17.1 What are we Doing?

We present a parser for German which deals with its free word order. The word order of German is not really free but at least freer than in English. Our approach to parsing should in principle also apply to other languages with free word order such as Dutch and ???.

Our parser is based on the framework of dependency grammar. This framework is attractive for specifying the word order in languages with ``free'' word order. The reason is that dependency grammar leaves word order unspecified a priori and supports constraints for its specification a posteriori. We argue that word order constraints provide natural specifications of word order in German (or other free word order languages). In particular, we believe that word order specifications by constraints are more natural than by transformations.

There is very little agreement about what might constitute a dependency grammar or the essence of dependency parsing. It might also be unclear whether dependency grammar/parsing differs much from other kinds of grammar/parsing? We will mainly answer the question what dependency parsing is about and speak less about the grammar itself. For parsing, it turns out that we can reuse many ideas and programming techniques which we have already seen for other parsers. Nevertheless, we will also introduce a bunch of new ideas, in particular concerning constraints for word order and constraints for constituent structure.

We will show how to express word order constraints for German by finite set constraints in Oz. This leads to an efficient implementation of a dependency parser within the paradigm of constraint programming. The parser we present has been developed by Denys Duchier [1998].


Denys Duchier, Claire Gardent and Joachim Niehren
Version 1.3.99 (20050412)