An ontology of systematic relations for a shared grammar of Slavic

An ontology of systematic relations for a shared grammar of Slavic Tania AVGUSTINOVA Computational Linguistics, Saarland University P. O. Box 151150 Saarbrücken, Germany, D-66041 avgustinova@coli.uni-sb.de Hans USZKOREIT Language Technology Lab, DFKI Computational Linguistics, Saarland University Saarbrücken, Germany, D-66041 uszkoreit@dfki.de Abstract Sharing portions of grammars across languages greatly re- duces the costs of multilingual grammar engineering. Related languages share a much wider range of linguistic information than typically assumed in standard multilingual grammar architectures. Taking grammatical relatedness seriously, we are particularly interested in designing linguistically motivated grammatical resources for Slavic languages to be used in applied and theoretical computational linguistics. In order to gain the perspective of a language-family oriented grammar design, we consider an array of systematic relations that can hold between syntactical units. While the categorisation of primitive linguistic entities tends to be language-specific or even construction-specific, the relations holding between them allow various degrees of abstraction. On the basis of Slavic data, we show how a domain ontology conceptualising morpho- syntactic "building blocks" can serve as a basis of a shared grammar of Slavic. Introduction In applied computational linguistics, the need for developing and utilising operational notions of shared grammars stems from multilingual grammar engineering. If considerable portions of existing grammars can be reused for the specification of new grammars, development efforts can be greatly re- duced. A shared grammar also facilitates the diffi- cult task of maintaining consistency within and across the individual parallel grammars. In machine translation, the specification of a shared grammar can furthermore be exploited for simplifying the transfer process. Without much ado, computational linguists engaged in multilingual grammar development have always tried to reduce their labour by importing existing grammar components in a simple "copy-paste- modify" fashion. But there were also a number of systematic attempts to create and describe shared grammars that are convincingly documented in publications. [Kam88] demonstrates the concept for a relatively restricted domain, the grammatical description of simple nominal expressions in five languages. [BOP88] were able to exploit the gram- matical overlap of two Slavic languages, for the design of a lean transfer process in Russian to Czech machine translation. In multilingual applica- tion development within Microsoft research, gram- mar sharing has extensively been exploited � [Pin96], [GLPR97]. However, all these approaches are rather opportun- istic in the sense that existing grammatical descrip- tions based on existing grammar models were ex- plored. We went a step further and started grammar design with a notion of a shared grammar for a family of related languages. Pursuing the goal of designing linguistically motivated grammatical resources for Slavic languages to be used in com- putational linguistics, one is inevitably confronted with primary problems stemming from the fact that different linguistic theories cut up grammars in quite different ways, and grammar formalisms differ in their degree of granularity. It cannot be expected, therefore, that the minimal differences between two languages or their shared elements form easily identifiable units in the available language-specific grammars. Therefore, an ontology conceptualising morphosyntactic "building blocks" would offer a …

Content truncated to 500 words