The problem outlined in the present paper presented itself in the planning
stage of a project aiming at constructing the IPI PAN corpus of written
Polish. The IPI PAN corpus is going to contain at least 75-100 million
words. It is also going to be annotated structurally and
morphosyntactically according to the suggestions laid out in the Corpus
Encoding Standard Guidelines (Ide et al. 1996). The corpus will contain
several sub-corpora divided according to the genre of the texts that make
them up (e.g., literary texts, dialogue transcripts, etc.), as well as a
balanced reference subcorpus that should be representative of modern
standard Polish, and a hand-verified subcorpus designed for the purpose of
training the morphosyntactic tagger. In what follows, we briefly report on
a class of design problems related to the use of the so-called stand-off
morphosyntactic and structural annotation, advocated by the Corpus
Encoding Standard.