Annotating Multi-media/Multi-modal Resources with ELAN

This paper shows the actual state of development of the manual annotation tool ELAN. It presents usage requirements from three different groups of users and how one annotation model and a number of generic design principles guided the choices made during the development process of ELAN. Introduction At the Max-Planck-Institute for Psycholinguistics (MPI) software development on annotation tools for the manual annotation of multimedia data has been going on since the early 90’s. Over this decade there have been large changes in enabling technology and insights in the nature of linguistic annotation. Media frameworks for the handling of digital audio and especially digital video files have matured, as has media streaming technology. XML has come to existence and has become highly relevant in a short time. Rendering and input of Unicode characters is now commonplace. Simultaneously, users made experiences with the first generation of video annotation tools and became aware of and got used to these new technologies. From this a new set of requirements arose. Finally, annotation tool builders are better aware of each other’s approaches, annotation models and annotation document formats. Clearly convergence is going on, leading to easier exchange of data between annotation tools. An important role in this process was played by the paper by (Bird & Liberman, 2001) that introduced Annotation Graphs. We are closely watching and trying to participate in standards initiatives, as for example ISO TC37/SC4. The first video annotation tool developed at the MPI was MediaTagger, a QuickTime based application that runs only on pre-OS X Macintoshes. It started as a first attempt to exploit the QuickTime Movie data structure, and especially it’s text tracks, as an informal model for linguistic annotation. Since then several new formal models where made, each one building on the experiences of the previous ones and considering new user requirements. The formal modeling languages that were used are Entity-Relationship diagrams and UML. A detailed presentation and evaluation of these models can be found in (Brugman & Wittenburg, 2001). The next chapters will discuss the requirements of several different groups of users and describe the latest state of ELAN functionality. We will then present our model for annotation in some detail and show how we can cover the needs of very different user groups with one relatively simple model. In the discussions plans for future development will presented. 1 http://www.mpi.nl User requirements ELAN is developed with a number of different user groups in mind. These users are situated both within the MPI and, in an increasing number of cases, outside the MPI. Often they are participating in externally funded projects (DoBeS, ECHO). We will discuss the main requirements per group, although there is of course a substantial overlap between each group’s needs.