Leveraging Closed-Access Multilingual Embedding for Automatic Sentence Alignment in Low Resource Languages