Building context: which words should count?

To predict the next word after "the big cat sat", the model needs a single context vector. How should it combine the four word vectors? Compare three strategies — and in weighted mode, drag the sliders to emphasize the words that matter.