NATURAL LANGUAGE PROGRAMMING

Natural language programming

Abstract

Many Natural Language Interfaces (NLIs) to knowledge bases have been developed in order to provide easy access to structured data for casual users. However, those that have reasonable performance are domain-specific and tend to require customisation for each new domain, which, from a developer's perspective, makes them expensive to maintain and unattractive for practical applications spanning different domains. This paper explores how the performance of existing NLI systems to knowledge bases can be improved without the extra cost of extensive customisation. Additionally, usability of NLIs to knowledge bases is explored from two aspects: that of the developer who is customising the system and that of the end-user who is querying it. We discuss existing methods for increasing the usability of NLI systems and their impact on the overall retrieval performance.

Introduction

One of the most prominent benefits gained from the emergence of Semantic Web technology is the possibility to access data more efficiently, through the use of ontologies. Querying such data requires using formal languages such as SeRQL or SPARQL. However, the syntax of these formal languages tends to be too “artificial” and complex, especially for domain experts who are unfamiliar with such machine-like languages. (Biermann,2005)

To minimise the learning curve mandatory for the access of such data, many user-friendly interfaces have been developed. Some of them provide a graphical interface where users can browse the data, others offer a form-based interface for performing search whilst hiding the complexity of formal languages, e.g., KIM Platform. The most sophisticated ones provide a simple text box for a query, which takes full-blown questions or a set of keywords as an input, and return answers in a user-understandable form. (Danica,2008)

According to the interface evaluation conducted in, systems developed to support Natural Language (NL) interfaces are perceived as the most acceptable by end-users. This conclusion is drawn from a usability study, which compared four types of query language interfaces to knowledge bases and involved 48 users of general background. The full-sentence query option was significantly preferred to keywords.(Esther,2007) However, using keywords for querying was preferred to menu-guided, or graphical query language interfaces.

Literature Review

The development of accurate Natural Language Interface (NLI) systems is “very complex and time-consuming task that requires extraordinary design and implementation efforts” . According to Tablan, a major challenge in building NLIs is to provide the information the system needs to bridge the gap between the way the user thinks about the domain of discourse and the way information about the domain is structured for computer processing. In the case of Natural Language Interfaces to Knowledge Bases (NLIs to KBs), the domain knowledge is in the knowledge base. The knowledge base is typically created by instantiating classes defined in the domain ontology and relating them as per ontology definitions. Therefore, it is very important to consider the ontology structure and content when building NLIs to KBs. (Samuel2007)

Another big challenge is building a robust NLI due to the very difficult task of automatically interpreting natural language ...