Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Test Set Construction of a Natural Language Interface to a Database

IP.com Disclosure Number: IPCOM000106486D
Original Publication Date: 1993-Nov-01
Included in the Prior Art Database: 2005-Mar-21
Document File: 2 page(s) / 72K

Publishing Venue

IBM

Related People

Winghart, OJ: AUTHOR

Abstract

Proposed is a method to build Test Sets so that a Natural Language Interface to a Database can be objectively evaluated for global breadth and depth in specific areas. A set of test cases is built up with increasing coverage of a subset enabling specific features to be tested such as processing from inputs in several natural languages.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Test Set Construction of a Natural Language Interface to a Database

      Proposed is a method to build Test Sets so that a Natural
Language Interface to a Database can be objectively evaluated for
global breadth and depth in specific areas.  A set of test cases is
built up with increasing coverage of a subset enabling specific
features to be tested such as processing from inputs in several
natural languages.

      With any application, e.g., the conceptual domain of
personnel-data in a company, several functions of information
retrieval from a database have to be fulfilled.  Additionally in a
particular natural language, e.g., English or German, these database
querying functions should be expressed as natural language queries.
To test how an interface to a datablase can accept and support
multi-language naturally occurring queries in a demonstrable way,
even though the possibilities of input in natural language are in
principle infinite and unforeseeable, the following scheme is
proposed.

      There are two goals that the disclosed methodology addresses.
The first is Construction;  to build test sets of real queries, i.e.,
spontaneous formulations by users.  The second is Segmentation;  to
enable testing of both a global test set and specific area subsets
without query redundancy.  Both types of test are necessary to
evaluate a natural language interface.

Method

  Global schema:

DB query --> User ---> Filtered --------
functions    queries   user queries    !
                                       V

Developer ------------------------>  Global ------> Segmented
queries                              set            global set

Tasks for construction (1):

a. Get familiar with application definition

b. Generate an abstract specification of what to ask within the
application.

c. Have an independent native speaker generate formulations on
   paper.

d. Type in and run these 'user queries'.

e. At this poin...