Week 2 quiz - word embeddings

One of the quiz questions was about whether or not word embeddings are learned by predicting surrounding words given a word. My understanding was that we could think of NLP tasks as follows:

  1. Learn the word embeddings from a word corpus. This steps learns a feature space into which each word is then embedded, instead of being considered independent.

  2. Use the word embeddings to predict things like which words come next in a sentence.

If this is the right intuition, I think the quiz question is misleading or incorrect. If not, and someone could explain why, I’d appreciate it! Thanks.

Learning and predicting are separate tasks. You don’t do any predicting when training from the word corpus.

Perhaps I do not fully understand your question.
I agree with your summary of how word embeddings are learned and used.

Note that “predict what comes next” is what the LSTM provides. The embeddings are data that the LSTM uses.