Learning Phrase Representations using RNN Encoder–decoder for statistical machine translation

1. The main task accomplished was the ability to translate English into French, using a encoder-decoder model, in which the sequence was transformed into a vector in the encoder RNN model. In decoder, a vector is transformed into an output sequence, and encoder-decoder can be used to add sequential information between words and words.

2. Another task is to express the sequence as a vector, using vectors to clearly see that semantically similar words gather together.

3. When designing the hidden layer of the RNN, the reset and update gates are added to the read-in or generate sequence, which gives a more meaningful result with the option to discard the memory information and update the memory information.

Learning Phrase Representations using RNN Encoder–decoder for statistical machine translation