Classification of Attributes in a Natural Language Query into Different SQL Clauses

Proceedings of Recent Advances in Natural Language Processing, pages 497–506, Hissar, Bulgaria, Sep 7–9 2015. Classification of Attributes in a Natural Language Query into Different SQL Clauses Ashish Palakurthi†, Ruthu S. M.†, Arjun R. Akula†,∗ and Radhika Mamidi† †Language Technologies Research Center, IIIT Hyderabad, India. ∗IBM Research, Bangalore, India. ashish.palakurthi@research.iiit.ac.in, ruthus.m@students.iiit.ac.in, arakula@in.ibm.com, radhika.mamidi@iiit.ac.in Abstract Attribute information in a natural language query is one of the key features for con- verting a natural language query into a Structured Query Language1 (SQL) in Nat- ural Language Interface to Database sys- tems. In this paper, we explore the task of classifying the attributes present in a natural language query into different SQL clauses in a SQL query. In particular, we investigate the effectiveness of various features and Conditional Random Fields for this task. Our system uses a statisti- cal classifier trained on manually prepared data. We report our results on three differ- ent domains and also show how our sys- tem can be used for generating a complete SQL query. 1 Introduction Databases have become one of the most efficient ways to store and retrieve information. Database systems require a user to have the knowledge of structured languages in order to be able to retrieve information from them. As a result, it becomes difficult for people of non-technical background to use databases. Natural Language Interface to Database (NLIDB) (Androutsopoulos et al., 1995; Catalina Hallett and David Hardcastle, 2008; Pa- zos et al., 2002; Popescu et al., 2003; Giordani and Moschitti, 2009; Gupta et al., 2012) systems provide an interface through which a user can ask a query in natural language and get the required information from the database. NLIDB systems translate the user’s natural language (NL) query into a SQL query, thereby allowing the user to re- trieve the answer from the database. However, 1Structured Query Language is a specialized language used for relational database management and data manipu- lation. NLIDB systems are not widely used because of their inability to process ambiguity and complex- ity of natural language, which makes them more error prone. Thus, it becomes very important to capture even the smallest of the information from a NL query before converting it into a SQL query. A relational database contains objects called ta- bles in which information is stored. These tables contain columns and rows. The column names in the tables are known as attributes. A SQL query is composed of different SQL clauses like SELECT, FROM, WHERE, GROUP BY, HAVING and ORDER BY. Since clauses in a SQL query have attributes, attribute information becomes very important for an effective conversion of a NL query into a SQL query. Explicit attributes are the attributes men- tioned by the user in the NL query text. When a NL query is converted into SQL query, explicit attributes may belong to different SQL clauses. In this paper, we use Conditional Random Fields (CRF) for classifying the explicit attributes in a NL query to different SQL clauses. 2 Related Work There have been significant …

Content truncated to 500 words