SimpleNLG-NL : Natural Language Generation for Dutch

This thesis presents SimpleNLG-NL, a Dutch adaptation of SimpleNLG. SimpleNLG is a Java-based surface realiser, which performs the last step in Natural Language Generation. Natural Language Generation is the process of transforming non-linguistic information into understandable texts. With SimpleNLG-NL, developers and researchers can generate Dutch sentences based on dynamically generated input. SimpleNLG-NL was developed using an iterative process recreating sentences from a Wikipedia corpus. After four rounds, out of 86 sentences for which the input was manually written, a total of 75 (87.2%) could be generated in an acceptable manner. 69 of those were exact matches (80.2%). A proof-of-concept demonstrates that dependency trees can be automatically converted into input code for SimpleNLG-NL. In the current state, the proof-of-concept can only handle basic sentences, but the coverage can be increased by continuing the same development method. A more advanced version can be useful to quickly generate multiple variants of a sentence, for instance. SimpleNLG-NL will be released as open source software. Several suggestions for further development are made.