All the specialists have agreed that the possibility of adding to multimedia WWW objects some sort of `conceptual' annotations describing their information content would greatly contribute to solve the problem of their `intelligent' indexing and retrieval. We propose to associate with the Web objects not the final conceptual annotation, but a simple natural language (NL) caption, in the form of short texts representing a general, neutral description of their informational content. The NL caption will then be converted into a conceptual annotation in NKRL (Narrative Knowledge Representation Language), making use of an automatic translation system like those we have implemented in the context of recent European projects.