12.3 Natural laNguage INterfaCe to uNstruCtured data 331
In order to tolerate lexical variations in the input questions, the prepro-
cessor builds a semantic knowledge base composed of interpretation rules
and semantic sets for all possible relationship and attribute names in the da-
tabase. This helps to build the same semantic representation, and hence, the
same SQL query for various questions.
In the preceding NLI2DB example, we saw that the link parser, which
collects information at a syntactic level, also can be used at a semantic level.
The focus of NLI2DBs today has shifted to handle problems at a higher level
of linguistic analysis. Therefore, the development of NLI2DB systems that
handle language-related phenomena is an active area of research.
12.3 NATURAL LANGUAGE INTERFACE TO UNSTRUCTURED DATA
The process of establishing an interaction between human being and ma-
chine was made successful in 1966 with the ELIZA system, which was
developed by Joseph Weizenbaum (Weizenbaum, 1966). ELIZA worked
by simple parsing and substitution of keywords into phrases stored in a
knowledge base. Though ELIZA did not employ any language-related
phenomena, it still remains a milestone simply because it was the rst time
a programmer had attempted such a human–machine interaction with the
goal of creating the illusion of human–human interaction.
Dialogue systems were historically the domain of artificial intelligence
(AI) researchers. These systems are a sort of natural language interface to
unstructured data. These systems do not restrict themselves to interacting
with data in database tables only; data from various sources can be used
and accumulated.
Moving forward through the history of the natural language interface to
unstructured data research brings us to SHRDLU and GUS. Both of these
systems are dialogue systems interacting on information about a restricted
domain. The difference between these systems and systems such as LUNAR
lies in their dialogue capabilities. GUS was designed to simulate a travel advi-
sor and had access to a database containing limited information about airline
flight times. SHRDLU is probably the better known of these two systems.
It controlled a robot arm in a virtual microworld consisting of a tabletop
strewn with colored blocks of varying shapes and sizes and a box into which
the blocks could be placed. While example conversations with SHRDLU
76473_CH12_Akerkar.indd 331 8/11/09 10:16:11 AM

Get Knowledge-Based Systems now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.