Building a Morphological Treebank for German from a Linguistic Database

Building a Morphological Treebank for German from a Linguistic Database Petra Steiner, Josef Ruppenhofer Institut f¨ur Deutsche Sprache {ruppenhofer, steiner}@ids-mannheim.de Abstract German is a language with complex morphological processes. Its long and often ambiguous word forms present a bottleneck problem in natural language processing. As a step towards morphological analyses of high quality, this paper introduces a morphological treebank for German. It is derived from the linguistic database CELEX which is a standard resource for German morphology. We build on its refurbished, modernized and partially revised version. The derivation of the morphological trees is not trivial, especially for such cases of conversions which are morpho-semantically opaque and merely of diachronic interest. We develop solutions and present exemplary analyses. The resulting database comprises about 40,000 morphological trees of a German base vocabulary whose format and grade of detail can be chosen according to the requirements of the applications. The Perl scripts for the generation of the treebank are publicly available on github. In our discussion, we show some future directions for morphological treebanks. In particular, we aim at the combination with other reliable lexical resources such as GermaNet. Keywords: treebank, morphology, word structure, deep-level morphological analyses, CELEX, German 1. Introduction German is a language with complex processes of word formation, of which the most common are compounding, derivation and conversion. The resulting lexical units usu- ally have long orthographical forms. Moreover, many word forms have more than one combinatorially possible analy- sis, as in Figure 1: Hauptbahnhof ‘central station’ consists of three morphs which can be combined in two ways on the level of immediate constituents but only the first combina- tion is the correct structure. Hauptbahnhof Haupt ‘main/central’ Bahnhof ‘station’ ♯Hauptbahnhof Hauptbahn ‘*main rail’ Hof ‘yard’ Figure 1: Ambiguous analysis of Hauptbahnhof ‘central station’ Other word forms have ambiguous boundaries of morphs as in Figure 2 where the word form Zugriff ‘grasp/access’ is the product of a conversion process from zugreifen ‘to grab/to grasp’ and not a compound of other forms which could be erroneously recognized by a morphological anal- ysis program. Zugriff zugreifen ‘to grab/grasp’ zu ‘at’ greifen ‘to grip’ ♯Zugriff Zug ‘train’ Riff ‘riff’ Figure 2: Ambiguous analysis of Zugriff ‘grasp/access’ Morphological splitters for German such as Gertwol (Haa- palainen and Majorin, 1995), MORPH (Hanrieder, 1996), SMOR (Schmid et al., 2004), or TAGH (Geyken and Han- neforth, 2006) generate many ambiguous analyses. Usu- ally, this problem is approached by filtering procedures on the output analyses. For the ranking of the different morphological analyses, the geometric mean is a common score (Cap, 2014; Koehn and Knight, 2003; Steiner and Ruppenhofer, 2015). Another method is pattern matching with tokens (Henrich and Hinrichs, 2011) or comparisons of lemmas (Weller-Di Marco, 2017) or strings (Daiber et al., 2015) with corpus data. Ziering and van der Plas (2016) use normalization combined with ranking by the geometric mean. W¨urzner and Hanneforth (2013) apply a probabilis- tic context free grammar for parsing adjectives. Ma et al. (2016) apply Conditional Random Fields modeling for let- ter …

Content truncated to 500 words