Glenn Fulcher

There is no published material in the language testing literature on the process of, or good practice in, developing an interface for a computer-based language test. Nor do test development bodies make publicly available any information on how the interface for their computer-based language tests was developed. This article describes a three phase process model for interface design drawing on practices developed in the software industry, adapting them for computer-based language tests (CBTs). It describes good practice in initial design, emphasizes the importance of usability testing, and argues that only through following a principled approach to interface design can the threat of interface-related construct-irrelevant variance in test scores be avoided. The article also charts concurrent test development activities that take place during each phase of the design process. The model may be used in CBT project management, and it is argued that the publication of good interface design processes contributes to the mix of validity evidence presented to support the use of a CBT.

The difficulty of speaking tasks has only recently become a topic of investigation in language testing. This has been prompted by work on discourse variability in second language acquisition (SLA) research, new classificatory systems for describing tasks, and the advent of statistical techniques that enable the prediction of task difficulty. This article reviews assumptions underlying approaches to research into speaking task difficulty and questions the view that test scores always vary with task conditions or discourse variation. A new approach to defining task difficulty in terms of the interaction between pragmatic task features and first language (L1) cultural background is offered, and the results of a study to investigate the impact of these variables on test scores are presented. The relevance for the generalizability of score meaning and the definition of constructs in speaking tests is discussed.