Computer science has long dreamed of using computers to translate. Even so, machine translation has only become a viable tool over the last ten years. With the advancements in natural language processing, artificial intelligence, and computing power, this technology has become increasingly useful. In layman’s terms, machine translation is the process of automatically translating content from one language to another.
Machine translation saw great leaps in improvement during the 1950s, but references to the subject date back to the 17th century. Unfortunately, the complexity of the task was much greater than early computer scientists had anticipated, requiring enormous amounts of data processing power and storage well beyond the capabilities of early machines.
Only in the early 2000s did the software, data, and hardware become capable of doing basic machine translations. Early developers used statistical databases of languages to teach computers how to translate text.
In 2016, Google had an experimental team testing the use of neural learning models and artificial intelligence (AI) to train translation engines. When the small team’s methodology was compared to Google’s current statistical machine translation engine, the AI proved faster, more accurate and effective. Google adopted neural machine translation as its primary development model afterwards. Microsoft and Amazon soon followed suit, and machine translation became a viable addition to translation technology.
Let’s take a look at different types of machine translation:
Statistical Machine Translation (SMT)
This method uses statistical models that rely on analyzing huge volumes of bilingual content. Basically, it searches for a match between a word from the source language and a word from the target language. Google Translate is a great example of SMT implementation. Their system is extraordinary for basic translations, but its biggest drawback is that it doesn’t take context into account. The output texts can often be misleading, since they are translated as separate keyword matches, rather than a single string, leading to incorrect translations.
Rule-based Machine Translation (RBMT)
RBMT translates texts by using the grammatical rules of both languages. It performs a grammatical examination of the source language, then analyzes and restructures the text in the target language to generate the translated text. In spite of its scientific approach, RBMT requires extensive editing afterwards. Its heavy reliance on dictionaries implies proficiency can only be achieved over time and supervision is mandatory.
Hybrid Machine Translation or (HMT)
HMT, as the term implies, is a mix of Statistical Machine Translation and Rule-based Machine Translation. It uses a translation memory to generate a higher quality result, but it still requires a translator to proofread the text, since its large database can’t always account for context. Unusual phrasing, extensive usage of slang words and inversions in the source text can mislead the algorithms, resulting in a nonsense text.
Neural Machine Translation (NMT)
A neural machine translation system uses artificial intelligence to learn languages and continuously improve their knowledge. It’s meant to mimic the neural networks in the human brain, thus reaching a lower margin of error. NMT is the current state-of-the-art technology in machine translation and achieves the highest quality results. The main advantage of NMT is that it provides a single system that can unravel both the source and target texts. Many multinational institutions now use NMT engines to aid in multilingual communications.
What are the benefits of machine translation?
Simply put, the main advantage of machine translation technology is speed, since computer programs can quickly translate vast amounts of text. A human translator may be more accurate, but can never match a computer’s speed. Another benefit of machine translation is its capability to learn important words and reuse them wherever they might fit.
There are alternative solutions, like Computer-assisted translation (CAT) software, that are quicker but have an entirely different approach to translating. We’ll discuss them in a later post.
We hope you find this information interesting and useful. There will be a sequel very soon. The world of MT is becoming increasingly vast!