Humans vs. Machines: Will Google Translate Replace Professional Translators?
Ray Kurzweil, an inventor and futurologist, once predicted that machines capable of performing automated translations would appear by 2019. He believed that translation machines would replace human professionals by 2029, though it would still be necessary to learn foreign languages.
The first steps on the way to a new reality have already been made. In early 2017, Google introduced a new neural, network-based translation tool. A neural network is a self-trainable artificial intelligence system that functions according to the same principle as a brain (a mouse’s brain is used as a reference model for now, though scaling up shouldn’t be difficult).
During a presentation of the new interpreting machine, Sundar Pichai, CEO of Google, used the following aphorism by Borges as a sample phrase to be translated:
“Uno no es lo que es por lo que escribe, sino por lo que ha leído”.
The old Google Translate would have translated this line into English as:
“One is not what it is for what you write, but for what you have read”.
The new Google Translate edition suggests the following version:
“One is not what one writes, but rather what one has read”.
Machine translators are “learning” from the language of fiction. Considering the multiple-word senses and shades of meaning and the versatility of artistic devices, this achievement is hard to overestimate, and the result is truly outstanding. The question is whether Google Translate is a potential replacement for human translators and how scientific progress is going to reshape the market.
Valuable Benefits for Business: Speed, Automated Processes, and Lower Translation Costs
Automated text processing solutions for businesses are already being developed: for example, ABBYY LS and Xerox have recently launched a joint product, Xerox Easy Translator Service, which enables multi-functional device s not only to recognize, but also to translate, scanned texts.
For the time being, however, neither Google Translate, nor Xerox Service can guarantee a sufficient level of precision in translating highly technical texts. So, when a machine translation is used, who should be responsible for the quality of the translation?
A translator’s job implies, above all, a responsible approach. The wrong interpretation of the terms of a multimillion contract could entail significant losses for the client company and trigger a court action.
Humans vs. Machines?
The evolution of neural networks suggests that lower-grade translators may be the first ones to fall out of demand, as the existing translation tools are already capable of processing printed material and will have their “skills” upgraded in the years to come. Human involvement is thereby being reduced to a minimum, as the program will perform the bulk of the work. However, like an inexperienced translator, a machine needs to be closely monitored to prevent lapses. This will be the responsibility of translators, editors, and industry experts for the following reasons:
- Neither a machine nor its developers can be held accountable for a poorly matched synonym or a blurry definition. Responsibility will lie with the translator or editor.
- An automated translator “learns” from materials provided by its developers. In the case of Google Translate, the learning material is being derived from fiction and texts used by the Canadian Parliament. However, for the purpose of fine-tuning the machines to accurately translate legal documents, one needs to use real life contracts, memorandums, and other documents. Yet even sanitized versions of such materials, with company and individual names deleted, could contain proprietary information. Will legal firms agree to sacrifice confidentiality and sensitive information to help Google?
While machine-translation technologies are still in their development phase, humans working with legal, technical, and medical texts are unlikely to become obsolete. Translation-bureau clients around the world are not yet ready to entrust the translations of multimillion contracts to artificial-intelligence programs.
Humans and Machines Working Together
A number of automated tools are already being used by translators.
With large projects of 2000 pages or more, we use Translation Environment and CAT tools, which are technical instruments enabling the use of unified terminology databases and translation memories within the limits of a single project. The memory stores every text processed within the project’s framework. Thanks to these tools it is possible to ensure the consistent translation of terms. For example, the term “article”, when used in an agreement, may be legitimately translated into Russia as both “пункт” and “статья;” however, once one of these options has been selected, all of the translators and editors participating in the project will use the same term throughout the text.
Programs similar to those mentioned above simplify work with texts, yet require human involvement. They are of help when it comes to the translation of individual words or lines. A phrase matching another already translated phrase at 70–80%, however, will need to be translated anew.
Things may look easier with Google Translate, but the problem of data security resurfaces once again: the translating prompts generated by the system as a result of reworking real life documents will now be available to all program users! By contrast, translation memory systems are available exclusively to the team working on the project.
In our opinion, the development of corporate document-translation technologies is largely client-driven. The reason for this is not research funding alone – client companies are also in the position to further inform the locally used translation environment tools by making their documents available as source texts. It is possible that the future of automated translation is in the hands of large corporations seeking to optimize their budgets. Conversely, translation bureaus may opt for seconding their experts and becoming technology partners of translation software developers.
Nevertheless, even having resolved the data security problem, it will be impossible to entrust translation projects fully to machines. Translators will still be required as production editors who are responsible for the accuracy and precision of a translation. An article by the New York Times recently claimed:
“It is important to note, however, that the fact that neural networks are probabilistic in nature means that they’re not suitable for all tasks. It’s no great tragedy if they mislabel one percent of cats as dogs, or send you to the wrong movie on occasion, but in something like a self-driving car we all want greater assurances.”
Machine translation is becoming a new reality, which is a positive development. Meanwhile, translation bureaus are likely to acquire new roles as providers of expert skills and technology partners to software developers.