Grammar-based Corpus Annotation

There is an increasing number of linguists interested in large syntactically annotated corpora (treebanks). Such corpora can serve as a base for statistical applications and, at the same time, may be used in theoretical linguistics as a source for investigations about language use. The most important treebank nowadays is the Penn Treebank (Marcus et al., 1993; Marcus et al., 1994). Many statistical taggers and parsers have been trained on this treebank, e.g. (Ramshaw and Marcus, 1995; Srinivas, 1997; Alshawi and Carter, 1994). Furthermore, context-free and uni cationbased grammars have been derived from the Penn Treebank (Charniak, 1996; van Genabith et al., 1999a; van Genabith et al., 1999c; van Genabith et al., 1999b). These parsers, trained or created by means of the treebank, very successfully parse unseen text with respect to correct POS tagging and chunking, and hence can be applied for enlarging the treebank. However, the situation is di erent for languages other than English. Ongoing projects are still in the process of building treebanks, e.g. for German (NEGRA corpus (Skut et al., 1997), now continued in the TIGER project; the German treebank in Verbmobil (Stegmann et al., 1998)), for Czech (The Prague Dependency Treebank (Haji£,