This $300,000 National Science Foundation project grant supports research at the University of Southern California to develop innovative domain-specific architectures for enabling simultaneous translation at the edge. The goal is to design hardware accelerators for transformer-based simultaneous translation models that can reduce computation requirements by orders of magnitude while maintaining translation accuracy. Specific lines of research include input-centric dataflow and data reuse, compute-proportional architecture to achieve linearly scalable attention calculation, and utility-driven linear transformation architecture to efficiently reduce dimensionality. Broader impacts include incorporating findings into graduate curricula and courses, as well as outreach activities to broaden inclusion in computational linguistics. This award falls under the NSF's Computer and Information Science and Engineering program to support investigator-initiated research and education in all areas of computing, communications, and information science.