A word2vec-style model is trained on the corpus below and on nothing else. It is shown one centre word at a time and must predict the words surrounding it, using a window of two words on each side. Training runs until the loss stops falling.
| # | sentence |
|---|---|
| 1 | the soup was hot |
| 2 | the soup was cold |
| 3 | she drank hot tea |
| 4 | she drank cold tea |
| 5 | do not touch the hot metal |
| 6 | do not touch the cold metal |
The two highlighted words never appear in the same sentence, and they mean opposite things.
Where do the vectors for hot and cold end up relative to each other?
Select all that apply.