In recent years a great deal of work has been devoted to the design of interactive database query systems suitable for 'end users'. By contrast, there appears to be little comparable work on interactive facilities for database modification. This is surprising for two reasons. Firstly, update operations are of great practical significance. Users of transaction processing systems often spend a greater proportion of their time modifying a database than retrieving information from it. Secondly, update is a hard problem. It is vital to safeguard database consistency, and that implies a thorough understanding of the semantics of the data in question. A quick glance at the facilities for maintaining consistency in many current database management systems is enough to demonstrate that this is not a trivial matter.This paper describes an interactive approach to database modification which exposes the semantic problems and overcomes them. Four conceptually simple update operations - insert, delete, change and copy-are formulated in terms of a model of database consisting of entity sets and mappings. These operations lead naturally to the idea of update-by-dialogue: a system in which each of them is implemented as an interactive dialogue. The user is guided in each operation, relieving him of the burden of remembering how the database is structured and also avoiding the effects of negligence. Furthermore, the dialogues are driven from a stored description of the database structure and thus are completely general. They can be applied to any database and are even immune to changes in the database structure.

The concept of normalisation and the definition of third normal form occupy a central place in the relational model of database. Normalisation has been described as an analytic process in which an initially unconstrained collection of relations is successively decomposed into a collection of smaller relations satisfying the constraints of the three normal forms. Normalised relations constitute the basis on which data manipulation languages - relational algebra, relational calculus and others - have been defined. As yet, however, there is no generally recognised data definition language for the relational model. A pre-requisite for this would be a constructive definition of third normal form, that is, a definition in terms of rules for building normalised relations from simple domains. This paper considers how such a definition may be achieved by applying two ideas derived from the study of programming languages; the notion of data type, and the notion of a selector as used in formal language definition. The result is an abstract syntax of relations. This is compared with the standard definition of normalisation and its completeness is critically examined.