Found in Translation

Beam Me Up, Babelfish

---

To find them, look no further than the “three stars”—Star Wars, Star Trek, and
Stargate. You can also see them in Doctor Who, Futurama, and of course, A
Hitchhiker’s Guide to the Galaxy. We’re talking about “universal translators,”
those futuristic machines designed to bridge language barriers between
earthlings and aliens.
Speech translation is not just the stuff of science fiction. It exists today, and
it’s a field that is growing. But machines cannot yet fully replace humans when
it comes to converting spoken language. A human can interpret simultaneously
—listening and speaking at nearly the same time. For now, a machine works
much more slowly. Actually, the machine has to complete three separate
processes. First, a speech recognition program comprehends what was spoken
in one language, converting it into text. Then, using automatic translation, the
written text gets translated into a second language. For the final step, the
machine vocalizes or speaks the translated version of the text.
Because there are so many variables involved, speech translation presents
even more obstacles to developers than text translation. Humans are fairly
adept at looking past a speech impediment or unfamiliar accent, but machines
are not. Just consider the fact that when Apple launched Siri, which stands for
Speech Interpretation and Recognition Interface, it had problems recognizing
and interpreting Scottish accents. Non-Scottish humans may have problems
deciphering Scottish accents, too, but humans are at least capable of making
better guesses.
Humans also know to listen more intently when someone is whispering, but
machines don’t instinctively do this. Humans can block out sound from other
sources—say, a baby crying in the background—but machines have a hard
time with that. If there are two male voices speaking the same language, a
human can usually discern between them quickly. Machines don’t find it so
easy.
But developers love a challenge, so there are plenty of speech translation
products on the market. Google has added speech options to its core translation
product. The U.S. Department of Defense has spent millions upon millions of
dollars over the years on various projects to automate the translation of
speech. There are some promising examples of technologies that do a decent
job when limited to certain settings or specific languages. There are even some
tools that work reasonably well (after significant time spent in training the
speech recognition portion) with a single user ’s voice. Yet, despite plenty of
investment from government organizations and private-sector firms,
automated speech translation today does not even come close to doing what
human interpreters can do.

---

Enabling human beings who speak different languages to communicate with
each other in real time without relying on a human interpreter is one of the
final frontiers of translation technology. Such an achievement would surely
represent the fall of the figurative Tower of Babel. Will it ever happen? Maybe
there’s a reason the humans in the science fiction films and television shows
always seem to speak the same language. Usually, they are also trapped in a
linguistic time warp, speaking a form of American English that strangely
ceased to evolve in the late twentieth and early twenty-first centuries. Perhaps
it’s easier for most people to imagine a world in which humans can seamlessly
communicate with aliens than for them to imagine instantly overcoming
human language barriers.
Speech translation will improve as time goes on. Of that we have no doubt.
In fact, we can envision an array of possibilities for customization—for
example, a technology that can translate from one generation to another. One
that can detect and filter out statements that a listener might view as culturally
inappropriate. One that slows down your speech and turns up the volume for an
older person with hearing loss. One that fixes your stutter in real time. One that
enables you to speak exclusively in rhyme. Or in the voice of your favorite
actress. Or in a perky voice when you’re feeling sick. All of these language
preferences are possible in theory, but plain old translation from one spoken
language to another is perhaps the most difficult of these to implement.
As long as human beings speak different languages, the need for translation
will continue. And as long as translation exists—even if it someday becomes
more fully automated—it will always require the skills, talent, and expertise of
human specialists. After all, even in science fiction, someone has to build and
maintain those universal translators. Indeed, Spock’s famed lack of emotion
was apparently due to a translation error. In the novel Spock’s World by Diane
Duane, Spock’s mother, who helped build the universal translator, was blamed
for a mistranslation that would cause Vulcans to be unfairly stereotyped. The
Vulcan word Arie’mnu, which translates to “passion’s mastery,” was
mistranslated by Spock’s mom as “lack of emotions.”