Collocations of Complex Nouns: Evidence for Lexicalisation

This paper combines a corpus-based study of noun+verb collocations with an attempt to distinguish compositional, regularly formed compounds from lexicalised ones. We claim that morphologically regular, compositional compounds share most of their collocational preferences with their compound heads, whereas lexicalised compounds have their own collocational preferences, distinct or only marginally overlapping with those of their heads. We test this claim on corpus data for noun+verb collocations of 85 German simple nouns and their compounds (1,058 types). The extraction relies on stochastic parsing, since this procedure provides syntactically homogeneous data and frequency counts, which are input to the log-likelihood association measure, to compute significance scores for the combinations. The experiments seem to confirm our hypothesis: a collocational analysis of this kind can serve to identify lexicalised compounds. Being able to identify lexicalised compounds is relevant for lexicography in different respects: in the definition of the nomenclature of a dictionary, because lexicalised compounds should be included with priority, and in work on the micro structure, because collocations of a compound which are shared with the compound head may be treated extensively under the head, with only a reference or a pointer under the compound.