Custom.MT is a generative AI and machine translation platform for localization teams, used to improve enterprise AI localization and evaluate machine translation quality.
Purpose The platform serves localization workflows, helping teams manage, connect, evaluate, and optimize machine translation and generative AI models, and can connect models to various language tools.
Main capabilities
- Services: enterprise machine translation evaluation and quality intelligence, machine translation model fine-tuning, translation memory cleanup (TMX), machine translation quality estimation, machine translation automatic post-editing, on-premises machine translation deployment, language dataset acquisition, language AI team training workshops.
- Products: AI translation platform, custom translation portal, plugins or extensions for Trados, Smartling, memoQ, Shopware, XTM Cloud, Smartcat, API, AI translation and machine translation pricing documentation.
- Management and connection: manage proprietary and preset translation models and connect them to language tools.
- Evaluation and comparison: use the Best Engine tool in Custom.MT Console to compare DeepL, Google, GPT-4, custom and other models, view BLEU and WER scores, sample segments and engine rankings to choose the model suitable for localization workflows.
- Tag handling: tag handling routines can preserve tags when AI translates software strings, or remove them.
- Model training and fine-tuning: improve model accuracy through terminology and style training, reduce edit distance and save localization costs.
- Machine translation evaluation: measure model performance by language, domain and linguist through automatic and human evaluation, with metrics including BLEU, WER and COMET.
- On-premises deployment: deploy machine translation on your own hardware to handle confidential content that cannot flow to the cloud.
- Translation memory cleanup: perform annual cleanup to keep translation memories useful to translators, or prepare data for model training.
- Automatic post-editing: replace human review with machine translation automatic post-editing, reducing language audit costs by up to 70%.
- Data acquisition: search for and acquire in-domain training datasets from public sources and the market, covering medical, legal, patent, and e-commerce data.
- Open-source machine translation: build your own branded machine translation models using open-source data and the latest neural network toolkits.
- Language data anonymization: automatically remove personal information such as names, phone numbers, social security numbers, while preserving language data for training.
How to use Users can start translating through Custom.MT Console , manage and connect models, compare engines using the Best Engine tool, configure tag handling, perform model training and fine-tuning, and connect to systems such as Trados, Smartling, memoQ, Shopware, XTM Cloud, Smartcat via plugins or extensions. The platform provides an API and related pricing documentation.
Use cases The materials mention applications including: machine translation for nonprofit organizations in the field of climate change, machine translation for player chat in video games, legal and corporate translation under regional budgets, and fully automated translation at scale for healthcare applications.
Limitations The materials do not provide details such as specific pricing amounts, supported language pairs, deployment environment requirements, free trial conditions, or performance guarantees, and are not provided for now.
