Syntactic Structural Kernels for Natural Language Interfaces to Databases

Abstract

A core problem in data mining is to retrieve data in a easy and human friendly way. Automatically translating natural language questions into SQL queries would allow for the design of effective and useful database systems from a user viewpoint. Interesting previous work has been focused on the use of machine learning algorithms for automatically mapping natural language (NL) questions to SQL queries. In this paper, we present many structural kernels and their combinations for inducing the relational semantics between pairs of NL questions and SQL queries. We measure the effectiveness of such kernels by using them in Support Vector Machines to select the queries that correctly answer to NL questions. Experimental results on two different datasets show that our approach is viable and that syntactic information under the form of pairs of syntactic tree fragments (from queries and questions) plays a major role in deriving the relational semantics between the two languages.

Generating SQL Queries Using Natural Language Syntactic Dependencies and Metadata

“…If we consider valid an answer given in the top three, our accuracy increases to 95%, achieving 99% on the 10-top ranked. Note that the accuracy at top-ranked answer can be improved by learning a re-ranker, as explained in [8], which can move correct answers in the top position.…”

Enhancement of Natural Language to SQL Query Conversion using Machine Learning Techniques

“…In this paper, a detailed comparison on different kernels were considered and appropriate combination were made to generate better results. Accuracies were considered for both datasets -Geo dataset (75.9%) and RestQueries dataset (84.7%) [8]…”

Semi-Supervised Learning of Statistical Models for Natural Language Understanding

“…There have been some prior works on learning semantic parsers that map natural language sentences into a formal meaning representation such as first-order logic [ 6 – 10 ]. However these systems either require a hand-built, ambiguous combinatory categorical grammar template to learn a probabilistic semantic parser [ 11 ] or assume the existence of an unambiguous, context-free grammar of the target meaning representations [ 6 , 7 , 9 , 12 , 13 ]. Furthermore, they have only been studied in two relatively simple tasks, G eo Q uery [ 14 ] for US geography query and R obo C up ( http://www.robocup.org/ ) where coaching instructions are given to soccer agents in a simulated soccer field.…”