In the previous issue of Interacting with Computers, I offered a critique of Task Analysis (TA) arguing that, in general, task analysis techniques had failed to appreciate the significance of the shift in emphasis -towards logical modelling of the system and away from physical modelling -which had taken place during the early 1980s in information systems. As a result TA (if used on its own) would produce poor system designs because it fails to achieve sufficient device independence (Benyon, 1991). In their commentary on this paper, Addison (Diaper and Addison, 1991) defended TA in general and TAKD (e.g. Diaper, 1989b) in particular against the criticisms and made a number of assertions about my stance regarding humans and computers.In this reply, I intend to clarify firstly what I mean by a data-centred view, secondly what I mean by device independence and finally why the datacentred view is a vital part of systems development. In doing this I hope to deal with many of the specific points raised by Diaper and Addison.Underlying this discussion is a view of how system designers should undertake their designs and a theoretical, or philosophical basis of human-computer interaction (HCI) provided by the data-centred view. Whilst both of these are important subjects, there is neither the space nor the time to give adequate coverage here. However, I am sure that both the theory and the practice of HCI will benefit from this continuing debate on the preferred role of task analysis in system design.My original concern about TA stemmed from the claim made by its advocates that 'task analysis is potentially the most powerful method in HCI...for producing requirements specifications' (Diaper 1989a, preface). I believe this view is unsatisfactory because TA does not focus sufficiently on the data in the system. This paper is aimed at explicating this position.The central part of this paper is an exposition of data models as conceptual models which capture both a human and a computer view of systems. This discussion is necessary in order to understand the notion of device independence which is presented in the following section. Before turning to the details of data, it is important to obtain a perspective on system development.
SYSTEM DEVELOPMENTThere are many characterizations of how systems are developed and how systems should be developed. The model used in (Benyon 1991) was chosen to emphasize the importance of conceptual modelling and task allocation rather than to prescribe a system development method. Many methods advocate a task allocation phase (such as Avison and Wood-