The Art of Validation
Henk van den Heuvel · 2000
This article serves to satisfy the curiosity of the latter group of readers, at least to some extent. Validation of SLRs may refer to a variety of actions: 1. checking a SLR against a fixed set of requirements; 2. putting a quality stamp on a SLR as a result of the aforementioned check. If the database passes the check, then we say that it has been "validated"; 3. the evaluation of a SLR in a field test, thus testing the usability of the LR in an actual application. 4. ... SLR validation, as carried out by SPEX (acronym for Speech Processing Expertise Centre) , typically refers to the first type of action: the quality evaluation of a database against a checklist of relevant criteria. These criteria are typically the specifications of the databases, together with some tolerance margins in case deviations are found. The validation of language resources in general, and SLRs in particular, is a rather new type of activity in the area of language and speech technology. As more and more SLRs are entering the market, the need for validation of these resources increases, and therefore the best ways to accomplish validation need to be established. Validation of SLRs is of particular interest to the European Language Resources Association and its distribution agency ELDA (http://www.elda.fr/). ELRA offers a wide range of SLRs in its catalogue. Before distribution can proceed, the products must be subjected to quality control and validation. ELRA has established manuals for validation and has been actively persuading producers of Language Resources to adopt these as a means of adding value to the marketability of their products. ELRA, therefore, has started instituting a system that, in the long term, will yield a specification and quality control document to be issued with every ...