Practical Experience with Grammar Sharing in Multilingual NLP

  • Michael Gamon ,
  • Carmen Lozano ,
  • Jessie Pinkham ,
  • Tom Reutter

MSR-TR-97-16 |

In the Microsoft Natural Language Processing System (MSNLP), grammar sharing between English, French, Spanish, and German has been an important means for speeding up the development time for the latter grammars. Despite significant typological differences between these languages, a mature English grammar was taken as the starting point for each of the other three grammars. In each case, through a combination of adding and deleting a modest number of grammar rules, and modifying the conditions on many others, a broad-coverage target grammar emerged. Tests indicate that this approach has been successful in achieving a high degree of coverage in a relatively short period of time.