An XML Coding Scheme for Multimodal Corpus Annotation

Multimodality has become one of today’s most crucial challenges both for linguistics and computer science, entailing theoretical issues as well as practical ones (verbal interaction description, human-machine dialogues, virtual reality etc...). Understanding interaction processes is one of the main targets of these sciences, and requires to take into account the whole set of modalities and the way they interact. From a linguistic standpoint, language and speech analysis are based on studies of distinct research fields, such as phonetics, phonemics, syntax, semantics, pragmatics or gesture studies. Each of them have been investigated in the past either separately or in relation with another field that was considered as closely connected (e.g. syntax and semantics, prosody and syntax, etc.). The perspective adopted by modern linguistics is a considerably broader one: even though each domain reveals a certain degree of autonomy, it cannot be accounted for independently from its interactions with the other domains. Accordingly, the study of the interaction between the fields appears to be as important as the study of each distinct field. This is a prerequisite for an elaboration of a valid theory of language. However, as important as the needs in this area might be, high level multimodal resources and adequate methods in order to construct them are scarce and unequally developed. Ongoing projects mainly focus on one modality as a main target, with an alternate modality as an optional complement. Moreover, coding standards in this field remain very partial and do not cover all the needs in terms of multimodal annotation. One of the first issues we have to face is the definition of a coding scheme providing adequate responses to the needs of the various levels encompassed, from phonetics to pragmatics or syntax. While working in the general context of international coding standards, we plan to create a specific coding standard designed to supply proper responses to the specific needs of multimodal annotation, as available solutions in the area do not seem to be totally satisfactory.