Next: Manual Validation Contents
Up: The Validation of Speech
Previous: Check List
This chapter describes those validations of the corpus data
that cannot be performed
automatically and therefore require considerable efforts in terms of manpower.
Typically, this will be carried out by a group of selected validators -
usually with special skills and native speakers of the corpus
language - under the supervision of a person
responsible for the logistics. This supervisor must be able to
Since the variety of annotations is large, we cannot give detailed
advice on how to validate all the different annotations schemes.
Instead we will concentrate on some practical hints that will most
likely be useful in all kinds of manual corpus validations.
- select (maybe also train) the validators
- organize the selection of validation material
- organize the distribution of work load to the validators
- collect and check the results
- do quality control (e.g. by taking random sample checks)
- calculate the results (some basic statistics)