Tag: human-aided machine translation

Adaptive MT – Trados 2017 New Feature

Adaptive MT – Trados 2017 New Feature


SDL Trados Studio 2017 includes new generation of machine translation.

How does it work?

It allows users to adapt SDL Language Cloud machine translation with their own preferred style. There is a free plan and it offers these features:

  • 400,000 machine translated characters per month.
  • only access to the baseline engines, so this means no industry or vertically trained engines.
  • 5 termbases, or dictionaries, which can be used to “force” the engine to use the translation you want for certain words/phrases.
  • 1 Adaptive engine.
  • Translator… this is basically a similar feature to FreeTranslation.com except it’s personalized with your Engine(s) and your termbases.

How does it help?

  • Faster translation with smarter MT suggestions.
  • Easy to use and get started.
  • Completely secure – no data is collected or shared.
  • Unique MT output, personal to you.
  • Access directly within Studio 2017.
  • No translation memory needed to train the MT.
  • Automatic, real time learning – no pre-training required.

What are the available language pairs?

Uptill now, Adaptive MT is available in these language pairs:

English <-> French
English <-> German
English <-> Italian
English <-> Spanish
English <-> Dutch
English <-> Portuguese
English <-> Japanese
English <-> Chinese

For reference: https://www.sdltrados.com/products/trados-studio/adaptivemt/

The Translation Industry in 2022

The Translation Industry in 2022

In this report, TAUS shares predictions for the future of the translation industry in line with their expectation that automation will accelerate in the translation sector during the coming 5 years. The anticipated changes will inevitably bring along various challenges and opportunities all of which are explained thoroughly in the Translation Industry in 2022 Report.

The report explains the following 6 drivers of change

1. Machine Learning

Machine learning (ML) was introduced in the 1950s as a subset of artificial intelligence (AI), to have programs feed on data, recognize patterns in it, and draw inferences from them. 2016 was the year when ML went mainstream, with a lot of applications that were almost unimaginable a few years earlier – image recognition and self-driving cars are just two examples.Computational power and unprecedented advances in deep neural networks will make data-driven technologies astonishingly disruptive. This might be also the case of MT.

As a rule, the growth of machine intelligence represents a threat to many human jobs as people will be replaced by intelligent systems.  The majority of creative jobs is relatively safe while sales jobs could be at risk. The forecast is dubious for technology jobs, but the more senior jobs being relatively secure, while computer programmers and support workers may likely be replaced.  The assumption that jobs requiring manual dexterity, creativity, and social skills are the hardest to computerize is already obsolete: new developments in deep learning are making machines more powerful than anticipated, especially in areas relating to creativity and social interaction. 

In the translation industry – as in other industries – many functions will be affected – whether enhanced, expanded or replaced – by ML.

2. Machine Translation

In the past years NMT has been said to be achieving impressive results, and it is more and more often presented as a replacement for SMT. Advances in artificial neural networks are bringing extremely high expectations, suggesting that NMT could rapidly achieve higher accuracy than SMT. Independent evaluators fnd that NMT translations are more fluent and more accurate in terms of word order compared to those produced by phrase-based systems. Better quality MT will mean that a broader range of document types and audiences can be addressed.
NMT will help the further expansion of speech-to-speech (S2S) technologies, now available mostly as English-based monolingual systems. Transforming these into multilingual systems implies many deep and expensive changes. Most S2S technologies are still at an infancy stage and confned to university labs. NMT will help bring speech-enabled devices to the streets.

MT will lead to the ultimate disruption in the translation industry when, only the premium segment of artsy—and possibly life sciences—
translation will remain tradable.

3. Quality Management

Due to the uncertainties intrinsically involved in translation quality assessment, and the fixity of the relevant concepts in the translation community, users seem now willing to accept good enough MT output, especially for large volumes, delivered virtually in real time. For serviceable MT output with no human intervention downstream, TAUS coined the acronym FAUT (Fully Automated Useful Translation) already in 2007. Investing in quality-related decision support tools has become essential to gain translation project insights and beneft from MT.
Applying machine learning to data-driven translation quality assessment will be a disruptive innovation that will call for a major shift in conception and attitude.  Data-driven applications in translation quality assessment will go from document classifiers to style scorers, from comparison tools to automatic and predictive quality assessment, from content sampling to automatic error detection and identification. The data-driven approach to quality will require another major attitude shift.

4. Data

There is a strong need for data scientists/specialists/analysts, but this profile is still absent from the translation industry.

Data has been the fuel of automation, and after entering the automation era at full speed, we are being challenged with many issues.  Translation data is typically metadata: data about translation that can be harvested downstream the closure of a translation project/job/task.  The analysis of translation data can provide a very valuable insight into the translation processes to find the best resource for a job, to decide what to translate and which technology to use for which content. Translation data will be more and more frequently generated by algorithms. More data will come from rating staff and KPIs. All these kinds of data will come from ML applied to translation management platforms, which will get rid of human involvement.

Erroneously, also data covering multilingual text resources is labeled as translation data. In fact, language data specifically consists of translation memories, corpora, and lexicographical and terminological collections. Of course, all these resources have metadata too, which could be exploited. Stakeholders should become more open and massively start sharing their translation metadata to make it the real big data of the translation industry.

There is a strong need for data scientists/specialists/analysts, but this profile is still absent from the translation industry. Translation companies should be looking out for these specialists who can mine and use data for automation. This will most probably lead to a further reduction of the number of translation companies that are able to float and thrive in a more and more competitive market. The challenge for the next few years might be the standardization of translation data in order to shape it and make it convenient for users to derive the maximum benefits from it.

5.  Interoperability

Interoperability is the ability of two different systems to communicate and work together through a common language or interface. While many other industries have flourished thanks to standardization which led to interoperability, automation and innovation, the translation industry has always suffered from a lack of interoperability. This has been costing a fortune for years, both on the client side (in translation
budgets) and on the vendor side (in revenues).
  Things have been changing a lot since 2011, when TAUS published a report on the costs from
lack of interoperability in the translation industry
. Many blame the lack of compliance to interchange format standards as the primary barrier to interoperability, and no one believes any longer that true interoperability in the translation industry can be achieved only through awareness programs, education, and certifications. Interoperability should come from the adoption of standards created by consortia and not from the dominance of a market leader.

The spreading of MT has forced a breakthrough in the interoperability dilemma, starting a wave of innovation and renewed efforts. Most of these efforts have still been focusing on APIs though, as XML has been established for years as the common language, bringing everyone the industry to find its child formats TMX and XLIFF essentially enough.  So far, most of the many APIs made available are meant to simplify the translation business process and reduce translation management and overhead cost. Only a few have been designed to help disintermediation and facilitate access to services.

In this case, we could expect that the most influential buyers of localization and translation services will advance their requests; the technology vendors with the necessary technological and financial resources will fulfill those requests or even introduce their own solutions on the market, just as it happened in the past.

6.  Academy

Translation education is vocational by definition: it prepares people to work in the trade as translators. None of the skills translation students acquire is anything but sophisticated.  Today, many players in the translation industry complain about the lack of good translators, but they seem to ignore that, more than in many other academic fields, translation education follows obsolete models that are still shaped for the 20th century. To make matters worse, the gap between the academic world and the industry is so wide that, when approaching the job market, translation graduates instantly and bitterly realize they don’t know much about the actual work they are supposed to do. They also discover that the world is not interested in their basic skills.

The future may not really need translators, at least not in the old way, as the audience will become even more forgiving for lesser quality of fast-moving content. A highly-automated localization environment will depend on human skills in quality evaluation, content profiling, cultural advisory, data analysis, computational linguistics, and gradually less and less in post-editing; translating plain text will indeed be a long-tail business.

The success of any innovation depends entirely on the people that are going to nurture, develop, and implement it; in times of exponential growth, education is vital to drive adoption and prepare the next generations of workers. Employers should play a part in closing the skills gap with continued professional training. It is never too early to prepare for the future; vast workforce and organizational changes are necessary to upend stale business models and related processes.

For more details, download the full report.

Machine Translation Post-Editing Types

Machine Translation Post-Editing Types

Post Editing is the next step after completing the machine translation (MT) process and evaluating its output. A human translator processes the document to verify that the source and target texts convey the same information and that the tone of the translation is consistent with the original document. The quality of machine translation varies and affects the subsequent effort required for post editing. There are contributory factors to the quality of the MT such as the clarity and quality of the source text; it is important to make sure that the source text is well-written and well-suited for machine translation beforehand. Other considerable factors that affect MT output quality include: the type of MT used, and the compatibility of the source and target languages.

There are two types or levels of post editing

Read More Read More

Evaluation of Machine Translation Output

Evaluation of Machine Translation Output

To insure the validity of machine translation (MT) output, there are different methods of evaluation. A rudimentary form of evaluation is to perform a “round-trip translation”, meaning that the original text is machine translated into the target language, and then the result of that process is translated back into the original language to test the quality of the machine translation. As the quality of machine translation continues to improve, a reliable method for evaluation will also be necessary. Currently, there are two main types of evaluation used for machine translation: human and automated.

Read More Read More

Machine Translation History & Approaches

Machine Translation History & Approaches

Machine Translation (MT) refers to automated language translation. The concept has been around since the 1600’s but has come into its own beginning in the twentieth century. Along with the invention of electronic calculators came the development of ways to adapt computer technology to language translation of documents. Research became prevalent at universities in the mid 1950’s to develop and test machines to perform tasks previously only possible by human translators.

Read More Read More