Skype Translate: Policy and Training Considerations
At Microsoft’s Natural Language Processing (NLP) laboratory, researchers are test-driving a new generation of natural language processing tools that promise to transform work in the coming decades. Among the most widely publicized and exciting breakthroughs is the lab’s much-hyped Skype Translator, which promises to provide real-time translation to collaborators working across languages, such as English, Mandarin, Spanish and Italian. But will the hype translate into workable cross-linguistic tools in the near future, and if so, what are the implications for the future of work?
What is Natural Language Processing?
NLP, if achieved, would enable computers to replicate human communication on every level. With the right software, our computers would be able to analyze, understand and generate languages in the same way we analyze, understand and generate languages. Unfortunately, even as Microsoft announces programs, like Skype Translator, we remain far away from achieving NLP. To consider how far, consider the qualifications of the average human translator or interpreter.
Human versus Machine Translation/Interpretation
First, translation refers to the practice of moving a text from one language to another language. Interpretation, on the other hand, happens in real time (e.g., at a conference or meeting). Skype Translator is both a translator and interpreter, since it enables you to hear your interlocutor’s words spoken in your own language and enables you to read a written transcript of the conversation. But if you think that this means that translators and interpreters are no longer required, think again.
The United Nations (UN) employs hundreds of highly skilled translators and interpreters to keep the organization running. Qualifications for UN translators and interpreters are rigorous. In order to even write the UN’s demanding qualification exam (if you pass, you are not guaranteed a position but considered eligible to apply), you must be fluent in at least three languages and with few exceptions, all of the organization’s translators and interpreters hold graduate degrees. Thus, while many people speak more than one language, working as a translator or interpreter in a high-stakes environment, like the UN, requires a level of fluency and level of literacy that far exceeds the average bilingual, trilingual or multilingual person and far exceeds the capacity of any machine. In short, in order to replace the UN’s large team of translators and interpreters, we would need machines that can translate documents and interpret in real time across multiple languages with a level of literacy on par with someone who holds a graduate degree in linguistics, literature or a related discipline. In addition, we would need machines that can do all this brain work with the sensitivity of living, feeling people.
Since Alan Turing developed the Turing test, computer scientists have been attempting and failing to develop machines that can communicate like humans and more notably, translate languages like humans. While we have come a long way since the 1950s, NLP remains one of computer science’s biggest challenges. Why? Because machines continue to lack the ability to truly understood language in its cultural context and lack the ability to interpret subtle language cues, such as something said or written in an ironic tone.
Reasonable Expectations for NLP
Despite the fact that there is no indication that NLP will replace human interpreters or translators at the UN—at least not any time soon—there is no doubt that moving forward, a growing number of businesses will adopt Skype Translator (and similar programs, which are bound to appear), in order to carry out business across linguistic boundaries. On the one hand, no longer expecting clients outside the English-speaking world to work in English (a default for many businesses, especially in the US) holds the potential to create a work world that is less ethnocentric and more culturally sensitive. Theoretically, with the right tools, everyone would be working across languages and no one would have a specific linguistic advantage. On the other hand, working across a growing number of languages and cultures—especially with shaky technology—is also bound to generate new types of cultural misunderstandings.
During a recent demonstration for Skype Translate’s Mandarin program, Microsoft’s executive vice president, Harry Shum, became Hairy Shum. While the mistake is understandable and in this case, humorous, one can imagine that in a high-stakes setting, a homophonic error of this nature could have a negative impact. What’s clear is that while embracing machine translation programs may be tempting, it should be done with caution and with appropriate training. Creating a world where everyone can communicate on the same grounds holds the potential to transform work but without appropriate policies and training, it also holds the potential to heighten cross-cultural misunderstandings.