6 CONCLUSION

We have surveyed error handling techniques for recognition, prediction, search, and other ambiguous or error prone systems. This surveys covers work in the areas of error discovery, error handling techniques, validation, and toolkit level support. Although we have uncovered extensive work in many of these areas, significant questions remain.

Error discovery

How can we improve the accuracy of error discovery? How should errors in error discovery be handled? What is the best technique for error discovery and how does this change depending on the situation?

Error correction techniques

Does error handling require new types of interfaces or widgets different from other interfaces? When should error handling occur? How integrated should error handling interfaces be with the normal workflow/interface?

Validation of techniques

How can we compare methods across applications? Is it possible to uncover general rules for the design of error handling interfaces?

Toolkit level support

Is it possible to separate out and encapsulate interface techniques for error handling? What techniques belong in such a toolkit? Is complete reversibility possible, and if not what are the alternatives? Are there efficient ways of keeping track of increasing numbers of probabilities?

In our own research, we are developing the PenPad system as a platform for answering some of these questions. Our first task is to develop a toolkit which supports the techniques uncovered in this survey. In addition to encapsulating standard techniques, we plan to continue to work on developing innovative new techniques and investigating existing HCI techniques which could be applied to error correction.