Delete search term

Header

Quick navigation

Main navigation

Machine translation for academic texts

At a glance

Description

The project consists in developing a prototype for a ZHAW neural machine translation system trained on academic texts. Freely accessible systems such as DeepL and Google Translate are not specifically trained on scientific texts and therefore often present issues regarding terminology, text cohesion, pragmatics and information structure. In this sense, the project is a follow-up to the DIZH-Fellowship Machine translation literacy for academics, which identified specific syntactic problems of machine translation from German abstracts into English. This previous project allowed to create a handout for German-speaking researchers to use NMT systems to produce English academic texts (available here: doi.org/10.21256/zhaw-22301). The combination of both project outcomes will provide ZHAW researchers with a solid framework to use NMT efficiently and safely.

The overarching goal is to help researchers at ZHAW to disseminate their findings and output more easily in English at the national and international level. Therefore, the prototype will be first trained for the language combination German-English. The proposed MT system will initially be hosted on a ZHAW-internal server to ensure optimal data protection. Indeed, new NMT solutions simplify the international dissemination and reception of scientific research and digital publishing accelerates the scientific dialogue. However, data protection is not guaranteed by free providers such as DeepL and Google Translate, which can be problematic for project proposals, for example.

This project aims at strengthening the place of the ZHAW in the digital international scientific dialogue and at the same time provide a better privacy of the data. This specialized NMT solution is intended to simplify the production of abstracts, papers, research reports and applications in English and thus to improve the international reception.

Publications