How does torchtext define the example that makes up the dataset?

How does torchtext define the example that makes up the dataset?