This is a quote from the DOM level 2 spec, explaining the normalize() function.
Puts all Text nodes in the full depth of the sub-tree underneath this Node, including attribute nodes, into a "normal" form where only structure (e.g., elements, comments, processing instructions, CDATA sections, and entity references) separates Text nodes, i.e., there are neither adjacent Text nodes nor empty Text nodes. This can be used to ensure that the DOM view of a document is the same as if it were saved and re-loaded, and is useful when operations (such as XPointer [XPointer] lookups) that depend on a particular document tree structure are to be used.
i.e. the idea of normalize is to make text nodes the same as they would be if the xml document was re-loaded into DOM. It's handy if you want to compare DOM nodes.