Treating German with a Provably Context-Free Grammar
Coping with Subcategorization, Unbounded Dependencies and Partially Free Word Order
Original title (German): Mechanismen einer kontextfreien Grammatik für das Deutsche
Several applications of natural language processing need short processing times while retaining a certain degree of linguistic adequacy. This article describes some techniques employed in a text-to-speech synthesis system for German which allow for the parsing of sentences containing linguistic phenomena such as unbounded dependencies, subcategorization, and free word order in the German Mittelfeld. The grammar is restricted to the unification of atomic terms only, and it is shown that every grammar using such a formalism is equivalent to a context-free grammar.
A condensed English version of the German original, Mechanismen einer kontextfreien Grammatik für das Deutsche, which first appeared in the student track of the 18th German Conference on Artificial Intelligence (KI-94). This version was presented at the 5th Conference of Computational Linguistics Students (TaCoS) in Saarbrücken.
The work describes the grammar and parser of the SVOX text-to-speech system, developed by the Speech Processing Group at ETH Zürich, where I spent three months in 1993 reworking both.