Multimodally Grounded Translation. Session 7 - Directions for Research
Résumé
In most cases computer systems processing symbols or language do not have access to the phenomena being referred to. In contrast, human beings can readily associate expressions with their non-linguistic experiences. As a direct consequence, the present day computational systems can only reason about the symbols themselves rather than about the inherent meaning or external references of those symbols. This problem can be addressed in various ways. A traditional solution is to formalize the domain, such that the symbols used are defined in relation to each other to allow algorithmic analysis. However, multimodal contexts are more naturally represented as patterns and signals which differ considerably from the discrete representation of symbols and expressions of symbolic languages. This finding has some important consequences on how theories of linguistic semantics and pragmatics should be formulated, and what is the methodology that is needed in the computational modeling of high quality language processing and translation. In this paper, contextuality, multimodality and subjectivity are discussed as foundations for high quality translation and future scenarios for language technology and machine translation are provided with suggestions on how to divide translation tasks between human translators and machine translation systems. Special emphasis is given on underlying philosophical assumptions that may often have decisive role in determining whether a technological solution is functional and effective or not. lien video : http://webcast.in2p3.fr/videos- multimodality_grounded_translation_by_humans_and_machines
Origine : Fichiers éditeurs autorisés sur une archive ouverte
Loading...