Predict multiple spans/answers from same input question with a separate token

[mayankagrawal93]

I have a dataset which contains multiple indices (or spans) as output. Example : If tweet is "Getting bored of walking up and down the stairs. Dont want to do it again!" and sentiment is negative, the answer given is : "getting bored of walking ; dont want to do it again".
My input becomes : "question: negative context: Getting bored of walking up and down the stairs. Dont want to do it again!"
And expected output is "getting bored of walking; dont want to do it again "
Here there are two spans extracted from same tweet which correctly points to negative sentiment and both of them are needed as output (not any one). I want to finetune the model in this way that it is able to extract both the spans separated by a token.
But I could not find any [SEP] token or any such similar token used by T5 in output.
Can I use EOS token after each phrase like "getting bored of walking </s> dont want to do it again </s>" ?