next previous contents
Next: Evaluation of the framework Up: Second round of testing Previous: Evaluation of the grammar

Conclusion

We have shown how a grammar checker may be tested, relying partly on its own specifications, partly on the requirements of a specific user group.

The test results mentioned above are based on so small amounts of data that no general conclusion can be drawn from them. However, a few general comments can be made: some grammatical phenomena are not handled very well, whereas others pose no problems. False flagging can be a problem on the sentence level, and no flagging as regards prepositions is an overwhelming problem -- at least if this is a common source of error of the type of user whose requirements are taken into account. The most important problem, however, is the lack of consistency. Sentences containing the same problem are not treated the same way as in the examples under Articles and Aspect above. However, the semantic problems we encountered cannot be ascribed to the grammar checker, as they are outside its scope.

To sum up, we have chosen the following strategy for testing a grammar checker:



ceditor@tnos.ilc.pi.cnr.it