[et_pb_section bb_built=”1″][et_pb_row][et_pb_column type=”4_4″][et_pb_text _builder_version=”3.6″]
One of the earliest goals for computers was the automatic translation of text from one language to another.
Automatic or machine translation is one of the most challenging artificial intelligence tasks given the complexity of human language. At first, rule based or statistic based methods were used for best translation. More recently, neural network models are being used to achieve high end results.
Basically in machine translation, a computer takes already given symbols and texts in one language and is tasked to turn them into symbols and texts in another language. However, there is no single best translation from language to language. The human language is much more complex than a computer can handle. Language has natural ambiguity and flexibility making it difficult for a machine to accurately translate.
Machine translation is evolving. Statistical Machine Translation uses statistical models to be best translate sentence A into sentence B. It’s statistical, given sentence A, the error in translation is minimized by choosing sentence B, the most probable. The source is data driven. It requires a large amount of examples in both languages to accurately find the best translation. The issue with Statistical Machine Translation is its narrow focus, the translation process goes sentence to sentence, thus losing the broader nature of the text. In Statistical translation the computer loses the syntax that would be easily recognizable by a linguist.
Neural Machine Translation is another type of evolving Machine Translation. Neural Machine Translation creates a neural network model to learn and source from. What’s cool about this is a single system can be trained on a source/target text. Similar to our own neural systems, the machine allows itself to learn and remember movements, context, etc. from a text to apply to future translations. However, the machine does run into issues when the translation need becomes too long, creating errors in translated words or skipping words overall. Although effective, the neural machine translation systems still suffer some issues. The machine has a hard time recognizing rare words and scaling to a larger vocabulary. Neural machine translation machines also face an issue of speed; their processing time is much slower than desired.
Machine translation is a modern technology that everyone can admit is pretty cool. The development and success of these machines can and will alter the language industry greatly. However, for now, the best translations out there are those done by humans.