Using Shooting Scripts for Indexing Moving Images

Introduction

This paper reports some of the results of a study that looked at three types of texts we believe are probably the richest sources for automatically deriving shot-level indexing to moving images. The three types of text are closed-captioning for the hearing impaired, audio description (a technology for the vision impaired), and shooting scripts. In this paper we focus on the results related to the text contained in the shooting script broken down for the shoot.

For the present part of the study, we needed a script broken down for the shoot so we could compare the terminology used for identification of costumes, props, and so on, with the text of the description we wrote for each shot as we analyzed them from videocassettes. For the other parts of our study, we needed to compare the terminology with the text of the closed-captioning and the text of the audio description. From such comparisons we are able to identify the indexing terms useful for retrieving individual shots and sequences from an information system housing this kind of information. The main objective of the global study was to look at the contribution each of these textual sources makes to the overall indexing picture.

The production we used to gather data for this study is titled Cher Olivier, directed by André Melançon, produced for and distributed by Avanti Ciné Vidéo Inc. in 1997. "Based on the life of the great Québec comic Olivier Guimond, this series recounts the highs and lows of his career and personal life" (Québec Audiovisuel 2004). The production was broadcast as a miniseries in five one-hour parts on television in the spring of 1997 and later that year won eight Gémeaux awards (iFrance 2004). Cher Olivier is a French-language production and was the only one for which we were able to gather all the pieces for our study: the text of the closed-captioning, the text of the audio description, and the breakdown of the shooting script.

Background

Our previous work in studying shot-level indexing to moving images has shown that moving images (and still images other than art images) seem to have characteristics not found in textual documents and other types of information objects (Turner 1994a, 1994b, 1995; Hudon, Turner, and Devin 2000). There is a rather direct correspondence between the information content of images and the words used to describe them, whether these words come from professional indexers, from associated texts such as descriptions in print catalogues or news databases, from descriptors provided by viewers, or from a number of other sources. People who have the task of describing the content of images name the objects, persons, or events they see in the picture. Although they use a variety of terms to describe the same thing, the top terms emerge rather obviously, are consistent from one context to another, and provide the most obvious subject indexing for the shots because they are the terms named most often.