How do sequence-to-sequence models generate outputs?

• A. They use an encoder to process input and a decoder to produce the output sequence.
• B. They classify input into predefined categories.
• C. They rely solely on the decoder without encoding.
• D. They convert text into audio signals.

Answer: (A)

Sequence-to-sequence models generate outputs by first encoding the input sequence into a representation that captures its information, then decoding that representation into the output sequence, token by token. The encoder processes the input and builds hidden states that summarize the input’s context. The decoder uses those states to predict the next token, conditioned on what it has already produced, and often with attention that lets it focus on the most relevant parts of the input for each step. During training, the model learns to maximize the probability of the correct output sequence given the input, enabling it to handle variable-length inputs and outputs. This encoder–decoder setup is essential for tasks like translation or summarization, whereas simply classifying inputs or skipping the encoder, or converting text to audio, would not produce sequential outputs in the same way.