Knowledge Representation Systems Based on Natural Language
Papers from the AAAI Fall Symposium
Lucja Iwanska, Chair
This symposium addresses the problem of knowledge representation (KR) systems that closely parallel the representational and inferential characteristics of natural language (NL). Its main goal is to discuss the arguments for and against developing NL-based KR systems.
Among the arguments for the NL-as-KR-system approach are:
- Such systems are easy for people to use.
- Most human knowledge is encoded and communicated via NL.
- Such a system can automatically create and update its knowledge base directly from NL input.
- Such systems provide a uniform symbolic representation.
- The same representational and inference mechanism could be used when utilizing previous knowledge for processing new NL inputs.
- It is hard to match expressiveness and precision of NL, particularly in not (well) formalized domains.
- Many scientists believe that mental-level representation of knowledge is close in form to NL.
Among the arguments against the NL-as-KR-system approach are:
- NL is (highly) ambiguous.
- NL has (very) complex syntax, semantics, and pragmatics.
- NL is non-systematic, non-algorithmic.
- NL is (highly) context-dependent.
- NL is (merely) an interface. Inferencing does not belong with NL.