This website uses cookies to store information on your computer. Some of these cookies are used for visitor analysis, others are essential to making our site function properly and improve the user experience. By using this site, you consent to the placement of these cookies. Click Accept to consent and dismiss this message or Deny to leave this website. Read our Privacy Statement for more.
Language Technology

The language services industry has been human powered for millennia and technology enhanced for decades. There is an endless list of technologies used in all facets of language in our personal and professional lives. This page focuses mainly on the common technological tools that help LSCs provide their services.

Translation Technology

Translation has seen the most innovation. Digital tools specifically created to help with this started and first grew in the 1980's. Computer-assisted translation (CAT) tools help human translators work. There are a wide range of these tools, and they aim to increase efficiency and even quality. Enterprise-level software and platforms have also been created specifically for LSC. Translation Management Systems (TMS) are digital environments that manage the entire translation or localization process, centralizing and coordinating team members, content, databases, glossaries, and other resources. Automated QA tools can also be used to scan and identify mistakes and errors in the target language file, such as incorrect terminology, missing segments, wrong formats, and more. AutoQA tools are particularly useful in localization.


Human Translation (HT)


Translation completed by humans, though they may use CAT tools or other resources to help their work.

Translation Memory (TM)


A system that scans a source text and tries to match strings of sentences or phrases against a database of strings that have already been translated for previous materials.

Machine Translation (MT)


Translation carried out exclusively by a machine or software.


The main types of machine translation include:

Rules-Based Machine Translation (RBMT)


Uses a lengthy set of preprogrammed rules about syntax, lexicon, and morphology in each of the languages in the pair (first developed in the 1970s).

Statistical Machine Translation (SMT)


Uses very large sets of parallel aligned bilingual texts to look for statistical probabilities to translate future source texts (although conceived of earlier, first developed in the 1980s).

Neural Machine Translation (NMT)


Uses one or more artificial neural networks or related deep learning models, which are themselves highly-specialized statistical systems (first developed in the 2000s).

Hybrid Machine Translation (HMT)


Combines two different basic types of machine translation systems, such as statistical with neural systems.


Speaking of Neural Machine Translation (NMT) – this new branch of technology is trying to innovate translation. And while it is impressive, it is still in the early stages, mostly research and hype. Media have latched onto this as the next thing that has “solved translation forever”, until the next breakthrough they will say the same thing about. Because of this, every large corporation wants to have their own on the market.

But “deep learning” and “neural networks” are still very broad terms. There are so many types of such systems. Facebook uses a convolutional network, Skype uses a long short term memory network, Salesforce uses a quasi-recurrent network, and Yandex uses a hybrid system. And in 2017, companies like Facebook and Google experimented with different systems and networks. We are still in the "Wild West" of neural machine translation. It is an exciting time for research, but businesses are still waiting to learn how to apply it all.

Artificial neural networks (ANN) don’t work like human brains, but they are inspired by the functionalities of human brains. NMT differs from SMT in that it works at the sentence level, not the word or short phrase level. Neural networks use vector representations for words and internal states, and then interpret the relationships of vectors to each other. Statistical engines are mostly trained with the corpora that are input. You need a large amount of pairs. Neural requires fewer pairs and, in some cases, perhaps no pairs. NMT is mostly trained with feedback loops, after it translates, gets feedback, makes adjustments, and then learns.  

Each existing neural network or deep learning system can vary greatly. And researchers keep experimenting with and creating even more types. Despite all of the products and platforms available on the market today, no one knows the direction that neural machine translation is going. Nothing out there yet is close to replacing LSCs. We should all keep an eye on these researchers and these NMT providers. There are some DIY neural network products you can try, but mostly, we all just have to wait and see.

Interpretation Technology

Everyone looks for technological innovation in translation, but not many think about such innovations in interpreting, mostly because there isn’t much. Research and development in this area is very new and most of it has focused on replacing the interpreter in so-called machine interpretation systems. But as with translation, this still leaves a large quality assurance gap. Unlike with translation though, real-time interpretation can’t wait until later for quality assurance.

The most common and best technological tools in interpretation are ones that help connect clients with interpreters at a distance. Video remote interpreting (VRI) allows a client to use a professional interpreter anywhere via an internet connection. This eliminates the geographical limitations in finding an interpreter who can arrive and work in person. Over-the-phone Interpretation (OPI) already helped solve this problem, but it was slow to catch on. And OPI did not support sign language, which helped lead to VRI. Advancements in bandwidth, video quality, and accessibility have allowed VRI to grow, for both sign languages as well as spoken languages.

One new form of VRI is Bring Your Own Device (BYOD) Interpreting. Earlier VRI systems were proprietary platforms. Then, general video messaging systems grew rapidly, and clients could use existing apps or messaging systems such as Skype if they could find an interpreter. Today we are starting to see apps as businesses that not only provide the visual communication platform but also supply the interpreters. Concerns around such apps, however, include lack of bandwidth, signal cutting out or stuttering, and confidentiality and security.

In terms of quality, onsite interpreters are still the best option. But VRI is becoming much more accessible and affordable and still often provides sufficient quality for most needs.

Language Training Technology

Various technological tools and platforms can be used in foreign language training. Educational institutions are now increasingly using a combination of technological or e-learning products with face-to-face teaching—known as blended learning or hybrid learning—to teach languages. Virtual classrooms make it easy to teach via webcam to deliver live online classes, providing students with convenient access to teachers online. Technology-enhanced language learning (TELL) may use digital tools or self-directed curriculums, but successful training still relies on well-crafted learning plans.

Language Testing Technology

The use of technology in language testing is still very limited. It is more commonly used in the delivery of the tests rather than in scoring.

But even with delivery, conclusions from a great deal of research on its efficacy or its impact on reliability are “mixed.” So using digital tools to give a test isn’t necessarily better. It really depends on the type of test and methods used to build it; technology does allow for one very useful type of testing platform—the computer adaptive test. This is a test administered by a computer in which the difficulty level of the next item to be presented to test takers is estimated on the basis of their responses to previous items and adapted to match their abilities. Based on item response theory (IRT), this means not everyone takes the same test, but each test can determine a more accurate rating of that individual.

The full automatic scoring of tests is even less common. Some types of test items can easily be scored with machines (e.g., multiple choice questions). But most items require an analytical eye, and for now still, research has shown that human evaluators actually retain more score rating validity than purely automated systems.

Learn more about language technology by talking to the experts who use it.

Find an LSC

Contact Us