I would say ‘no comment’ to the decision by the U.S. Citizenship and Immigration Services (USCIS) to use Google Translate in vetting refugees’ social media activities, but, as a professional translator, this is a slap in the face. For a link to the referenced article, please click here.  Many a translator would still be offended should USCIS use non-professional linguists. That would, however, be easily understood based on the overwhelming number of applicants and the fact that the linguists would at least be humans.

Machine Translation (MT) is beneficial both to the translator and the end-client in that it improves translation speed, reduces costs and helps memorize terms for future use. There are diverse opinions within the translation and interpretation industry vis-à-vis MT. While some view it as a threat to the profession based on a potential loss of work and professional prestige, others consider it a job aid and are happy to serve as human editors in Machine Translation Post Editing (MTPE).  

Machine Translation should not however, be confused with Computer-aided translation or Computer Assisted Translation (CAT) which involves the use by a human translator, of computer hardware or tools to support and facilitate the translation process. These include among others, translation memory, terminology, project management, dictionaries and quality assurance tools. They improve speed, quality and consistency in terminology and style. Some CAT tools include controlled Machine Translation. Examples of the range of tools, could be found on this link.  

As a modern-day translator in the era of artificial intelligence, I commend the efforts and progress made towards improving Machine Translation in general and Google Translate in particular. The evolution from Classical or Rules-based MT (uses customizable grammar and language rules), to Statistical MT (uses models that learn to translate from examples) to the acclaimed Neural MT (uses a large neural network) has produced far-reaching results and is revolutionizing the industry. 

Despite all the progress made so far, Machine Translation is very unlikely to replace the human translator. It is yet to be proven capable of circumventing semantics, polysemy, colloquialisms, syntax, ambiguity, context, fluidity/flexibility of human language, culture… Evaluating snippets from social media would be a daunting task especially for a machine, without a brain. Apart from the linguistic and technical challenges, there are information security and privacy concerns with using MT, even for secure closed platforms. MT engines are still very far from being able to support the about 3900 written languages of the world. Google Translate for example, supports just about a hundred and the quality of the output varies greatly per language pair and content type. Machine Translation has been proven to produce better output with straightforward technical or scientific texts for example. Like literary and other complicated content, social media content is a different ball game. Even in those areas of relative success, MT output still needs to be edited by a human translator to correct errors: lexical, morphological, syntax, semantic, etc.

Based on the foregoing, it is unfathomable for the USCIS or any government agency to use MT, without a human translator’s input for such consequential work. It could be argued that the essence of the exercise isn’t for high quality translation but for understanding a message. Nonetheless, a change in meaning could be catastrophic. MT isn’t efficient or accurate enough to replace human translators.