Although many systems for assisting individuals with aphasia have introduced photos as a supplement to their icon-based vocabulary, they only provide a limited set of additional images and, thus, the key limitations with using icons as the core vocabulary remain. Furthermore, the images employed in the existing assistive devices are mostly photographs taken by the system users, which, similar to having artists design icons, has only shifted the burden of generating the needed language representations from the icon designers to the system users. One alternative to breaking this “effort” bottleneck is to utilize the abundance and diversity of internet images. In this case, images still need to be found, but this is faster than image generation. In addition, the abundance of web images available makes selecting the most appropriate image easier.
Both anecdotal reports from speech-language therapists plus systematic investigation suggest that, unlike written words, cultural differences do not prevent people from identifying images (photographs) and icons (line drawings) in a similar way, although different cultures may resolve ambiguities differently. This suggests that no training is required for understanding images; in fact, experiments have found that infants are able to understand images without training.
Despite the increased use of images for communication, no research has focused on how the irregular quality and inconsistent information complexity can influence the perception of an arbitrary selection of images from the web, in contrast, to carefully and deliberately designed icons. In order to provide a basis for using web images instead of icons as the principal visual representation in assistive technologies for people with aphasia, we first need to conduct a set of studies illustrating that the web images are, on the whole, as effective as icons for communication.
Thus, in this work, we have directly compared the effectiveness of images and icons for communication with our target population, 50 aphasic individuals. This large subject pool is uncommon in research on assistive technologies because it is much more time-consuming to run studies using language deficient subjects and also more difficult to obtain subjects.
Nevertheless, it is important to run comparisons such as these with a larger number of subjects for two reasons. First, our aim is to investigate if images are equal to icons in efficacy. As such, we need to look at the power of the test when conducting statistical inferences that show no difference. This approach demands a larger N. In addition, because of the wide variability in the impact of the traumatic event that has caused aphasia and also in the subjects themselves who maybe affected cognitively in other ways, a larger N is needed to handle this variance.
Unlike earlier studies, the vocabulary we examined was not restricted to concrete nouns but extended to illustrate different parts of speech, including concrete and abstract nouns, verbs, and adjectives. Moreover, our source of images is relatively random. It is compiled from the Internet and not painstakingly annotated. The hope is that this work will demonstrate that this cheap and plentiful source of images is “good enough” to use in communication systems developed for aphasics.