By Matt Hauser

ComputerWorld.com – May 26, 2011

Back in February of this year, the folks at Jeopardy set up the first ever man-versus-machine version of their game show.  And much like many over-hyped match-ups in sports, it wasn't even close; Watson wiped the floor with its human challengers, more than tripling the cash winnings of the second place finisher.  Watson also picked up a cool $1 million for the total prize.

So, given the fact that a supercomputer can win convincingly at a game show that is structured around more conceptual thinking, what does this mean for machine translation (MT) as an alternative to a human-based process?  In the short term, not much. 

But it certainly outlines future scenarios in which MT becomes more and more viable and higher quality as computers are able to mimic human brain functions related to concepts and subjectivity.  It also potentially gives rise to a "Skynet" or "Matrix" type of future where humans are ultimately enslaved by machines or turned into batteries, but let's look at the bright side while there is still time.

In the marketplace today, machine translation is becoming more and more accepted as a means of communication.  This can be attributed to two main factors - the sheer amount of information that is being created online each day, and the increasing use of statistical MT to improve overall results.

On the first point, businesses are being flooded with inbound content in other languages and also being pressured to provide mutilingual outbound content.  Inbound content includes things like support requests, customer service, product reviews and interaction with social media like Facebook or Twitter.  Outbound content includes responses to all of the aforementioned components, as well as more obvious items like websites, product documentation, marketing materials, HR materials, etc.  With this constant two-way flow of communication, organizations must make decisions as to which content warrants higher-quality human-based translation, which content can be handled effectively via MT, and which content may benefit from a hybrid approach of MT and human translation.

So when does MT make sense?  This goes back to the risk assessment and content structure point listed above.  High risk content can be defined in a few different ways.  It can be literally "high-risk" as in a mistranslation can be life or death (think instructions for a portable de-fibulator or drug interaction information) or it can have significant financial or legal impact (annual reports, contracts, etc.).  It can also have a risk factor associated with brand affinity (translation of a tag line, slogan or marketing campaign).  In these cases, human translation is still the preferable option.  And human translation is still more able to handle marketing content that may not translate literally.

For more structured communications, like manuals and documentation, MT can be a potential option to reduce costs and accelerate turnaround times, but a post-MT edit by a human translator is recommended to ensure overall quality.  This is a fairly common practice with technical documentation in industries like IT, automotive and aerospace.  And it makes good sense when the projects are large, frequent and consistent versus ad hoc (since you can train the MT engine over time to produce higher quality results).

While MT has advanced considerably in a short amount of time (and continues to do so), it still is most effectively used as either a way to communicate the gist of general information (e.g., sort through support requests to see which require human interaction) or as a complementary piece to process where human editing is involved.  And for higher risk, higher profile content, human translation is still the recommended approach.

My firm, as well as virtually every other firm in the language services space, continues to invest in and research MT technologies in efforts to reduce overall costs without sacrificing quality, so I would anticipate that MT will become more and more involved in the translation and localization process as the technology continues to evolve.

Let's just hope we have a few more years ahead of us before we are all working for Watson.

Matt Hauser is VP of Technology for TransPerfect Translations.