What is a reference corpus, and what can it do?

The 'Deutsche Referenzkorpus (DeReKo)' of the Mannheimer Institut für Deutsche Sprache currently contains over 28 billion words, and it is constantly being expanded. The sheer size of the corpus makes it impractical for researchers to analyze its entire content. On the other hand, the DeReKo offers the possibility of taking seriously the principle that every research project needs its own corpus - by acting as a 'reference corpus' that can be used in combination with special corpora. This paper addresses the question of whether a corpus should contain complete texts or only statistically relevant extracts; it also discusses the uses and necessity of 'small corpora'