What is a SentenceSet?

[tastyminerals]

I do not understand the following SentenceSet paragraph:

A DataSet used for language modeling. 
Takes a sequence of words stored as a tensor of word IDs and a Tensor holding the start index of the sentence of its commensurate word id (the one at the same index). 
Unlike DataSets, for memory efficiency reasons, this class does not store its data in Views. 
However, the outputs of factory methods batch, sub, and index are Batches containing input and target ClassViews.
The returned batch:inputs() are filled according to Google 1-Billion Words guidelines.

So words stored as two tensors, a "tensor of word IDs" and the other as "tensor holding the start index of the sentence of its commensurate word id" what?
I am not even asking why words are stored as two tensors in the first place, because this is even more confusing. Can somebody explain how are words actually stored?