The Relo-KT Process for Cross-Disciplinary Knowledge Transfer

Abstract

Digital humanities research, by its nature, is collaborative and interdisciplinary. A key aim when undertaking cross-disciplinary research is to integrate insights from two or more distinct disciplines using both formal and informal methodologies to achieve effective knowledge transfer. Such collaboration can lead to new ideas, creative solutions and innovation within both disciplines, in a way not possible within single discipline research and work. Whilst collaboration happens regularly in academic, creative and work environments, methods of cross-disciplinary knowledge transfer in interdisciplinary research are not often documented.

This thesis explores synergies between the linguistics discipline and the extensive science around machine language translation. While both disciplines have their own distinct approach to solving problems, combining these disparate skills within a particular application affords exciting opportunities to develop.

The multi-step relo-KT process was developed during this thesis to formalise and codify collaborative cross-disciplinary knowledge exchange. The process incorporates establishing an interdisciplinary question; acquiring a corpus of data suitable for analysis and extracting domain specific understanding from it. The process is iterative in nature as the cross-disciplinary knowledge codification and transfer develops between the discipline experts.

To rigorously examine the relo-KT process, it is applied to the RF-MT (rhetorical figure-machine translation) use case, in which linguistic understanding of rhetorical figures is codified to facilitate a tangible transfer of linguistic knowledge to the machine translation (MT) domain.

A multi-faceted, mixed method approach is taken to enact the relo-KT process. The Rhetorica software is deployed to automatically detect rhetorical figures from a corpus of political statements. Key rhetorical figures explored include epanaphora, epistrophe, polyptoton and polysyndeton, each well understood within the linguistics field, but each posing challenges for effective machine translation.

Quantitative findings from the application demonstrate the complex nature of persuasive speech. A repository of exemplary codified rhetorical figures for persuasion is developed and improved over a series of semi-structured, collaborative interviews with experts from the field of machine translation. Qualitative findings from the iterative series of interviews indicate that the MT domain is primed to integrate linguistic nuance, and a potential application is in the area of automated post-editing of machine translations.