Shallow Post Morphological Processing with KURD

B m | m m | m m m m m Shallow Post Morphological Processing with KURD Michael Carl and Antic Schmidt-Wigger email: cad,antje@iai.uni-sb.de Institut fiir Angewandte Informstionsforschung, Martin-Luther-StraJ~e 14, 66111 Saarbrticken Germany Abstract In this paper we describe a constraint based formalism that manipulates se- quences of morphological analyses in or- der to Kill, Unify, Replace or Delete parts of the structure. We compare the formalism to a similar approach (CGP) and describe two applications. 1 Introduction In NLP applications an input text undergoes a number of transformations until the desired in- formation can be extracted from it. Typically, such transformations involve part of speech tag- ging, morphological analyses such as lemmatiza- tion or full derivational and compositional analy- ses, context-dependent disambiguation of tagging results, multi-word recognition, shallow, partial or full syntactic parsing, semantic analyses and so OU. It is not always evident what level of analysis should be involved. For instance, whether a cer- tain task reqnizes a full parse or whether some 'shallow' operations may be sufficient is often dif- ficult to determine. The choice of tools can be guided by the data or the requirements and the prerequisites of the goal to be reached. These considerations may depend on the availability of a grammatical model, the required standard of the results, and processing time constraints. However, the optimization of this task remains an unre- solved area until now. The interest of the NLP community for 'shal- low' processing has grown recently (cf. (Karls- son, 1990),(Abney, 1996), (Deelerek and Klein, 1997)). In this paper, we describe a simple formal- ism (KURD x) that is designed to perform some X KURD is an acronym representing the ftrst letters of the implemented actions: K(ill)-U(nify)-R(¢place)- D(elete) Carl and Schmidt-Wigger 257 'shallow' operations on morphologically analyzed texts. The output can be used directly, or be redi- rected to further processing. Typical tasks for such shallow processing include • Tagging (disarnbiguation of multiple mor- phological analyses) Often a set of simple rules that runs in a set order over the results of the morphological analyses is sufficient to disambiguate multiple analysis of a word due to its morphosyntactic context. • Syntax checking Grammatically erroneous sentences are de- tected by a set of rules describing common weak points such as missing punctuation marks or ill-formed agreement. • Style checking Highly complex constructions or heavy phrases can disturb the reading and under- standing process. To avoid this, style check- ers can recognize such patterns so that the author can readjust his text for better com- munication. • Shallow parsing Shallow parsing can help to simplify the data before full parsing is undertaken. It recog- nizes syntactic phrases, mostly on the nomi- nal level. • Segmentation The morphological analysis deals with words which are presented in texts. High level processing deals with units between the word level and the text level, mostly with sen- tences. Thus, sentence segmentation is a typ- ical shallow process, but other subunits could be equally interesting. KURD Michael Carl and …

Content truncated to 500 words