Dataset Preparation for Q&A FineTuning

Hi all, this is my first post! Happy to join the community here.

On Q&A Pretrained models, i was studying the Datasets especially used for popular models, and the pre-processing of the datasets seems highly important.

take for example

In the answers column, how does one direct the answer: to the exact token? are there some steps i am missing, was looking through all the relevant tutorials but i reckon this maybe a beginner problem. Any help is appreciated!