Hands-On Artificial Intelligence for Beginners

Neural machine translation

Machine translation represents a sequence-to-sequence problem; you'll frequently see these networks described as sequence-to-sequence (or Seq2Seq) models. Instead of utilizing traditional techniques that involve feature engineering and n-gram counts, neural machine translation maps the overall meaning of a sentence to a singular vector, and we then generate a translation based on that singular meaning vector. 

Machine translation models rely on an important concept in artificial intelligence known as the encoder/decoder paradigm. In a nutshell: 

• Encoders parse over input and output a condensed, vector representation of the input. We typically use a GRU or LSTM for this task
• Decoders take the condensed representation, ...