2 months ago
February 13, 2020
Use of ModernMT in IP and Life Science Translations: an ASTW Case Study
In the past four to five years, with a view to increasing productivity and enhancing linguist training, ASTW has implemented and relied upon a variety of Machine Translation (MT) solutions available on the market. Regardless of the pros and cons of each tested solution, we immediately understood the potential to include MT in our workflow. However, some problems persisted, both being inherent in the type of MT solution adopted.
- Large amount of time dedicated to creating glossaries and to setting morphology > Rule-based machines
- Difficult integration with assisted translation systems (CAT tools) > Rule-based and statistical machines
- Impossible or difficult output customisation > Statistical machines relying upon neural networks
- Non-optimal output quality > Statistical machines created only with their own data
- High price > Some statistical machines relying upon neural networks
- Data privacy issues > Some statistical machine relying upon neural networks
The recently launched ModernMT seems to virtually overcome all these disadvantages. We interviewed Davide Caroselli, the ModernMT Product VP.
Greetings Davide, can you tell us what you do?
Hi! I am the Product VP at ModernMT and responsible for our translation engine, its functionality and technical implementation.
How was the ModernMT project born and developed?
ModernMT was born of a European project, and, more specifically, in the context of Horizon 2020. The goal of the project was to design and develop a new machine translation technology that was able to adapt and learn from the feedback of professional translators. Having therefore openly an Enterprise-first approach, we modelled the service from its earliest stages on the basis of the needs of the translation market. The team of four founding members was certainly our biggest advantage: TAUS, the University of Edinburgh, the Bruno Kessler Foundation and Translated were the right state-of-the-art mix in terms of research in the field of artificial intelligence and strong knowledge of the market, as well as the needs of the movers and shakers of the translation world. Just think that two of the three founders of the technology behind Google Translate were part of the ModernMT team.
Today, ModernMT is a bona-fide company that has created and makes available on the market a product that relies on this innovative technology.
How does your solution “capture” the context of each translation, starting from the document in its entirety?
Whenever our translation engine is used, whether through a plug-in for a CAT tool or directly through its API, a fundamental step takes place even before the translator starts working on a specific project: Context Analysis.
In this step, completely transparent to the plug-in user, a component called the Context Analyzer analyses in a few milliseconds the entire document text to be translated; this process seeks out the distinctive terminology and intrinsic style of the document. This information is therefore used to automatically select the most suitable private memories loaded by the user for that particular document; that is, the memory inventory that best reflects the right terminology and writing style. It is precisely this inventory that the translation engine leverages in order to customise output in real time, for each single sentence of the document.
How do you guarantee the privacy of customer data?
Any content sent to ModernMT, whether a “.tmx” memory or a correction from a professional translator, is saved in the user’s private area. In fact, only you will be able to access your resources and make ModernMT adjust to them; in no way, will another user be able to utilise that same inventory for his/her system, nor will ModernMT itself be able to use those contents, other than to exclusively offer your personalised translation service.
In addition, ModernMT uses state-of-the-art encryption technologies to provide its cloud services. Our data centres, employee processes and office operations are ISO 27001:2013 certified.
The price of ModernMT is affordable for any pocket: how did you manage to create such an advantageous solution?
Being Enterprise-first since its inception, ModernMT’s goal has always been to create a product that was not only able to take the quality of machine translation to the next level, but that could do so by offering the market a competitive service. Packaging a product of excellent quality, but with a disproportionate cost would have meant the failure of ModernMT. For this reason, throughout its development, the search for algorithms, or the technical solution to improve quality, have always been accompanied by the quest for efficiency.
The technical solution at the heart of ModernMT, its “Instance-based adaptation”, not only guarantees all the functionality necessary for an enterprise translation engine, but it does so by reducing computation costs. For this reason, indeed, there is no “hidden” cost for the preparation, or training, of the customised translation engine, and our translation costs are a fraction of those charged by our competitors.
In your opinion, what are the competitive advantages compared to other commercially available MT solutions?
Without doubt its superior quality, and you don’t have to take our work for it. In a report published last year, “Inten.to”, a leading company in research in the field of automatic translation, presents ModernMT as the best engine on the market, beating names such as DeepL, Google, Microsoft and SDL.
What are the possible developments of ModernMT?
Without a doubt, expand the number of CAT tools supported. At the moment, ModernMT is available for SDL Studio 2017/2019 and Matecat, but there are many requests from potential customers to support a larger number of translation software. This is currently one of our biggest priorities.
In our case, we have been able to verify that ModernMT offers the following advantages:
- Fast response times
- Easy customisation by importing existing translation memories (as .tmx files)
- Machine learning ability verified as translation proceeds
- Quality output that is consistent with preferred terminology and one’s translation style
- Less post-editing efforts than other machines (e.g., Google Translate and DeepL)
- Continuous machine updating and learning from post-editor corrections
- Perfect integration with CAT tools (in our case, Trados Studio) using the dedicated plug-in
ModernMT is currently our favourite MT engine, especially in patent translations and in the Life Science sector, because it proves reliable, efficient, qualitatively better than its competitors, easily customisable and advantageous in terms of cost.
Domenico Lombardini, CEO ASTW