![](bg5.png)
• The students passed the exam.
all(student(X), the(exam(Y), pass(X,Y)))
In the first example, the first entity in the relation is
quantified by the universal quantifier and the second
by the existential quantifier. Based on our current ex-
perience, and the examples encountered, usually only
one occurrence of the variable quantified by the exis-
tential quantifier is involved in the relation. In such
cases the LF quantifier “ex” is interpreted as a unique
existential quantifier. We are therefore in a case were
many occurrences of the first variable are related to
one occurrence of the second variable. By definition,
this is a many-to-one relationship. In the second ex-
ample, the interpretation is much more stronger since
we have a unique existence interpretation for the sec-
ond “the”. We have again a case of a many-to-one
relationship. Let us now consider the following set of
sentences and their LFs:
• The company maintains a description for each item
of stock.
all(item(X,stock), the(company(Y), ex(description(Z),
for(X,maintain(Y,Z)))))
• The student passed all exams.
the(student(X), all(exam(Y), pass(X,Y)))
The NP each item of stock, in the first sentence,
suggests that we are talking about a particular stock
that contains many items. Therefore wa have a one-
to-many relation between the entity “stock” and the
entity “item”. In general, sentences where the first
entity is singular and quantified by the definite article
and when the second entity is quantified by the uni-
versal quantifier define one-to-many relationships. A
typical example is the second sentence. Let us con-
sider now the following sentence and its LF:
• The student passed the exams.
the(student(X), the(exam(Y), pass(X,Y)))
The previous rule does not apply because the sec-
ond entity is itself singular and quantified by the
unique existential quantifier. As this example suggest,
we are talking about a particular student who passed
a particular exam. In this case, we infer a one-to-one
relationship between the entities.
These are the main cases where our approach can
help in identifying the degrees of the relations from
the LFs of the sentences. In other cases, when it is
difficult to predict the degree of a relation, we let the
user determine it.
5 CONCLUSION AND FUTURE
WORK
This paper has presented a novel approach that can
help an analyst produce an initial ERM from specifi-
cations written in NL. The approach makes use of NL
analysis techniques to translate sentence to LFs which
are then used as a basis for identifying the entities
and relationships. The quantifiers in the LFs also en-
able the identification of the degrees of relationships
in some common cases.
The approach has been implemented in Prolog and
tested on some examples. To date, the most interest-
ing application has been to a specification of a flight
planning system that was written independently of our
work (Hepworth, 1988; Vadera and Meziane, 1994).
In that case study, the approach worked well in that:
• The majority of entities and relationships were cor-
rectly identified. The system identified 55 entities
of which only 1 was thought to be spurious and
none had been missed. It identified 52 relationships
of which were incorrect and none overlooked.
• Most of the degrees were correctly identified. The
degrees for 49 of the 52 identified relations were
correctly predicted.
Future research aims to develop the techniques so that
a wider range of sentences and more structured ob-
jects, like tables, can be handled by the NL process-
ing phase. This should enable a broader evaluation
of the approach on larger specifications. The results
obtained with our current implementation are encour-
aging and suggest that further research may lead to an
invaluable practical aid for producing ERMs.
REFERENCES
Allen, J. (1987). Natural language understanding. The
Benjamin/Cummings Publishing Company, Inc.
Ashworth, C. and Goodland, M. (1990). SSADM: A practi-
cal approach. McGraw-Hill Book company.
Bowers, D. (1988). From data to data base. Van Nostrand
reinhold (U.K) Co. Ltd.
Gane, C. and Sarson, T. (1979). Structured System Analysis.
Prentice-hall Software series.
Hepworth, B. (1988). An introduction to Z. Technical Re-
port BAe-WIT-RP-GEN-SWE-152, Systems Comput-
ing Department, British Aerospace Ltd.
Hess, M. (1985). How does natural language quantify? In
Second Conference of the European Chapter of the as-
sociation for Computational Linguistics, pages 8–15.
McCord, M. (1990). Natural language processing in Prolog.
In Adrian, W., editor, A logical approach to expert sys-
tems and natural language processing Knowledge sys-
tems and PROLOG, pages 391–402. Addison-Wesley
Publishing company.
Meziane, F. (1994). From English to Formal Specifications.
PhD thesis, University of Salford.
Vadera, S. and Meziane, F. (1994). From English To Formal
Specifications. The Computer Journal, 37(9):753–
763.
ICEIS 2004 - DATABASES AND INFORMATION SYSTEMS INTEGRATION
642