Accuracy: systems trained with customer memories and terminology
We train SMT (Systems based on statistical training) and NMT (Systems based on neural networks) engines with customers’ own materials: translation memories that hold only revised content and validated terminology. This way, the resulting translation reflects to a large extent the desired style and wording, making the work of post-editing lighter. Furthermore, a variety of supporting glossaries are used for untranslatable words (proper names, brands, constants etc. ), and corporate terminology is updated at regular intervals.
Confidentiality: customer memories hosted in a non-shared environment
Each customer has their own language repository, and the assets are stored and managed separately per customer. Training is carried out subject to confidentiality clauses, both with machine translation system providers (encrypted secure connection) and internally, in order to ensure the utmost privacy of data. We do not share translation memories or terminology between customers.
Quantity: human-controlled post-editing
A team of translators, experts in the domain and accustomed to working with the customer, revises and corrects the texts that have come through machine translation. We provide exact post-editing instructions geared to rapid detection of potential errors. Moreover, all files go through final quality checking using automated tools which will highlight discrepancies or inconsistencies in the translations. We use evaluation metrics that measure the effort by the translator (post-edit distance) in order to continually improve the engines.
Solutions for your multilingual challenges
Linguaserve and OctaveMT, powered by Lucy offer solutions for fast and profitable translations:
- User-support service content