Training Trains Internship
Student login
← Natural Language Processing learning path
Natural Language Processing · Module 6 of 10

Sequence-to-sequence tasks

Translate, summarise and answer questions.

📖 Key ideas

  1. Seq2seq models map input text to output text.
  2. Encoder-decoder Transformers (T5, BART) suit translation and summarisation.
  3. Machine translation powers Google Translate.
  4. Extractive summaries copy sentences; abstractive summaries rewrite.
  5. Metrics: BLEU (translation), ROUGE (summarisation).
💡 Real-life example: Seq2seq is an interpreter listening to a full sentence, then speaking it in another language.
Seq2seqSequence-to-sequence model
Encoder-decoderReads input, then generates output
BLEUTranslation quality metric
Abstractive summaryRewritten summary