Tell me Where it Hurts: Army R&D Improving Translation Technologies for OCONUS MedicsPR Web
Aberdeen Proving Ground, MD (PRWEB) October 25, 2012
When providing medical care, it is critical that medics are able to converse with patients about their medical needs. Language barriers make communication a challenge and can hinder the delivery of effective medical care. Without the assistance of a human translator, a medic may not be able to accurately capture enough information to fully address a patients needs.
Engineers from the U.S. Army Research, Development and Engineering Commands Communications-Electronics Research Development and Engineering Center, or CERDEC, have spent the past year developing strategies and methods for improving machine foreign language translation software in support of military medical translation needs.
The Medical Application of Speech Translation, or MAST, is a collaborative research project between CERDEC, the U.S. Army Medical Research and Materiel Commands, Telemedicine and Advanced Technology Research Center, or TATRC, and the U.S. Army, Southern Command, or SOUTHCOM. TATRC is the sponsor and program manager for MAST, CERDEC provides the engineering support, and SOUTHCOM facilitates access to the operational environment and end-user base.
TATRC has identified a need to conduct research on capabilities that will enhance communication for Department of Defense health engagements outside the United States, said Ray Schulze, Chief, Information Management Branch, for CERDECs Command, Power and Integration Directorate, or CP&I.
Medics are seeking a small portable solution, basically, translation software that runs on a mobile device that can be used without an internet connection said, Cynthia Barrigan, MAST Program Manager and portfolio manager for global health engagement at TATRC. In addition, we know that while users are interested in using a translation technology, they are concerned about how it will integrate into their clinical routine in the field and how a patient will react to it. They are also aware that they will need the translation to be very accurate to be useful; giving them a vested interest in seeing real improvements to the current capability.
CERDEC has worked in the field of language translation for a number of years, supporting Product Director, Machine Foreign Language Translation Systems, or PD MFLTS, since its inception.
We specialize in language translation in the disconnected environment and doing so on various mobile platforms, said Schulze. After hearing reports about translation challenges in Haiti, following the 2010 earthquake, TATRC recognized that translation was a pressing need for the medical community and that we could assist with the engineering research and development to help accelerate a medical capability.
The combination of automatic speech recognition, which takes spoken word and converts it into text, and foreign language translation [technologies] already exists, but the accuracy of those technologies is mediocre, at best, when used within the medical domain, said Schulze.
In some locations, Soldiers currently have foreign language translation technologies like the Phraselator, initially developed by the Defense Advanced Research Projects Agency, or DARPA, in partnership with CERDEC back in 2001. Those systems were developed for expedience in particular Soldier scenarios using, initially, one-way translation, and utilize generic phrases that require the user to stick to a script. Commercial translation applications also exist, but they are made for tourists in foreign countries. Medical providers tend to use complicated terminology that can easily be misinterpreted by a machine, said Schulze.
If youre really talking Western medical terms [to a commercial translation app], take mesothelioma for example, it misunderstands that word as Miss Ophelia, he said.
The MAST project aims to conduct research and development that can support medical care in the field.
Armed with a variety of mobile devices loaded with a commercial translation software application called Jibbigo, the MAST team gathered useful operational data and observations, and demonstrated the technology to users.
To date, collected data has all been scenario-based, meaning engineers created scripts based on their observations of doctor and patient interactions and recorded the scripted conversations between a native Honduran, volunteers and medical students, and SOUTHCOM medics.
The more data you give the program, the better it becomes. Thats the purpose for us going and doing all these recordings, said Daniel Yaeger, a CACI contractor supporting CERDEC CP&I.
Translation software today learns, or becomes more accurate, in much the same way a small child learns, explained Yaeger, a subject matter expert in the area of machine language translation who has traveled to Honduras three times in the past six months performing technology demonstrations and collecting simulated data for the MAST project.
A child learns by listening to conversations, they absorb it, said Yaeger. Its very similar to statistical machine translation, which is what were doing. Essentially, you tell the program that this sound means this text. You do that enough times and the algorithms behind the machine translation software actually learn those new phrases.
CERDEC CP&I engineers followed SOUTHCOM medics on several medical readiness training exercises, or MEDRETEs, to observe how medics interact with patients. This includes observing the process of registration, triage, observing the types of conversations between doctors and patients, common questions a doctor asks, and how human interpreters are being used, said Yaeger. By understanding how medics interact with patients, engineers can determine the specific requirements that would be needed to put a machine translation system in place.
Each trip to Honduras has used a different set of hardware, microphones, devices and form factors, in order to gain feedback from doctors.
Wireless was bigthey dont want the wire to get in the way or have to disconnect from anything, said Yaeger. So we brought wireless microphones and packaged everything in neat set up where you can just pick it up and take it where you need it. There are no wires, its ready to go.
The observations also revealed a significant challenge to MAST and other translation programs. Shy, quiet patients in a noisy environment make it incredibly difficult for speech recognition software to hear what a patient is saying, let alone translate it.
Its mostly women and young children that come to these events in Honduras. Its very difficult to get them to speak loud enough and then interact with an iPad that theyve probably never even seen before. A lot of them are hesitant to touch it, said Yaeger.
While simple hardware improvements to the translation device, like an improved microphone, have dramatically improved performance in noisy environments, those same improvements do little to make a patient feel more comfortable.
To combat this problem, a concurrent CERDEC machine translation project in Thailand has attempted to change the paradigm about what translation software can do for doctors. The common idea is to turn doctors into bilinguals by giving them translation applications. The twist in this program has experimented with taking an individual from the local population and turning them into the translator.
Instead of turning the doctor into an interpreter, we turned someone who was monolingual into an interpreter in that language, said Yaeger. We call them monolingual facilitators.
These facilitators, local volunteers at medical exercises, are trained by a bilingual translator to use the translation software. The doctor interacts with the facilitator in the same way he would with an interpreter, but the device is used to communicate between the two languages.
The facilitator is there as a cultural filter between the technology and the patient, said Yaeger. The patient doesnt have to interact with it [the translation device] at all. Theyre going to have a conversation with someone who lives in their country, speaks their language and knows their culture. That facilitator will talk back to the doctor using the technology.
Not only does this set-up make a patient more comfortable, but it also proves to be more efficient. The same two people, doctor and facilitator, interact and use the technology all day, rather than having to teach each new patient how to use the translation system.
Machine translation is not expected to replace a human interpreter, especially for emergency or complicated medical practices, said both Schulze and Yaeger. Machine translation is meant to augment the number of human interpreters that currently exist.
While the recordings from MAST over the past year were scripted, they were all spoken by native Spanish speakers, as opposed to doctors providing medical terms in Spanish, said Schulze. The Spanish language sounds different when spoken by a native speaker compared to a native English speaker. Those differences impact the accuracy of translation software, so it is critical to gather data from native speakers for any machine translation program.
These efforts are noteworthy advances in CERDECs expertise in language translation, Schulze said. Were not just collecting data. Were trying to perfect the process of collecting data for the purpose of improving translation accuracy. An example of this is the work were doing with TATRC within the medical domain.
As the process of collecting data is perfected, the technology can be transitioned to other languages and other niche areas outside the medical domain.
The key is improving the processthats how it will be transitioned to other domains, said Schulze. We talk to computers using a keyboard and mouse, everyday of our lives. It takes longer than simply speaking like we do to human beings, but its just the most accurate way to interact with machines. Accurate speech recognition and language translation could revolutionize the way Soldiers will interact with computers and frankly the entire world.
Read the full story at http://www.prweb.com/releases/2012/10/prweb10057153.htm