Daniel Halliday
Sep 29 · Last update 2 days ago.
Will machine translation become sophisticated enough to replace human translators?
Research into machine translation goes back to the 1950’s, however accurate computer aided translation remains elusive and translation is still a mostly human endeavour. This has led some scholars to question the possibility of achieving a fully automated high quality machine translation of any kind. Will technological advances help us reach this linguistic milestone?
Stats of Viewpoints
Deep learning is allowing AI to develop to the point where it threatens to overtake human translators
0 agrees
0 disagrees
Accurate translation will remain a human venture
1 agrees
0 disagrees
Machine translation could augment the translation process making human translators more efficient
0 agrees
0 disagrees
Viewpoints
Add New Viewpoint
Deep learning is allowing AI to develop to the point where it threatens to overtake human translators

'Deep learning' refers to a range of machine learning algorithms and methods that allows multilayered, nonlinear processing to occur in a similar way to how a biological network of neurons function, for example that of the human brain or central nervous system. They allow multiple layers of abstraction as algorithms use a cascade of non-linear processing units to extract and transport multiple layers of information, a style of data processing that would lend itself to the subjective nature of linguistic communication. Deep learning is currently aiding artificial intelligence programming in developing competent machine translator software according to One Hour Translation company CEO Ofer Shoshan.

According to Shoshan, presently only a small number of corrections need to be made to such deep learning machine translations to make them accurate, and in the future he believes machines will largely replace humans as translators outright. Shoshan, who deals with various Fortune 500 clients, claims that such software requires on average only 10% human correction, a 70% increase of efficiency from just two years ago before the large scale implementation of deep learning methods. Projecting forward it would seem that such advances in artificial intelligence and machine learning pose a real threat to human translators in the very near future.

forbes.com/sites/bernardmarr/2018/08/24/will-machine-learning-ai-make-human-translators-an-endangered-species/#d7c789a39023

Agree
Disagree
Latest conversation
Daniel Halliday
Jan 22
Approved
DH edited this paragraph
According to Shoshan, presently only a small number of corrections need to be made to such deep learning machine translations to make them accurate, and in the future he believes machines will largely replace humans as translators outright. Shoshan, who deals with various Fortune 500 clients, claims that such software requires on average only 10% human correction, a 70% increase of efficiency from just two years ago before the large scale implementation of deep learning methods. Projecting forward it would seem that such advances in artificial intelligence and machine learning pose a real threat to human translators in the very near future.
Accurate translation will remain a human venture

After nearly seventy years of development machine translation is still in its infancy, with computer aided translation only managing fairly basic word for word translation or statistical prediction of relevant meaning. Linguist Yehoshua Bar-Hillel was the first in academia to work on machine learning full time in 1952, and organised the first international conference on the subject in the same year. Before his death in 1975, Bar-Hillel expressed doubt on whether automated high quality machine translation would ever be feasible.

It could be argued that a machine or computer would never be able to handle the very high level of ambiguity, nuance and inferred meaning as is common in natural human speech. Not to mention the vast array of accent and vocal variation, a machine may be able to recognise a human speaking clearly and slowly with an unnatural voice, but understanding natural speech will arguably never come to a machine. Additionally, language evolves rapidly, with slang terms being introduced to a language regularly and such language changing rapidly as it either forms a permanent place in the language or falls out of favour.

mt-archive.info/Bar-Hillel-1953.pdf

Agree
Disagree
Latest conversation
Daniel Halliday
Jan 22
Approved
DH edited this paragraph
It could be argued that a machine or computer would never be able to handle the very high level of ambiguity, nuance and inferred meaning as is common in natural human speech. Not to mention the vast array of accent and vocal variation, a machine may be able to recognise a human speaking clearly and slowly with an unnatural voice, but understanding natural speech will arguably never come to a machine. Additionally, language evolves rapidly, with slang terms being introduced to a language regularly and such language changing rapidly as it either forms a permanent place in the language or falls out of favour.
Machine translation could augment the translation process making human translators more efficient

AI and advancements in technology will revolutionise translation but not to the danger of human translators, they may just augment the process of translation. Arguably, future advances in technology will become more sophisticated and integrated, but understanding the wider context of speech may remain outside of the scope of machine learning altogether. Therefore it is more likely that quick machine translation could serve the human translator by carrying out the monotonous bulk of translation, while a human translator would still be needed to correct for nuances, deeper context or more accurate phrasing.

At present Google is developing neural machine translation (NMT) technology, utilising AI machine learning to pick up on a larger range of nuances and subtleties that have eluded machine translator programming so far. However in February 2017 at a translation competition at Sejong University in Seoul, Google’s NMT had the opportunity to compete against human translators from Korea’s International Interpreters & Translators Association and ultimately lost, with 90% of the NMT’s output being grammatically awkward. It seems that contrary to purely mathematical tasks, language has too many subjective selections that require a human touch, leaving applications of even NMT needing human input to be useful tools.

digitalistmag.com/digital-economy/2018/07/06/artificial-intelligence-is-changing-translation-industry-but-will-it-work-06178661

Agree
Disagree
Latest conversation
Daniel Halliday
Jan 22
Approved
DH edited this paragraph
https://www.digitalistmag.com/digital-economy/2018/07/06/artificial-intelligence-is-changing-translation-industry-but-will-it-work-06178661
Translate