Neural Machine Translation has moved into a primary spot in the world of language translation services. The software promises to produce more accurate and meaningful media and marketing translations than previous processes.
Artificial Intelligence and Machine Learning
Artificial intelligence (AI) is a broad category label that essentially refers to computer-generated solutions that generate results normally produced by people. Various techniques fall under the umbrella of AI, and for years, language translation software has worked through one such approach known as statistical machine learning.
Statistical machine translation takes a linear approach to translations. Essentially, dictionary translations of words are entered into the system, and the software produces a rote translation on a word-for-word and short phrase basis.
The enhancement of this process lies in a statistical decision-making process: the software chooses the translated word based on statistical likelihood of correctness. The software learns through human intervention confirming its translation, and as more corrections are entered into the system, the software can reduce its statistical errors for more accurate translations each time.
The challenge of statistical translation is the lack of contextual understanding, idioms, and cultural nuance. The software can get close to maintaining the meaning of a sentence, but it can have difficulty with less formal language and differences in syntax from one language to another.
Enter Neural Machine Translation
Also falling undering the AI label, neural machine translation (NMT) supersedes statistical machine learning by orders of magnitude. Instead of a strictly algorithmic, statistical approach to translation, NMT forms a network of related layers that mimic the way the human brain processes information.
Each piece of data that passes through the system is weighted for appeal. That weighting then influences the next interaction between the layers until a final decision is made. The process more closely resembles human thought than statistics do.
The leap in translation accuracy with NMT is due to contextual awareness. By comparing and understanding the relationships among words as opposed to simply identifying definitions, NMT teaches itself context, which in turn produces more natural and exact translations.
Major Advancements in NMT Are Reaching the Public
As technology connects the world, more sophisticated ways of communicating across language barriers are in demand. Major players in the tech world are improving their services to capture the increasingly vast amounts of information and data that a connected world creates and makes available.
Google, for instance, has its own version of neural translation called Google NMT. The goal is to replace Google Translate’s statistical approach with the neural. In 2017, Google NMT made considerable breakthroughs in translating from one language to another.
In a major achievement, it accomplished what’s known as “zero shot” translation. To date, most translations have to be filtered through English (i.e, in order to go from Spanish to French, the text is first translated into English). Google’s neural machine taught itself how to go directly from one language to the other without an intermediary translation.
Meanwhile, Facebook has been focusing on the social end of language translation, which typically involves less formal speech and grammar rules. Meaning becomes more important in conversational language, and Facebook’s neural network–which is fully deployed on the platform–now takes into account context and syntax.
Not to be left out, Microsoft offers Translator Hub, an app that lets users build their own dictionaries and input business jargon directly into the system. From there, the AI-driven app can produce more meaningful and accurate translations within industries.
Human Translators Have Unique Opportunities with NMT
For language service providers (LSPs), NMT does not represent the end of their industry, but rather an enhancement to their services. By positioning themselves as experts in the growing NMT markets, LSPs can bridge the language gap for their clients in new, more sophisticated ways.
For content marketing, the field remains an exciting opportunity to expand their expertise. NMT has yet to tap into the emotional components of good writing that intrigue readers and consumers.
Using NMT is an improved first step for translation and, depending on the system used, may be the perfect solution. For the nuance of voice, style, and and the “human touch,” human translators still need to step in. For marketers with multiple language and culture experience, NMT can be used as a springboard for employing their knowledge for subtlety and sensitivity in local markets.
Localized knowledge in business and culture will still be of paramount importance, as will expert consultation services in consumer marketing, business, and law.