Learning Constraint-Based Grammars from Representative Data - 2008

Explore a comprehensive lecture on constraint-based grammars and their learning from representative data, presented by Smaranda Muresan at the Center for Language & Speech Processing (CLSP), Johns Hopkins University in 2008. Delve into the integration of syntax, semantics, and learning in natural language processing systems, focusing on a new grammar formalism called Lexicalized Well-Founded Grammar (LWFG). Discover how this approach combines deep language understanding with scalability, utilizing compositional and ontology constraints to access meaning during parsing. Learn about the unique search space for grammar learning, represented as a complete grammar lattice, and its implications for solution uniqueness. Examine the practical applications of LWFG in populating terminological knowledge bases from medical texts. Gain insights into Muresan's research interests, including language learning and understanding, machine translation, and relational learning, as well as her vision for the next generation semantic web.

Introduction
Research interests
Natural language processing
Next generation semantic web
Example
Semantic Representation
Grammar Formalism
Grammar Formalism Architecture
Semantic Representations
Formalism
Semantics
ConstraintBased Grammar
Learnability Theorem
Inductive Logic Programming
Background Knowledge
Representation Lattice
Grammar Approximation
Different Types of Algorithms
Concrete Example
Approximation
Problem formulation
Concept identity
Qualitative evaluation
Advantages
Conclusion
Future Directions
Automatic Population of Knowledge
Machine Transmission