Multilingual Item Generation
The n-layer model is a flexible structure for item generation, thereby permitting many different but feasible combinations of embedded elements. In addition to generating more diverse items, one possible application of n-layer modeling may be in generating multilingual test items. Different languages require a different grammatical structure and word order (Higgins, Futagi & Deane, 2005). With a 1-layer model, the grammatical structure and word order cannot be easily or readily manipulated because the generative operations are constrained to a small number of elements at a single level. However, with the use of an n-layer model, the generative operations are expanded dramatically to include a large number of elements at multiple levels. Language, therefore, can serve as an additional layer that is manipulated during item generation.
Earlier in this chapter we described a method for using the plausible values specified in a cognitive model to generate new items by systematically replacing the item model content using computer algorithms. These replacement values are specified in the cognitive model as elements. As item models become more complex due to the requirements specified in cognitive models and in the linguistic complexity required for adapting items into different languages, the number of elements used for item generation dramatically increases. The increase in the number elements is problematic because it complicates the programming task and it affects the computation time required to run IGOR. To address this problem, Gierl, Lai, Fung and Zheng (in press) introduced the concept of a linked element as a way to facilitate the IGOR programming task and to increase IGOR’s computational speed.
Recall that the use of layered elements permits content to be embedded within content in an item model (see Figure 21.6). Layered elements, therefore, have a “vertical” function for item content (i.e., content within content). Linked elements also expand the capabilities of item modeling by permitting content to be transformed within an item model. For multilingual AIG, the transformation is from one language to another. Linked elements, therefore, have a “horizontal” function for item content (i.e., content in language 1 is transformed to content in language 2). The linked elements used for language transformations can function in four different forms: words, key phrases, single sentences and multiple sentences. These four forms are then used to adapt words, phrases and sentences from one language to another to permit multilingual AIG.
In our current example, we generated infection and pregnancy items in English. However, Canada is officially bilingual. Therefore, the Medical Council of Canada, the agency that licenses physicians, must administer items in both English and French. To accommodate item development in this scenario, we demonstrate how items can be generated simultaneously in English and French. The multilingual AIG example was created with the help of a bilingual medical content specialist. Four types of linked elements were identified and used for multilingual AIG in our example. First, linked elements are specified in the form of a word. These elements require the direct translation or adaptation of a single word between languages in the n-layer item model. Second, linked elements are specified in the form of a key phrase. These elements require the direct translation or adaptation of key phrases between languages. Third, linked elements are specified in the form of a single sentence. These elements require the direct translation or adaptation of words and key phrases as well as the coordination of these elements to produce a coherent sentence. Because the literal or direct combination of words and key phrases can produce awkward expressions, some linguistic refinement may be required to produce a more precise sentence. Fourth, linked elements are specified in the form of multiple sentences. A multiple-sentence linked element could be the entire test item. Because words, key phrases and single sentences have been carefully adapted prior to assembling multiple sentences, only small adjustments should be required for this linked element transformation. However, as with the linked elements at the single sentence level, care must be taken to coordinate these elements so a coherent whole is produced.
Taken together, linked elements specify content in four different forms that provide the translation or adaptation necessary to program IGOR so item generation can occur in multiple languages. Our example is constrained to two languages, but three or more languages can be developed using the same linked element logic to permit simultaneous multilingual item generation. Moreover, IGOR is character set-neutral, meaning that characters from any language can be used to generate test items. Once the four-level linked elements are completed, a multilingual AIG linking map is produced. The map summarizes the necessary links for words, key phrases, single sentences and multiple sentences (for more details, see Gierl et al., in press). Then, IGOR is programmed using the item model content in Figure 21.6 as well as the linking map to produce new items. Using this approach, a total of 2,906 items were generated—1,453 English and 1,453 French items.