Negative words are sampled from the vocabulary. The real algorithm uses a smoothed unigram distribution.