The Semantic Web relies on two layers: XML and RDF. XML documents are merely trees representing structured data or documents and accessed using XPath. RDF is used for a Web of data and to provide metadata about data or documents; it is organized as graphs made of collections of elementary triples and can be queried using SPARQL. Based on these two paradigms, there exist tools and platforms that produce and process both XML and RDF. When doing information integration and mash-up applications, there are scenarios where we need to query, compare and integrate data coming from both worlds. In this report we present a seamless way of mixing both paradigms in SPARQL. Generic extensions to SPARQL are explained, and then we provide use cases and an application in semantic annotation of textual documents using NLP techniques.