keyword
recurrent continuous translation models
Recurrent continuous translation models are neural machine translation frameworks that translate sentences between languages by operating entirely on continuous vector representations rather than discrete phrase tables or explicit word alignments [cite: 0]. In these architectures, a source sentence is first mapped into a dense vector space, typically using a convolutional sentence encoder to capture semantic and syntactic structure across varying lengths [cite: 0]. A target recurrent neural network language model then generates the output translation word by word while conditioned on this continuous source representation [cite: 0]. This design enables end-to-end probabilistic modeling of translation that remains sensitive to global sentence context, grammar, and word order [cite: 0].
1 item
