Sources in BigML store the information needed to parse correctly the data that you uploaded. Parsing is paramount to ensure that algorithms interpret correctly the numbers (decimal separator) and the texts (language, stop words, etc.) that are going to be used in the training and prediction phases. Also, for source composites, the source configuration decides the kind of features that will be extracted from your image files.
To learn about all their available configuration options using BigML's Dashboard, you can check the corresponding section of the Sources documentation. Also the Sources API documentation contains the information about the attributes that need to be specified to configure your source programmatically.