In recent years, artificial intelligence has allowed the development of tools that automatically translate documents. These systems make it possible...
4 min read
12/11/2014
With the growing popularity of digital content and customers' expectations for quality translations in their local language, translation companies experience great pressure from international businesses to localize content to gain optimum online presence. Human resources (translators) alone cannot meet the emerging translation needs of global businesses—technology must help in this process. Machine translation is the key to meeting these needs and aiding global businesses in staying ahead in a competitive arena.
First, it is important to understand what machine translation is and how it works. Machine translation is a process in which content is translated with the help of a computer. What is interesting about the process is that the translation of massive amounts of data requires no human intervention. Engines are set up and configured with training data or with dictionaries and rules. The thinking (i.e., the translation work) is done in seconds or minutes by algorithms and computer scripts which relate one language to another either using statistics or rules that match one language to the structure of the other. Machine translation work is mainly categorized into two types: rules-based systems and statistical systems. A rule-based system uses a combination of language and grammar rules along with dictionaries for common words.
You can count on this system for consistent translation, but historically it has been criticized for making translations sound too stiff and corseted and many times incomprehensible as they reflected the structure of the source language too much. The majority of translation systems were laboriously developed creating rules and examples between one language and another, working out the parallels and relations between them. This was fine when languages were related or at least belonged to the same family, but produced not so good results when languages were not related at all. Imagine trying to develop a Finnish-to-Portuguese translator. Development times and costs were massive, just as the linguistic resources required. One of the major problems is that eventually some rules contradicted each other - a ceiling had been reached. On the other hand, a statistical system does not take note of any language rules. 
Moses is a popular Open Source Translation Software
A statistical system, such as the ones used by Google or Microsoft Bing requires large amounts of parallel data (corpora), which in turn requires "cleaning" (pruning and discarding the parts that are not fit for machine learning, which is considered "noise") and customization (giving more weight to certain expressions, for example, or adding particular data from a specific domain).
A statistical machine translation system can translate large amounts of data fast and efficiently, and it is the technology behind so many new translation offerings we have on the web, mobile devices, etc. This type of machine translation system also delivers more fluent-sounding translation than consistent translation due to the fact that they analyze chunks of data in groups of 3, 4, or 5 words (optimally groups of 5, but this varies across languages).
Hybrid machine translation is a combination of the above two methods of developing machine translation programs. It is characterized by using multiple machine translation approaches but within a single machine translation system. The motivation for developing hybrid machine translation systems stems from the failure of any single technique to provide a 100% satisfactory level of accuracy, and so by combining the best of techniques, researchers have hoped to solve the shortcomings of rule-based technologies and statistical (or data-based) technologies. It has been proven in research and reported in many publications that a hybrid approach can improve the accuracy of the final translations in certain language combinations, and there are several popular machine translation systems that employ hybrid methods.
PangeaMT published two academic papers on hybridization for the English-Japanese language pair in the Asian Association for Machine Translation, which proved that some hybridization techniques improve the output when language pairs are very remote, like in the case of Japanese and English. The results were also presented at the Japan Translation Festival in 2011.
Our company's efforts on hybridization received the acknowledgment of the European Union, and we are now part of an EU research program on hybrid approaches to machine translation (EXPERT). Organizations operating in the international market working on sensitive projects often choose some kind of machine translation API or connector to translate low-priority content automatically. Typically, this is content like user comments or user-generated content which the organization does not want to translate and it is not core to the company documentation effort, yet it can be useful to a community of users. If you are a company with a large amount of data to be translated and it has to be done in a short time, machine translation services must be high on your list.
Free machine translation services can leak data, as it was reported by TCWorld and the European Union is taking measures on security in that respect. Therefore, a good machine translation system should be hosted and completely secured if it is to handle private data, for example, data that may be sensitive from an industrial point of view for intelligence services, the military, hospitals, insurance, etc. As a user, you must have the complete peace of mind that your information is not disclosed to third parties. Sensitivie content must be secured from getting into the hands of a careless or unscrupulous translator. This way, confidentiality is completely secured. Large corporations, e-commerce stores, government agencies, and content writers are getting the benefit of hiring and optimizing machine translation services. There are many more added advantages of using machine translation. They result in enhanced productivity, faster and cheap translation.
You need not have to muddle your head in finding whether the vendor has good team of translators or not. Nor do you have to worry whether they have the expertise to understand your market or not. All your translation work will be completed easily and without any hassle. They are flexible enough to suit any type of project. Hope you are convinced now why you should use machine translation services for getting your translation job done. The only thing that you need to consider here is choosing good machine translation services.
You should ensure that they use the latest technology in their core job. Plenty of vendors have an online presence, but Pangeanic works on technology transfer via its division PangeaMT. You can visit the website to learn more and try the technology for your own purposes, even build new business thanks to our API. Read the reviews of past clients and presentations with large corporations in the last few years. You will get an insight into the customer service of the chosen vendor. The popularity of machine translation services is increasing by leaps and bounds. Why wait any longer? Call a Pangeanic team member to see how we can best help you with machine translation when human translation cannot satisfy all your needs.
2024 Update
Do you know that Statistical Machine Translation was superseded by Neural Machine Translation systems? The ECO Platform provides complete independence to clients when it comes to managing their technology affairs. You can add your data in TMX format in complete privacy and choose the level of impact (aggressiveness) in the re-training of the engine, create clones from your engine, and further specialize them for a different client or use. Pangeanic is an active player in the AI space: we released our custom LLM ECO Chat in February 2024.
