#sequence-to-sequence

Question 1: Single choice

mediumAI-checkedNLP Tasks

Part-of-speech tagging is framed as sequence labeling (one output label per input token), but machine translation is framed as sequence-to-sequence generation. Why can't translation be done by labeling each source word with its translation?

Input and output can differ in length and word order, with no one-to-one alignment1
A sequence labeler cannot look at neighbouring words, so it cannot use context2
Translation needs a softmax over the whole target vocabulary, which a tagger cannot have3
Sequence labeling is only defined for a single language, not for pairs of languages4

#sequence-to-sequence

Timed test