Identifying predicatively used adverbs by means of a statistical grammar model

This paper presents work on the corpus-based acquisition of predicatively used German adverbs by means of a statistical grammar model. The predicative use of adverbs is lexically restricted to local adverbs, temporal adverbs, and to a limited number of other adverbs, some of which are part of idiomatic phrases. Besides listing local and temporal adverbs we are mainly interested in identifying the list of ‘other’ adverbs. These adverbs cannot be subsumed under a common semantic class. Their predicative function is an idiosyncratic property of the lemma. The identification task was pursued by extracting predicatively used adverbs and prepositional phrases with an adverb complement from a statistically trained grammar model. The results were sorted by frequency and ranked according to a statistical association score. We then manually classified them into four classes: (i) local adverbs, (ii) temporal adverbs, (iii) other adverbs, and (iv) parsing errors. We discuss the extraction results from a linguistic point of view .