The goal of this project is to demonstrate that speech translation technology can be used to remove language barriers between nurses and non-English-speaking patients in a hospital setting. In Phase I, the hypothesis is that speech translation technology that has been developed for the U.S. military can be adapted for use by nurses to communicate effectively with their non-English-speaking patients. For Phase I, Sehda will create and implement a 2-way speech translation system that will enable nurses at Kaiser Permanente Medical Center in San Francisco to communicate with Spanish speaking patients for five common patient-care activities, including assessing and managing pain. The goal is to use the system with at least 5 nurses and 25 patients for a minimum of 50 nurse-patient encounters and establish a baseline of system accuracy. ? Phase I will involve five tasks: (1) Sehda will define the scope of the interactions the system will enable by getting input from nurses through key informant interviews and focus groups. (2) The company will then collect data to model nurse-patient interactions by recording actual nurse-patient interactions in the hospital and by recruiting Spanish speakers to call Sehda's telephone-based data collection system. Sehda will transcribe the recordings and use the data to build language and translation models. (3) Sehda will then build the speech translation system on its S-MINDS platform. (4) Sehda will monitor the system's performance as nurses use it with their Spanish-speaking patients. During this stage, Sehda will record and transcribe all interactions for later analysis, and nurses and patients will be surveyed on their experiences with the system. (5) At this point, Sehda will isolate the conversation segments and code them for accuracy to objectively assess the system's performance. The company will also evaluate nurse and patient attitudes toward using the speech translation system. The goals for Phase II are to demonstrate that speech translation technology can be more accurate than ad hoc interpreters and as accurate as trained interpreters when used for specific types of interactions and that nurses and patients will adopt the technology as a routine part of patient care. In Phase II, Sehda will add Cantonese to the system and expand the range of nurse activities for which the system can translate. Sehda will conduct a similar set of activities as in Phase I but on a much larger scale-collect speech data, build and test system, monitor system use with patients, and evaluate results. In this phase, Sehda will also be comparing the system's performance with the performance of human interpreters, including professional in-person interpreters, telephone-based interpreters, and ad hoc interpreters such as friends and family of the patient and non-clinical bilingual hospital staff. ? ? ? ? ?
