1. A Comparative Framework for End User Development
Alistair Sutcliffe
(Nikolay Mehandjiev, Darren Lee)
Department Of Computation
UMIST
Manchester, UK

2. Objectives
to identify the contribution from different strands of EUD research;
to expose research issues and indicate a future research agenda;
to act as a critical evaluation tool for EUD environments and systems;

3. Approach
A literature survey across the spectrum of disciplines associated with End-User Development.
Requirements
Engineering
HCI &
Cognitive
Science
User
Communities
e.g. Bioinformatics
Software
Engineering
Artificial
Intelligence
Information
Systems
Psychology
of programming
CSCW
Critically evaluate research and industrial state of the art, create a review paper

4. Framework Dimensions 3 Dimensions Representation
The nature of the representations
used in the programming task
between computer and end-user.

5. Framework 3 Dimensions Representation Scope
The nature of the representations
used in the programming task
between computer and end-user.
Scope
The breadth of EUD problems an
environment can support.

6. Framework 3 Dimensions Representation Scope System Initiative
The nature of the representations
used in the programming task
between computer and end-user.
Scope
The breadth of EUD problems an
environment can support.
System Initiative
The degree to which the user is
required to directly specify software.

7. (Abstract Vs Concrete)
Representation
(Abstract Vs Concrete)
Medium for exchange of information between user and machine.
Natural or designed language- text, icons, graphics
Non-linguistic representations- graphics, interactive microwolrds, TUIs, physical artefacts, augmented reality
Linguistic representations= abstract syntax + lexicon, text, symbolic, iconic, natural language
Modality of communication, audio, speech, visual, haptic
Domain and representation interact. e.g. Mathematics Research.
Expressed via a language.
Irrespective of sensory channels.
Language (and attributes) +sensory channel = modality (Bernsen)
Non-Linguistic models
Iconic/Visual Programming has syntax and semantics therefore a language.
Interesting Question
Visual Programming arguing visual is inherently natural (concrete) because it visual!, claim that the lack of language (in the textual sense) also makes it more concrete.
Origin Of Abstraction
Generalisation – to generalise need to be abstract.
False metaphors- Forcing thinking about domain in terms of objects when more natural in terms of flows.
Nature of Domain- Highly abstract in real world.

8. (Task or Domain Specific Vs General Purpose)
Scope
(Task or Domain Specific Vs General Purpose)
Problem coverage of an EUD tool.
General purpose
Turing-complete, any task in any domain.
Architecture in the sense of Fischer’s DODEs.
Modular environment with programmability.
Specific Task - Domain
Horizontal domains- e.g. finance, training, CAL systems
Vertical domains- e.g. ecommerce, bioinformatics, medicine
Information systems tasks, support for tasks in the real world
Dimension interaction inevitably occurs. Notably General Purpose and Abstract.
Problem coverage
Programming as problem solving.
The task the end-user is using the EUD tool to support.
Views of GP
Turing complete – powerful as mainstream languages a matter of skill and effort on part of programmer.
Views of Task and Domain
For a task is user: searching for information OR searching for Biochemical information.

9. Scope and domains End user Software system Information systems Sensors
& devices
Modelled
world
Controlled
world
Real
world

10. System Initiative
(Active Vs Passive)
degree of user responsibility for specifying the solution
Passive
User driven- language or composable artefacts
Possibly supported by a critiquing system
Active
The machine tracks user behaviour and infers a software solution- assumes an existing artefact
System intervenes to help design –partial initiative
Machine generates solution from natural language instructions or observes RW and learns- frontiers of AI
Passive
Therefore DODEs and derivatives are passive.
PBE - Macro recorders.
Active
Automatic programming/ PBE – inference
Still the question of declarative programming paradigm at its purist Nik argues that it is possibly active.

11. Other issues Social Context of EUD
-Collaborative EUD and EUD of CSCW systems
-Autocratic Vs Democratic and Individual Vs Group EUD.
- Communities of EUD developers, markets for components, scripts, etc
2. Communication Model
- Sensory channel (Modality)- audio, visual, haptic
- Modality + Representation = EUD Paradigm
- Usability, User Cognition and Cognitive Dimensions
Machine Model
- Transparency and Execution Vs Interpretation.
- Integration with Software Engineering Paradigms- OO, Agents, etc
- Software reuse .legacy integration
[This is an additional slide, recommended by Nik to back up the other work we are doing]

12. Psychological issues Expressability and complexity
Learning or display based cognition
Intuitiveness, affordances, and transparency
User cost, effort and errors

13. Develop or Adapt ?
Increased
Learning
and effort
Tailoring, customisation, personalisation- low effort, limited scope
Configuration, design by composition- more effort, improved scope but limited by components
Instruction led design- high effort, good scope
Communicate requirements + machine designs- low effort, good scope

14. DODEs
An architecture for reducing the conceptual distance between the problem-domain and solution artefact.
Progamming as construction and design.
Views programming as argumentative process between system and user. Critiquing systems support this activity.
Model world, editors, interpreter, solution templates, tutors, critics
DODEs can evolve as the end-user becomes more skilled or the domain or task changes.

15. AgentSheets
Extends the DODEs concept by as a general purpose environment + language.
Introduces true programmability through graphical if-then- else rules and simple message passing.
Architecture and domain independent tool for simulation and building executable programmes.
Build interactive graphical micro world
Once a set of agents are implemented it becomes a domain specific tool.
Abstraction can be controllable by the end-user.
The user can work with predefined agents representing concrete metaphors
or drop down to the more abstract to define new agents as needed.

16. TAMBIS
Bioscience databases use multiple query languages from SQL to natural language.
Ontology, thesaurus, domain specific lexicon.
TAMBIS provides a visual ontology in which queries are built up through navigation of a 2D space representing the biochemistry domain.
TAMBIS infers this navigation into the query language program for each of the databases.
Visual ontology is produced from textual DL called GRAIL
TAMBIS uses concrete representations straight from the domain.
The scope is highly task and domain specific.
The system takes the users requirements and specifies the software that will be executed. Therefore active initiative.

17. Using The Framework
DODEs “Domain-Orientated Design Environments”. Fischer Univ of Colorado.
AgentSheets. Repenning Univ of Colorado.
TAMBIS “Transparent Access to Multiple Bioinformatics Information”. Goble et al Univ Of Manchester
TAMBIS- Baker Gobel et al. IMG at Manchester working with school of biological science. Task and domain specific EUD tool.

18. Position In Framework

19. Position In Framework
DODEs
Domain Specific
Passive
Concrete

20. Position In Framework
AgentSheets
General
Passive
Concrete/abstract

21. Position In Framework TAMBIS Task and Domain Specific Semi-Active
Concrete

22. Conclusion Framework based on common themes across research areas.
Permits the comparison of tools and theories.
It will help to identify trends and areas for future work.

23. Challenges
Merging paradigms- NL with interactive worlds, inference to help development
User freedom, expression and assistance-where is the trade off ?
Device sensor integration, solution power and learnability trade offs
Sharable composable development worlds
EUD or design support environments ?

24. Challenges II
End user requirements languages- integration of symbolic programming and clarification dialogues (NL research with SE/HCI)
Abstraction trade offs- general or domain specific EUD environments
Escaping from the expert knowledge bottleneck- learning or intelligent
acquisition envs
Evolutionary computing and EUD- users set parameters for generated solutions
Socio-technical solutions- design for user motivation and reward

25. Future work Review framework and research issues
Use the framework in a critical review of existing EUD tools and theories.
Conduct an industrial survey to identify successful commercial applications of EUD tools and techniques.

26. Position In Framework
Java
General Purpose
Passive
Abstract