In Defense of Symbolic NLP

The paper examines the benefits and the drawbacks of two competing approaches to natural language processing: statistical (probabilistic) and symbolic (deterministic). While the statistical approach is gaining popularity, better results may often be obtained using symbolic methodologies. The paper argues that the benefits of consistent deterministic parsers and lexica are well worth the time and effort required for their development. The Meaning Theory is specifically recommended as the best theoretic framework for cross-lingual NLP applications including machine translation and text retrieval. The paper concludes that best results are obtained using a combination of the two approaches when statistical methods are applied to the output of a deterministic parser.