Direct contact
Robin Gegenheimer
+49 (0)7248 925 425
info@across.net
AcrossMT
Train your own machine translation engine and retain data sovereignty
AcrossMT is a highly specialized machine translation engine developed by Across. Each engine is individually trained with your data and therefore delivers outstanding results, as AcrossMT takes into account the language and terminology of your company. AcrossMT can be used as a standalone solution, no matter which TMS you use. Thanks to the web interface, AcrossMT can also be made available to all employees. In this way, we guarantee maximum security when using AI-based machine translation.
Wondering how you could use AcrossMT?Not sure whether it would be useful for your organization and your translation processes? |  |
Book your pilot phase
Benefit from our standardized process, which guides you through the pilot phase step by step and helps you to assess the benefits of AcrossMT in your specific setting.
In the pilot phase, you train AcrossMT with your own data from the translation memory and the terminology database. With our review tool, you can objectively compare the quality of your translations with those of your competitors.
Why is AcrossMT the best choice?
Data sovereignty protected
With freely available online translation engines such as DeepL, all data are used to further train the engines. The data for the training comes from various sources, e.g. scientific papers and technical documentation. Even security-critical or sensitive corporate data are included in the training. In contrast, AcrossMT guarantees maximum security for proprietary corporate data.
Terminology
Corporate terminology is a vital aspect of every good translation. Apart from training AcrossMT with the customer's data, the customer's terminology can be directly integrated. This is another area in which you as a language service provider can generate income. After all, the quality of AcrossMT depends on the due preparation of the data prior to deployment.
Predictability
Owing to the flat-rate pricing, the costs of the AcrossMT always remain unchanged, regardless of the number of characters translated or the number of employees who use the web interface.
Higher quality through systematic retraining
To ensure maximum quality of AcrossMT, retraining is not triggered automatically while using the system. Rather, it is up to you to decide when you have sufficient data for systematic retraining. With this approach, you can boost the quality of the AcrossMT engine.
Standalone
AcrossMT can be operated as a standalone solution. Regardless of which TMS you use, AcrossMT can be integrated in existing systems and be used by all employees in the company via a web interface.
No user limitation
AcrossMT is not hampered by complicated licence rules. Rather, AcrossMT is available to all employees.
FAQ
A trained AcrossMT engine knows your corporate terminology and translates in your tone of voice. This greatly reduces the post-editing workload. Using the retraining feature, your AcrossMT engine continues to learn on the basis of new crossTank translation units. Across does not expect you to decide for either the one or the other. You can continue to use DeepL for languages not yet supported by AcrossMT. In v8.0, this also includes the use of stored DeepL glossaries.
The approach is straightforward: We train our AcrossMT with your data in one language for you. You merely pay for the training overhead, but do not assume any obligation and do not sign any contract. After training AcrossMT with your data, we will select a number of text segments with you and have them translated by various sources (human translator, DeepL, Microsoft, ChatGPT, etc.). Important: The text segments must not be taken from the MT training data. These segments will then be randomly sent to the reviewers assigned by you. The reviewers will rate the translation quality without seeing whether the texts have been translated by a human, by AcrossMT, by DeepL, or by any other engine you would like to include in the rating.
With the help of the Review Tool, you can objectively assess which machine translation engine delivers the best quality for your translations. Use valid data to decide which engine works best for your texts. We are convinced that AcrossMT will get top ratings.
Contact us to book your pilot phase!
Depending on the volume, we need 1-2 weeks for one training run.
The Basic package accompanies the pilot phase for the purpose of ensuring the best possible quality of the AcrossMT engine by selecting suitable training data. Additionally, the Plus package includes a thorough ROI analysis on the basis of the Across Data Cube. The review tool can be added to both packages for quality reviews. This comprises the creation of quality metrics by our analysts.
AcrossMT is capable of delivering good to excellent results for all text types. If trained with several hundred thousand segments, machine translation is able to deliver a quality that is on par with human translation.
This greatly depends on the quality of the training data. As a rule of the thumb, we recommend 100,000 segments for excellent quality. The more the better. At least 50,000 segments should be provided.
No. AcrossMT is already supported from version 7.0 15980.
We can help you to boost the acceptance in your organization. For example, the pilot test provides an ideal setting for convincing language service providers and internal critics of the quality of AcrossMT. For this, we offer the review tool in order to document and compare the quality objectively. Moreover, your decision-makers surely want to see some figures. For this purpose, you can book our ROI analysis. Based on your archived translations, we can prepare a forecast of the potential savings for your translation process.
No, the review tool is made available via an anonymous link. It can be booked along with the four-week pilot phase and does not require any user license.
AcrossMT can be used via a web frontend or via a plugin in the Across Language Server. If necessary, third-party systems can directly access AcrossMT via the REST API.
Users who are logged in benefit from a dashboard that presents statistics on the character count per language and over time. Moreover, the volume can be displayed by source, i.e. web frontend, Across, or third-party system.
The automation of the retraining is the next feature in the pipeline. In this way, new translation units will be transmitted from the Across Language Server to the AcrossMT engine. Additionally, we plan a dynamic MT connection for crossDesk that will propose translation options while editing a segment.
Technically speaking, definitely yes. Third-party systems can easily be connected via the API. Theoretically, the solution can also be used in other CAT tools. However, free retraining is only available in conjunction with the Across Language Server. Henceforth, the dynamic MT feature will support post-editors in crossDesk, the editing environment of the Across Language Server.