Next: Instantiations of the interpretational
Previous: Methods for combination of
The final test-kit includes at least a test suite, i.e. a carefully
composed and structured set of inputs. The test suite will correspond
to the set of error examples we have already discussed, as follows:
The correct inputs can be constructed according to general language
rules or taken from excerpts of corpora, or be devised from a
combination of these two. Likewise, the erroneous inputs as well as
the deliberately misleading type of input can be constructed in the
same manner. We do, however, stress that all erroneous or misleading
input should be checked against non-proofed texts, and ideally
supplemented with real-life occurences.
- For each attribute there is a horizontal tripartition: a set of
erroneous inputs, a set of correct inputs and possibly a set of
deliberately misleading inputs (setting `traps' for the system);
- For each attribute there is a vertical ordering: the inputs
cover the range from the most simple instantiation of a construction
to the more complex realisations of it.
One obvious shortcoming of this method is its limited ability to
control and substantially check the effect of interacting phenomena.
To that end, a set of random sample texts (non-proofread real texts)
must also be included in the test-kit.