NATURAL LANGUAGE DATA BASE ACCESS WITH PEARL 
Wendy Lehnert and Steve Shwartz 
Department of Computer Science at Yale University, New Haven, 
Ct., and Cognitive Systems, Inc. of New Haven, Ct. 
Data bases access is typically approached as a prog~anm- 
ing problem rather than a user interface problem. Wez. conseq 
ently see situations where only programmers can access inform- 
ation in data bases, in spite of the fact that the people 
who need this information are typically non-pro~'a~ners. The 
most desirable way to extend data base access capabilities to 
naive user populations is by means of a natural language inter- 
face or "front-end" system that can process requests stated in 
the user's natural language. PEARL is a domaln-lndependent 
system for natural language analYsis that can be customized 
to handle domain-specific queries to existing data bases. 
PEARL analyzes English input with expectation-driven p~rsing 
techniques end question answering strate@les developed by 
Artificial Intelligence researchers (Lehnert, /1978/, Sohank 
and Riesbeck /1976/, Riesbeok /1975/, Gershmsn /1979/, Birn- 
baum and Selfridge /1979/, and produces code in whatever query 
language is currently being used to access a specific data 
base. PEARL can therefore be adapted to any data base that 
has a query language access module. 
In any realistic application of natural language anal- 
ysis, we can expect users to make mistakes, forget to specify 
necessary parameters, and use words that are not in the 
system's vocabulary. PEARL is designed to handle these ine- 
vitable occurrences with interactive user dialogues that 
- 167 - 
allow the user %0 augment, verify and alter the system's 
interpretation of an input request. PEARL accepts requests 
in unrestricted English: no constraints are placed on the 
user to stay within a restricted vocabulary or artificial 
syntax. 
The following examples of PEARL in action are taken 
from a version of PEARL that has been customized to handle 
requests made by geologists to a data base conte/ning inform- 
ation about oil wells. The geologists in this user population 
axe oil exploratlonlsts with acoess to a °~phlcs package 
that can generate maps on the basis of information in the 
database. The map requests given be.low are highly domain- 
specific and therefore contain many technical terms used by 
oil explorationist s. 
Yale TOPS-20 Command processor 4(560)-I 
. (explorer) 
~XPI~P~R Version 02 9/23/81 
E~ADY 
*I want a map of all wells drilled before May 1, 1980 but 
asince April 30, 1970 that show the tog formation below 2000 , 
~s-~e themselves deeper than 5000", and are operated by Emitt. 
I only want to see wildcat wells where the operator reported 
9a drilli~ problem, and that have macha~Lical logs, blew out, 
and were drilled by Lexaco within the area defined by 
%latltude 30 deg 20 mln 30 sec to 31:20:30 and 80-81. Make 
sthe scale 1" = 300 meters. 
User requests a POSTED map_ 
LATITUDE 30.34167 - 31.34167 
LONGITUDE 80 - 81 
output device : PAPER 
filters: DRILLING DEPTH > 5000" 
COMPLETION DATE > = 4/30/1970 
GOMPLETION DATE < 5/1/1980 
LEXACO ORIGINAL OPERATOR 
BLOWOUT WELL 
MECHANICAL LOG PILE WELL 
DRILLING PROBLEM 
WILDCAT WELL 
- 168- 
E~ITT CURRENT OPERATOR 
PENETRATION OF ~CG WHERE TOP DEPTH > = 2000 • 
scale of 1" = 984.252 
READY 
* Contur a third ordr trend difference on the tcg. 
By CONTUR, do you mean CONTOUR (Y or N) . y 
By ORDR, do you mean ORDER (Y or N) * y 
DIPFERENCE not understood 
Would you like to propose a synonym or correction? (Y or N) *y 
Enter word: *residual 
A map region was not specified. 
Do you want the same geographic region as before? (Y or N) *y 
Please specify contour interval. 
.100 feet .~. 
A scale was not specified, 
Would you like the scale f~om the previous request? (Y or 1;) 
Previous scale = 984.252 
*y 
User requests a STRUCT map 
LATITUDE 30.34167 - 31,341 67 
LONGITUDE 80 - 81 
output device : PAPER 
statistics: 
TREND-EDITTING 
3RD-ORDER RESIDUAL 
on TCG 
CI = 100 • 
scale, of I" = 984.252 " 
PEARL uses conceptually-driven techniques for language 
ana~sis which mast be customized for specific domains and 
task orientations. To customize PEARL for a new data base 
application, we mast conduct a thorough study of typical l~- 
uage use for that application, so that PEARL will be able to 
handle technical or idiosyncratic constructions and vocabulary 
items. Once PEARL is capable of processing a realistic spect- 
- 169 - 
rum of possible language use, PEARL automatically adjusts to 
the expertise of individual users. Expert users are allowed 
to state their requests in as concise a manner as desired, 
while inexperienced users are carefully queried about possib- 
le ambiguities and insufficient specifications. This inherent 
flexibility is a natural by-product of a~7 competent lan6~age 
processing system, and a necessary featul~ for effective and 
satisfylng m-~/m~ch~e conznuzLtcation, 

References

Birnbaum, L. and Selfrldge, M. (1979). l>roblems in Conceptual 
Analysis of Natural Ienausge. Yale University, Computer 
Science Dept., Research Report No. 168. Also in R. 
Schsnk and C. Riesbe0k, (eds.), Inside Oompt~ter Under- 
Standln~, Lawrence Erlbaum Associates, Hillsdale, NJ, 
1981. 

Gerabman, A.V, (1979). K~owled~e-Based Pars:Lug. Ph.D. Thesis. 
Yale University, Computer Science Dept., Research Re- 
port No. 156. 

Lehnert, W. (1978). The Process of ~uestion Answering. Law- 
fence Erlbaum Associates, Hillsdale, New Jersey. 

Riesbeck, C. (1975). Conceptual anal~sis. In R.C.SchanE, (ed.) 
Conceptual Informatio D 1>;rocessin~. North Holland, 
Amsterdam. 

Riesbsck, C.K. and Schank, R.C° (1976). Comprehension by com- 
puter: Expectation-based analysis of sentences in con- 
text. Yale University, Computer Science Dept°, Research 
Report No. 78. Also in W.J.M. Levelt and G.B.Ylcres d" 
Arcals (eds.), Stpdies in the Perception of Ie~u~ua~e. 
John Wiley and Sons, Ltd., Chichester, England, 1979. 
