogation of huge quantities of the filmic document. The extraction description process relies mainly on the probabilistic LDA model and Dublin Core. To identify the description, a combination method applied both weighing score and position of each extracted theme. Qualitative and quantitative experiments were performed to evaluate the effectiveness of the proposed approach. The performance of the extracted semiotic description is better than that of the descriptions presented in the literature. The good results were obtained due to good choice of the information sources, the adaptation of the LDA, and the importance of the identification process.