Machine translation has long occupied a place of contention among translators. However, machine translation will only improve. For translators, the key to maintaining the crucial human element in the translation industry is to stay updated on the most current translation technology available.
A fellow translator advised that translation technology is a tool that should be used, a tool that makes our jobs easier and faster, and more accurate. Instead of fearing how the technology might replace humans, we should instead focus on how we can make the technology work for us, and on how it can be a balm for manual translation irks.
One of the newest and most advanced forms of machine translation (MT) involve neural networks, which holds some fascinating prospects for making the translation process less laborious.
Neural Machine Translation (NMT) became more efficient than Statistical Machine Translation (SMT) or classical machine translation. Due to the NMT’s ability to learn directly from the source text and target text, NMT does not need a pre-loaded bilingual corpus pre-programmed into its software. NMT is able to study any text and decipher how the source text is associated with the target text without the intermediary of a bilingual corpus. It’s software also does not need to be continuously modified in order to recognize grammar rules and exceptions.
This is an advantage over SMT, which always needed a bilingual corpus to learn from. However, SMT became inefficient when it needed to translate into rare languages because a bilingual corpus for rare languages was usually difficult to find.
NMT is able to learn rules by analysis, which then teaches it to decipher context and avoid ambiguity, leading it to be able to translate faster and more accurately. The target text output then becomes a product that is meaningful rather than being merely full of words.
NMT works by partly having neural networks that are connected together with nodes, resembling a human brain. The neural networks learn and retain information. The more it learns, the quicker it’s able to recognize the relationship between concepts.
But similar to the human brain, the NMT networks cannot always store new information quickly enough, causing it to occasionally skip rare words. NMT is also not advanced enough yet to recognize and make calculated decisions such as when translation or transliteration should be used. Sometimes meaning can still be misinterpreted, or whole chunks of text could be omitted. The important decisions, such as choosing whether to translate or transliterate or omit parts of a text, are still a very humanistic aspect of translation, and which humans will continue to be responsible for.
Though NMT will continue to advance, its advancement could only have a positive effect on the translation industry. NMT’s translation ability might be approaching that of a human (such as deciphering context and translating for meaning), but what the NMT will mostly replace is the hard, tough labour work of translation rather than the intellectual and curious side of the translator. The NMT’s job is to complement the relationship that humans have with machines, while the human translators’ time is freed to continue to develop and harmonize the relationships between people working together on the same project. People will continue to gatekeep the quantity of work that machines produce and continue to improve the quality of the work that machines and people will produce together.
Brownlee, Jason. “A Gentle Introduction to Neural Machine Translation.” Machine Learning Mastery, 29 Dec. 2017, https://machinelearningmastery.com/introduction-neural-machine-translation/.
Kravariti, Alexandra. “Machine Translation: NMT Translates Literature with 25% Flawless Rate.” Translate Plus, 9 Feb. 2018, https://www.translateplus.com/blog/machine-translation-nmt-translates-literature-25-flawless-rate/.
Zetzsche, Jost. “Neural Machine Translation.” The ATA Chronicle, https://www.ata-chronicle.org/highlights/neural-machine-translation/#sthash.T90xM77n.Oin7mW77.dpbs.
Leave a Reply