A Divide-and-Conquer Strategy for Shallow Parsing of German Free Texts

A Divide-and-Conquer Strategy for Shallow Parsing of German Free Texts Giinter Neumann* Christian Braun t Jakub Piskorski ~ Abstract We present a divide-and-conquer strategy based on finite state technology for shallow parsing of real- world German texts. In a first phase only the topo- logical structure of a sentence (i.e., verb groups, subclauses) are determined. In a second phase the phrasal grammars are applied to the contents of the different fields of the main and sub-clauses. Shallow parsing is supported by suitably configured prepro- cessing, including: morphological and on-line com- pound analysis, efficient POS-filtering, and named entity recognition. The whole approach proved to be very useful for processing of free word order lan- guages like German. Especially for the divide-and- conquer parsing strategy we obtained an f-measure of 87.14% on unseen data. 1 Introduction Current information extraction (IE) systems are quite successful in efficient processing of large free text collections due to the fact that they can provide a partial understanding of specific types of text with a certain degree of partial accuracy using fast and ro- bust language processing strategies (basically finite state technology). They have been "made sensitive" to certain key pieces of information and thereby pro- vide an easy means to skip text without deep anal- ysis. The majority of existing IE systems are ap- plied to English text, but there are now a number of systems which process other languages as well (e.g., German (Neumann et al., 1997), Italian (Ciravegna et al., 1999) or Japanese (Sekine and Nobata, 1998)). The majority of current systems perform a partial parsing approach using only very few general syntac- tic knowledge for the identification of nominal and prepositional phrases and verb groups. The combi- nation of such units is then performed by means of domain-specific templates. Usually, these templates * DFKI GmbH, Stuhlsatzenhausweg 3, 66123 Saarbriicken, Germany, neumann@dfki, de t DFKI GmbH, Stuhlsatzenhausweg 3, 66123 Saarbriicken, Germany, cbratm@dfki, de DFKI GmbH, Stuhlsatzenhausweg 3, 66123 Saarbriicken, Germany, piskorsk@dfki, de are triggered by domain-specific predicates attached only to a relevant subset of verbs which express domain-specific selectional restrictions for possible argument fillers. In most of the well-known shallow text process- ing systems (cf. (Sundheim, 1995) and (SAIC, 1998)) cascaded chunk parsers are used which per- form clause recognition after fragment recognition following a bottom-up style as described in (Abne), 1996). We have also developed a similar bottom- up strategy for the processing of German texts, cf. (Neumann et al., 1997). However, the main prob- lem we experienced using the bottom-up strategy was insufficient robustness: because the parser de- pends on the lower phrasal recognizers, its perfor- mance is heavily influenced by their respective per- formance. As a consequence, the parser frequently wasn't able to process structurally simple sentences, because they contained, for example, highly complex nominal phrases, as in the following example: "[Die vom Bundesgerichtshof und den Wettbewerbshfitern als Verstofi gegen das Kartellverbot gegeiflelte zentrale TV- Vermarktung] ist g~ngige Praxis." Central television raarketing, censured by the German Federal High Court and the …

Content truncated to 500 words