Expert Evaluation of AI-Based Translation Solutions for the Canton of Bern Administration

Description

On behalf of the Canton of Bern Administration, the Institute of Multilingual Communication at ZHAW conducted an independent expert evaluation of AI-based translation solutions for use by the language service and across the entire administration. The aim of the project was to identify suitable translation systems, assess them systematically, and compare their suitability with regard to the specific requirements of a multilingual public administration. Both neural machine translation (NMT) systems and solutions based on Large Language Models (LLMs) were taken into consideration.

In addition to translation quality in the official languages, the evaluation focused in particular on data protection and information security, adaptability to domain-specific terminology, user-friendliness, integration possibilities into existing applications and workflows, as well as scalability, maintainability, and cost structures. Furthermore, different deployment models (on-premise, private cloud, and hybrid approaches) were examined. Particular attention was also paid to the solution's ability to specifically control or deactivate security-relevant functions in order to reliably meet regulatory and organisational requirements.

The results of the evaluation provide a basis for selecting a future-oriented, secure, and flexible translation solution that meets the requirements of a modern digital administration.