Event details
| Date | 24 November 2026 |
| Time | 16:00 CET |
| Duration | 1 h (10-15 minutes for Q&A) |
| Format | Webinar |
| Language | English |
About the session
With new machine translation models emerging every day, how do you know when it's worth switching engines — and which one will deliver the best results for a specific use case, language pair or subject domain?
In this webinar, we'll show how to make informed, data-driven decisions by combining automatic metrics, human evaluation and the analysis of how closely their results correlate. We'll introduce an evaluation platform developed by imaxin that brings these processes together in one environment, helping language service providers compare MT systems, identify the best fit for each project, and reduce post-editing and human review time.
Contents
- The evolution of machine translation systems: from rule-based to neural models and LLMs.
- The evolution of evaluation methods: automatic metrics, human evaluation and their limitations.
- Challenges of evaluation in professional environments and imaxin’s experience with different models and language pairs.
- Vera: integrating automatic and human evaluation into a single platform.
- The evaluation process with Vera: reference preparation, metrics, MQM annotation and results analysis.
- From results to decisions: statistical significance, inter-annotator agreement, correlation and selecting the most suitable system.
Speaker
Sofía García
Computational linguist
About Vera
Vera is a web-based platform developed by imaxin to evaluate and compare machine translation systems. It brings together automatic metrics (BLEU, COMET, chrF++, TER) and human evaluation in a single environment, enabling translation quality to be assessed from both quantitative and linguistic perspectives.
The platform facilitates collaborative work among evaluators, analyses the correlation between automatic and human evaluation results, and generates reports based on the data collected. This makes it possible to compare translation engines and make evidence-based decisions when selecting the most suitable system for each language pair, subject area or use case, through a more efficient and traceable process.
Do you have a project?
Request a no-obligation quote.