Tag Archives: machine translation

Our first time at SATT!

by Giannis Nistas, Linguist at Commit

The 6th edition of the School of Advanced Technologies for Translators (SATT) took place on September 14-15 in Milan, Italy, on the premises of the International University of Languages and Media (IULM). It was attended by 120 participants, with 20% of them coming from abroad. Also, more than 50% of the participants work in the language industry.

Organized by the Bruno Kessler Foundation, this year’s school revolved around Machine Translation (MT) and other advanced technologies for translators, with lectures and labs spanning across two days, and speakers coming from the realms of research, academia, and the language industry itself. The first day was dedicated to lectures, whereas on the second day we received hands-on training in the university labs.

The keynote lectures were given by Sharon O’Brien and Renato Beninatto. The former, coming from the academia, tried to include MT and the skills associated with it into different translation competence models, and set some food-for-thought questions about how to fit MT in the training of translators and how to future-proof their careers. The latter, an industry veteran with extraordinary communication skills, provided us with an overview of the translation technology landscape with particular reference to developments in MT. His lecture was enhanced by personal experiences as well as tips for translators he shared with us.

Researchers Marco Turchi and Luisa Bentivogli introduced us to MT and MT quality evaluation respectively. Turchi gave us a detailed presentation of how Phrase-Based Machine Translation (PBMT) and Neural Machine Translation (NMT) systems work and drew comparisons on the performance of the two approaches. Bentivogli discussed about the importance of MT quality evaluation in deciding whether to use MT or not, and which system to select. She also described the various evaluation methods along with their pros and cons.

Industry people Tony O’Dowd, Laura Rossi and Konstantin Savenkov talked about KantanMT, a use case of MT in patents, and MT evaluation from an industry perspective respectively. All three lectures provided useful insights around MT.

The lecture day came to an end with a panel discussion among Renato Beninatto (moderator), Diego Cresceri, Tony O’Dowd, Laura Rossi, Paloma Valenciano (panelists). All active industry professionals shared their points of view about what skills translators should possess in our highly technologized industry.

During the labs we had the opportunity to attend hands-on courses on SDL Trados Studio 2019, MateCat, Smartcat, BootCat and MultiTerm, as well as focus on MT post-editing requirements and practical tips. I attended a lab on post-editing with Smartcat (led by Diego Cresceri), and another one on the use of such terminology tools as BootCat and MultiTerm (led by Claudia Lecci).

Overall, I enjoyed both days of the SATT 2018, was impressed by the passion of all my colleagues for our job, was excited to meet interesting people from our industry, and got to know as much of the wonderful city of Milan as I could on foot!

Congratulations to all those involved in the school’s organization!

I am looking forward to attending the SATT 2019 edition!

What to keep in mind when assigning your first post-editing task

by Dimitra Kalantzi , Linguist at Commit

Maybe your business or translation agency is toying with the idea of experimenting with Machine Translation (MT) and post-editing. Or maybe, after careful thought and planning, you’ve developed your own in-house MT system or built a custom engine with the help of an MT provider and are now ready to assign your first post-editing tasks. However simple or daunting that endeavor might seem, here are some things you should bear in mind:

  1. Make sure the translators/post-editors you involve are already specialized in the particular field, familiar with your business or your end-client’s business and its texts, and willing to work on post-editing tasks. Involving people with no specialization in the specific field and no familiarity with your/your client’s texts, language style and terminology is bound to adversely affect your post-editing efforts. Ideally, the post-editors you rely on will be the same people you already work with, trust and appreciate for their good work.
  1. Forget any assumptions you might have about the suitability of texts for MT post-editing. For example, IT and consumer electronics are often among the verticals for which custom MT engines are built, and it’s usually taken for granted that software texts are suitable for post-editing purposes. However, this might not hold true for all your software texts or even for none at all, and should be judged on a case-by-case basis. For instance, some software texts contain many user interface (UI) strings that consist of a limited number of words (in some cases only 1 word) and are notoriously difficult to translate even for professional translators, especially when the target language is morphologically richer than the source language and there’s no context as is often the case, leading to a multitude of queries. It would seem that such texts are hardly suitable for post-editing or should, at the very least, be not prioritized for post-editing purposes.
  1. Define your MT and post-editing strategy. If your overall goal is to get the gist of your texts and you’re not concerned with style and grammar, then light post-editing might be right for you (but you’ll always need to clearly specify what constitutes an error to be post-edited and what falls outside the scope of post-editing, which might be tricky). If, on the other hand, you’re after high-quality translation and/or the output of your MT system is (still) poor, then full post-editing might be best for you. Also bear in mind that post-editing the MT output is not your only choice. In fact, instead of giving translators/post-editors the machine translated text, you can provide the source text as usual in the CAT tool of your choice and set the MT system to show a suggestion each time the translator opens a new segment for translation.
  1. Offer fair prices for post-editing. As a matter of fact, the issue of fair compensation and how post-editors should be remunerated for their work is still hotly debated. Some argue for a per-hour rate, others for a per-word rate. Some believe that post-editing always involves a reduced rate, for others it means a normal, or even increased translation rate. It all depends on the type of post-editing used (light vs full, normal post-editing vs translation suggestions), the quality of the MT output and its post-editability, the suitability of a particular text for post-editing, the language pair involved, etc. And, of course, translators/post-editors should be paid extra for providing further services, such as giving detailed feedback for a post-editing task.
  1. Last but not least, if you’re a translation agency, you should always have the approval of your end-client before using MT and post-editing to translate their texts. It also goes without saying that if you’ve signed an agreement with a client which forbids the use of any kind of MT or if the use of MT is expressly forbidden in the purchase order accompanying a job you receive from a client, you should comply with the terms and conditions you’ve accepted and should not make use of MT.

Post-editing MT output is by no means a straightforward endeavor and this post has barely touched the tip of the iceberg. Let go of our assumptions, find out as much as you can, involve everyone in the new workflow and ask for their honest feedback, be ready to experiment and change your plans accordingly, and let the adventure begin!

4 tips for getting started with Machine Translation

by Dimitra Kalantzi , Linguist at Commit  

There is no doubt that Machine Translation (MT) is nowadays one of the major trends in the translation and localization ecosystem. Everyone is talking and debating about it in social media, blogs, newspapers and at conferences and almost everyone, including businesses, government bodies, translation agencies, technologists and even freelance translators, is trying their hand at it. If your business or translation agency is also considering getting on the MT bandwagon, you might find the following tips useful:

  1. Remember that MT is an investment and should form an integral part of your localization and overall business strategy. That is, unless you have your own IT/NLP (Natural Language Processing) department or are big enough to set up such a department, you’ll have to turn to the pros, in this case MT providers. With their experience, they‘ll help you determine what your needs are and how best to fulfill them in terms of system (rules-based, statistical, neural, hybrid), languages, types of texts, confidentiality, availability (onsite or in the cloud) and pricing, among other things.
  1. Make your market research as thorough as possible. You might be surprised, but as you’ll find out the market is rather huge with lots of alternatives on offer. Ask around and more importantly, ask from each MT provider you contact to provide you with a list of criteria they consider the most important in choosing an MT solution. This way, you’ll be able to collate the information you gather into a single list of criteria that are important to you and make an informed decision based on your own needs, capabilities and aspirations.
  1. Set realistic expectations. No MT system will work out of the box, no matter the amount of initial training it receives. You’ll have to invest time and money in order to reap the benefits of MT. In addition, be realistic regarding the adoption of post-editing by your freelance translators and beware of losing your most valued partners. Putting aside the gross generalisation that translators dislike MT and technology in general, many translators are indeed reluctant to take on post-editing tasks for a variety of reasons, the most important of which is the fact that because of the way they are currently practiced by some in the translation industry, MT and post-editing are often viewed as tools mainly targeted at lowering translation rates.
  1. Bring in the translators and/or agencies you work with from the outset, even before committing to MT and a particular system. Their collaboration and input might make all the difference to the success or failure of your MT venture. Bear in mind that although the role usually reserved for translators as far as MT is concerned is that of the post-editor, translators can also be of immense help in other related areas, such as MT evaluation and the maintenance and clean-up of translation memories (TMs) used in the training of MT engines.

Hopefully, these tips will help you in your first exploratory steps with MT. But remember, adopting MT is by no means obligatory and you’ll be able to review your circumstances and decision further down the road. And whether you decide to go down the rabbit hole or not, rest assured that your trusted Commit linguists are here to help you deliver your products and services, as well as market your brand in the local language, and who knows, accompany you on your MT journey.

What is Machine Translation and how can your business benefit from it

googletranslate

by Eftychia Tsilikidou, Project Coordinator at Commit

A question we are often asked as Language Service Providers is whether we use Google Translate in our work. This comes as no surprise as Google Translate is the most popular and well-known Machine Translation engine and many users turn to it when they need to understand a text in a language they do not speak. However, using this automated translation engine, one can quickly understand that the quality of the output can vary. Sometimes it will seem to work fine, giving results that resemble a human translation and other times the output is not at all satisfactory. But why is this happening? How does Google Translate actually work?

First, let’s have a look on how Google Translate defines itself:

Google Translate is a free translation service that provides instant translations between dozens of different languages. It can translate words, sentences and web pages between any combination of our supported languages.

When Google Translate generates a translation, it looks for patterns in hundreds of millions of documents to help decide on the best translation for you.

In order to explain the above statement more clearly, think of any Machine Translation engine as a massive “pool” that keeps inside thousands of millions of documents in all sorts of language combinations. Parallel texts, translated by humans, are gathered from different online sources and stored into this “pool”. Whenever users enter text that needs to be translated into a language they do not understand, the machine, through various processes, searches for matching patterns from the texts it contains in the pool and brings up the most relevant results based on certain statistical models. The process of creating this pool of training data is called “machine translation education” and there are various technologies in place for this purpose, with Statistical Machine Translation Technology being the most common.

Now, when it comes to high-level professional translation, Machine Translation should be examined as a very useful tool. Translation is a very complex activity that involves analyzing, interpreting and synthesizing elements of text and transferring them via the same process into another language – an ability, computers definitely do not possess, at least for the time being, no matter how many complex models they process. This is the reason why translation agencies rely largely on human translators. On the other hand, the human brain has certain weaknesses as well, as it is unable to obey to many strict linguistic rules.

Given that there is no MT in place that could imitate the function of a human brain and its ability to analyze and synthesize data, and that even human work needs corrections and entails certain weaknesses, a combination of both human input and automated translation systems could lead to some very good results.

Contrary to the very general and vast content used in Google Translate, when Translation Companies choose to build Machine Translation Systems, they can customize them based on domain, language combination and even customer specific needs. The more specific the content entered into these MT systems, the better the results.

But how can Machine Translation help YOUR business?

Our digital and global era has tremendously increased the amount of ready-to-publish content that needs to be translated into various languages as soon as possible. There is a constant demand and pressure to reduce prices, to introduce new services more quickly and effectively, to maintain the highest levels of customer satisfaction at the optimal turnaround time, cost and quality.

Machine Translation systems can help companies facing those challenges, if they are implemented wisely and offered as a complete solution rather than just as a mere translation process. Taking into account that the MT systems can be customized to meet specific customer needs, and combined with Translation Memories and human input, the time for the translation of large volumes can be significantly reduced, leading, consequently, to reduced costs. Projects that would normally require months to be completed following the traditional human translation path, can now be completed within weeks or less with the implementation of an MT system.

Another circumstance to consider an MT system as a translation solution is when you are faced with a large volume of content and need to get a rough idea in order to decide which content to translate.

However, Machine Translation Systems cannot be used for all kind of texts and in every situation.

Content that involves translations of technical documentation, manuals, software interface, help content as well as automotive, mechanical, medical and legal documents are just some examples of texts with repetitive patterns, specific styles and certain rules, which can effectively train an MT system and subsequently be used for the translation of new content.

The main purpose of technology is to ease people’s life and raise its quality. Machine Translation cannot replace human translators, since language is live and constantly evolving. However, it can certainly aid the entire translation process and, if used wisely, can provide significant benefits for both translators and their customers.