THE IMPACT OF NATURAL LANGUAGE ACCESS ON DATABASE DESIGN AND IMPLEMENTATION 
CHAIRMAN: 
Dr. S. Jerrold Kaplan 
Teknowledge Inc 
Palo Alto, CA 
PARTIAL LIST OF PARTICIPANTS: 
Mr. Jeffrey Hill 
Artificial Intelligence Corp. 
Waltham, MA 
Mr. Francisco Corella 
Symantec 
Sunnyvale, CA 
Dr. Jim Davidson 
Stanford University 
Stanford, CA 
Dr. Steven Shwartz 
Cognitive Systems, Inc. 
New Haven, CT 
Prof. Robert Wilensky 
University of California, Berkeley 
Berkeley, CA 
The central question to be addressed by this 
panel is how the provision of natural language 
access influences the design and implementation 
of database systems. Within this, there are 
(at least) the following four issues to discuss: 
i. The augmentation of data models to support 
the specific requirements of natural language 
access. 
This is perhaps the broadest category. Among 
Ehe particular demands of natural language access 
are: treatment of lexicon and lexical ambigu- 
ities; use of parenthetical expressions which may 
require simultaneous update and query; support 
for the generation of natural language responses; 
storage of discourse information; representation 
of text structure models; accurate processing of 
natural language updates; and other interactions 
that, though not unique to natural language 
access, have special problems in that context, 
such as natural language queries to dynamic 
databases requiring natural language statements 
of alerting conditions, or answering "meta- 
questions" - questions about structure. Current 
work in all these areas suggests the need to 
augment data models in particular ways. 
2. Ideas for data models that arise from the 
study of natural language Semantics. 
This category is related to the first, but 
the emphasis here is that researchers in natural 
language are developing ideas about modelling that 
can be very helpful in database modelling. One 
example of this is the uniform use of first-order 
logic as the "logical form" of natural language 
sentences as the representation of knowledge and 
data, and as the medium for deductive retrieval. 
Another example is the development of concepts of 
intensionality that support the notion of an 
"historical" database system. 
3. The organization of information in database 
systems to enhance the portability of natural 
language interfaces. 
Many researchers are investigating how to 
reduce the difficulty of moving a natural 
language interface from one database system to 
another. The problems in doing this include what 
information is needed, how the information should 
be divided into modules, and what algorithms are 
needed to acquire the necessary information for 
a new database. There are obvious ramifications 
on database structure for achieving various 
levels of natural language interface portability. 
4. The requirements imposed on conventional 
database systems that function as the back end of 
natural language processing systems. 
Looking at natural language research itself, 
most researchers develop their own information 
storage and retrieval mechanisms specifically 
tailored to the kinds of data structures that 
arise in natural language processing. However, 
there is some recent work in investigating how 
conventional database systems could be used in 
this role. This work is bound to have important 
effects in the future design of databases that 
deal in less regular structures than we are 
used to. 
ACKNOWLEDGMENT: 
The concept for this panel and the list of 
issues was developed originally by Dr. Jonathan 
King of Hewlett-Packard Laboratories, Palo Alto, 
California. 
