Section: New Results
Optimising Surface Realisation using Polarity Based Filtering
Surface realisation (the production, on the basis of a syntactico-semantic grammar, of all the phrases verbalising a given semantic content) is known to be exponential. However, polarity based filtering has been demonstrated by Guy Perrier to efficiently prune the search space when parsing. We applied polarity filtering to the surface realiser GenI (developed during the INRIA ARC GenI) and showed that it significantly improves the performance of this realiser  . We have also developed techniques to account for zero-literal semantic lexical items such as function words and pronouns within the polarity filter framework  .
For the purpose of wide-scale evaluation, we have developed a tree-anchoring interface which allows the realiser to use a common grammar, lexicon, tree selection and anchoring mechanism with parsers, and testing is well under way. The grammars we use are automatically generated using the XMG metagrammar compiler.