Machine Translation Services
Machine translation (MT) involves the use of computers and software programs to automatically generate a translated output. There is no point in arguing whether human parity has already been reached. The truth is that this technology has evolved to a point where, under the right conditions, it can be a valuable asset and a powerful tool.
Trusted Translations has specialized teams dedicated to implementing different types of machine translation solutions. Our teams are on the cutting edge of new MT technology and understand exactly how to integrate MT solutions with other translation tools, as well as human post-editing and further reviewing services. By thoroughly understanding the benefits and limitations of the latest MT technology, we are able to determine whether MT can be effectively used in a particular situation.
Our experts act as consultants to evaluate whether an MT solution is right for a particular client, domain and project scope. After following a few steps that may involve sample testing, and once both the client and our experts have determined that an MT solution works, our teams will start working to design the right workflow and technology to be implemented. This initial effort is well worth the investment, for it will ensure the maximization of our client’s ROI while producing the highest quality output adapted to the end-user’s needs.
For valued recurrent clients that need to frequently process high-volume content, our teams will further analyze all stages to implement the best MT service offer. This includes an analysis in the medium and long term in order to ensure that our client’s goals are being met at every stage during that implementation. In other words, different stages may require a different set of tools to achieve the optimal level of consistency and reliability. For example, an earlier stage may require the use of more human translators to train the engine while later stages can be managed with a more simple Quality Control (QC) process (e.g., by specifically checking terminology or running a spell and grammar check).
Neural Machine Translation
MT solutions were historically built on two basic and distinct models: rule-based and statistical. Developments in MT followed a logical path. At first, experts in the budding science of computational linguistics considered grammar rules to be the foundation for working with corpora. A few years later, statistics and statistical models started to creep in. Next came the merging of both concepts into hybrid (rule-based and statistical) models.
The law of diminishing returns in place began to be reflected in small, incremental improvements. When many of the early MT users were starting to lose faith, a quantum leap was achieved with the application of neural network concepts. Neural Machine Translation (NMT) made a stellar intro in the market. NMT showed a competitive edge when demonstrating that all combinations of languages were working much better compared with the old technologies. Another extraordinary benefit was that it broke down the wall that prevented adoption for all players in the game. That “psychological” switching cost was covered in large part. Users started relishing the “usability” of a faster and more cost-effective solution. Clients buying translation services started challenging their suppliers with an unbelievable claim: “machine translation is providing me with the same, or even better quality than you’ve delivered.” And, on the other end, linguists started praising the machine output they were receiving for processing from their project managers.
Trusted Translations Machine Translation Engines
Machine translations are based on previously translated text commonly referred to as bilingual text corpora. For older technologies, a sufficiently large corpora helped build a machine translation engine that could be used with impressive accuracy. With the advent of NMT, high-quality corpora is what really matters. Large volumes of translation units are no longer needed for algorithms used are no longer the actual differentiators. In summary, democratization of MT is here to stay.
Trusted Translations has the capabilities to offer you a customized Neural MT-based engine in case a general domain Neural MT one would not suit your needs. Nonetheless, our experts will help you first determine whether an MT solution is a viable path or not. Next, we will help you define which of those two main kinds—a General Domain Neural Machine Translation solution based on the use of a General Machine Translation Engine (GMTE), or a Customized Machine Translation solution based on a Custom Machine Translation Engine (CMTE)—is what you need for your globalization strategy.
The main difference between a GNMTE and a CNMTE is that a GNMTE is usually an “off-the-shelf” engine not customized for a specific client’s needs, while a CNMTE is a customized engine built and tailored for a specific client. Both options can be complemented with human post-editing and even further review steps to improve quality. However, a customized machine translation solution allows for the engine itself to become more intelligent over time, based on the specific needs and content of the client. General machine translation solutions can normally be somewhat improved for a specific client by being supplemented with the use of a translation memory server, but never at the same level as when training customized engines.
Based on our client’s goals, we will determine the best approach, create a plan, and implement the MT solution from A to Z. Trusted Translations is one of the few companies in the world that understands this technology and has the knowledge to implement a solution that produces high-quality machine translations to improve overtime with human customization and post-editing. Contact us today for a free evaluation of whether your project is right for an MT solution.
Translation Memory Servers and General Domain MT Engines
As touched on above, Translation Memory Servers combined with General Domain Neural MT Engines can make a simpler and lower-cost solution compared to a high-quality and Customized Neural MT one. This is accomplished by integrating a Translation Memory Server with the appropriate translation memory. A “Translation Memory Server” (TMS) refers to a server where translation memories are stored and can be accessed by other translators. The server houses the translation memory in a centralized location, allowing human translators access to past translations, and in some cases granting them the ability to modify and improve the translation memory. A TMS should not be confused with a General Domain Neural Machine Translation Engine (GNMTE). These are two separate tools that can be combined or used separately.
As the results from GNMTEs alone are generally not as reliable for a specific subject matter (even with human editing), we often recommend integrating a Translation Memory Server into the workflow to improve the quality of the output. Trusted Translations has one of the most experienced teams in the industry when it comes to implementing translation memories and managing translation memory servers as part of a General Domain Neural MT solution.