As the business world has gone global, the need for language translation services has increased dramatically. For companies that work in different countries, with different languages and cultures, creating a smooth transition from one language to another can be a challenge.
Human translators have traditionally been the primary means of achieving quality translations. A person knowledgeable in multiple languages with an understanding of local culture provides a critical lynchpin for success.
A failure in an understanding can have disastrous consequences. If you’re in the eLearning industry, for example, a poor quality educational translation can destroy your reputation to the point of causing your business to fail.
For over a decade, Google has been at the forefront of translation technology. Google Translate works with over 100 languages and over a half of a billion people use it everyday. Until recently, the process Google Translate used was a form of machine learning, a branch of artificial intelligence that processes a vast amount of data to understand and produce language translations.
Machine language translation software has become a cost-effective means for businesses to leverage their existing materials into their global markets. The software is faster than human translation, and usually cheaper.
The downside of this method is that it results in too literal an approach to translation, essentially performing a dictionary process to determine word choices. Subject context and cultural nuance cannot be addressed with this type of software. Thus, the need for human translators to review and revise materials still exists.
In 2017, Google switched to a new underlying process for Google Translate called Google Neural Machine Translation (GNMT). GNMT is an evolution in AI programming that mimics the way humans process information: the neural network.
While machine learning takes a linear approach to translation, a neural network works by connecting multiple nodes of information in a web-like structure that looks at relationships among words rather than just a one-to-one translation. The result produces translated material that can take into account the concept of context.
In the AI world, neural networks are referred to as Deep Learning. Where machine learning comes to its conclusions thorough statistical analysis, deep learning uses a type of observational skill in conjunction with machine learning. In other words, it learns. The more information deep learning has to process, the more it finds unique and efficient ways to produce results.
For instance, GNMT, and Google Translate in general, works by translating a language into English and vice-versa. To go from Spanish to French, the software first translates Spanish to English, then the translated English into French.
GNMT goes one step further and has figured out on its own how to go directly from Spanish to French without that English pit-stop in between. GNMT and its neural network has learned something fundamental about language structure that does not need a line-by-line rules approach. Known as “zero shot” translation, the algorithms necessary to perform these translations offer a new mode of efficiency and accuracy to the world of translation and communication.
For businesses, the advent of zero shot translation opens the door to even faster and more accurate translations. For particularly technical matters, a human translator should still be employed for quality assurance. Language service providers often bring subject area expertise to the table, whether for processing patent filings or claims, understanding local building codes, or bringing an awareness to cultural distinctions in human resources management.
By combining deep learning software and human translators, businesses can save costs on underlying translation while freeing their human translators to focus on business and legal matters, content, and process.