Emerging applications for automated translation
I asked Dr Seligman about the potential application of Spoken Translation’s technology into digital engagement channels such as social media. He explained that this is where he started out in the mid-1990’s, working on automating chat translations and it is certainly something that he hopes Spoken Translation will return to in the future. The company has a vision for applying their technology of today to live, verifiable, chat translation:
Spoken Translation’s vision of the future: live, verified chat translation
Other innovation in the pipeline includes server-based technology that would allow Converser to be used from portable and mobile devices. Dr Seligman hopes this will be available for iPhone and Blackberry during 2011.
Purpose built automated healthcare translation
During my research for this article, I was pleased to learn from TAUS, a think tank on translation strategies, about customized machine translation systems (aka automated translation) which are designed for use in specific sectors. This is in contrast to Google translate which can be unreliable for specialist areas such as healthcare. These customized engines are trained using database of previous translations from a specific industry and include features to ensure consistent use of terminology. The result is better quality, more accurate translations.
Just such a sector specific solution is used by the Pan American Health Organization (PAHO). Their system was originally developed in the 1970s, and today covers all combinations of English/Spanish and Portuguese and is being used daily for 90% of all PAHO’s translation needs, as well as by 75 clients. Another example is a Danish customized machine translation provider, Languagelens, whose purpose built solutions are used during clinical trials by pharmaceutical companies. Whilst human translators are needed to ensure that the final text is up to the high quality needed, the use of such customized automated engines drives down the cost of translation, increasing the amount of translation that can be done, and speeds up time to market.
TAUS also told me that it is possible to create automated engines rapidly when needed – within three weeks of the recent crisis in Haiti both Google and Microsoft added Haitian to the list of languages supported by their automated engines.
Real-time translation crowdsourcing
As automated translation technologies are deployed into healthcare environments, other innovative approaches to solving the automation challenge are emerging. New York, US based SpeakLike has developed a process that is enabling social media engagement to take place across 37 languages. Sanford Cohen, SpeakLike’s CEO told me that they were looking for a solution to enable real-time chat amongst people speaking different languages.
“We explored machine translation and found it was not good enough for our needs,” explains Cohen. “So we thought, ‘if machine translation were perfect, it would be integrated into everything we use – it would be in our email systems, in our chat systems, and in our content management systems; but it’s not. But why can’t we have something that can be integrated into everything we use, with good quality translation?’ That’s when we started looking at crowdsourcing.” Cohen says this idea was how SpeakLike started:
“We got a large number of translators on our system, and users could send in a request when they needed it, 24/7, and then whatever translators were available or logged in first would provide the translation.”
The first application of the process was live chat, and in a 2008 beta SpeakLike demonstrated live, real-time chat between three users speaking English, Spanish and Chinese. This technology was implemented by PETLinQ, a provider of radiology imaging software management tools, to enable their user base of 71,000 doctors to collaborate in their own language.
After experimenting with the physician-patient interaction, where a dental reconstructive surgery in New York could support its worldwide patients pre- and post-surgery via international chat, SpeakLike started to develop other applications of the process. The translation platform was expanded to integrate with email, website content, and social media applications.
Today, a Twitter connector automates the translation of tweets, enabling either a single, multilingual Twitter feed or separate feeds for each language. Meanwhile, for bloggers using WordPress, a plugin automatically posts translated content into languages selected by a content author.
SpeakLike’s system is designed to manage the end to end process, automatically notifying translators, managing translated content, and publishing based on user options.
Cohen told me of a customer who was previously waiting typically for two weeks to have website updates translated into nine languages, but their content was out of date within four days. By integrating SpeakLike into their content management system, they were able to publish translations within less than 24 hours.
Responding to international health crises
The potential for transformation that can be achieved when people from different countries collaborate to solve healthcare challenges is exemplified in the work of international aid organisations such as Médecins Sans Frontières, where healthcare specialists from around the world work together in response to a crisis. But when the international team leaves an area of need, local physicians are often left without access to the international knowledge pool that exists during an aid mission.
Murdo Bijl, Founder and Executive Director of Health Connections International, saw this situation first hand when working with Médecins Sans Frontières in the former Soviet Union. The experience inspired him to set up an organisation focused on facilitating and promoting communication between professionals through multi-lingual exchanges of information. Health Connections International operates on a non-profit basis and focuses on improving responses to the HIV, tuberculosis and drug use epidemics in developing countries and resource-poor environments.
The organisation’s online knowledge and information sharing platform has been designed to allow healthcare professionals across the globe to share their experiences and exchange information, quickly and easily across multiple languages.
Healthcare professionals register as members of My Health Connections and can ask medical questions in their own language. Most questions are then manually translated and labelled by subject area (such as HIV/AIDS, treatment, medication) before being routed to an appropriate expert to be answered. Once answered, the response is translated back into the language in which the question was originally asked. It’s a laborious process but the result is a rich and growing knowledge base accessible in multiple languages.
MyHCI includes a growing, multilingual expert medical information knowledge base
I spoke with Murdo Bijl about his vision. He told me that in the proof of concept that has been running since April this year, 600 unique questions and answers have been posted. He said that as the number of questions and answers continues to grow, the knowledge base will be able to provide the answers to most commonly asked questions:
“There will be a moment when the knowledge base will have enough information for people to find the answers to their questions. Then all the questions will be translated into Russian, Spanish, Arabic and Chinese. Right now we have 500 Q&As online, translated into Russian.”
