Semantic Parsing via Paraphrasing

What is Semantic Parsing ?

machine is able to learn that the constructions"What does X do for a living?”, “What is X’s profession?”,and "Who is X?”, should all map to the logical predicate Profession.

Profession is called a logical predicate here.

Further more, based on the predicate, machine is able to give a corresponding answer from the corpus.

What if no match can be found thoughout the corpus ?

Tradictional methods

use data which pairs natural language with the KB.

‍Drawback‍

this leaves untapped a vast amount of text not related to the KB. For instance, the utterances “Where is ACL in 2014?” and “What is the location of ACL 2014?” cannot be used in traditional semantic parsing methods, since the KB does not contain an entity ACL2014, but this pair clearly contains valuable linguistic information.

‍Approach in the paper‍