September 2020
·
45 Reads
Artificial Intelligence (AI) applications are usually built on large trained data models that can recognize and label images, provide speech output from text, process natural language for translation, and be of assistance to many individuals via the internet. For those who are non-verbal or have complex speech and language difficulties, AI has the potential to offer enhanced access to the wider world of communication that can be personalized to suit user needs. Examples include pictographic symbols to augment or provide an alternative to spoken language. However, when using AI models, data related to the use of freely available symbol sets is scarce. Moreover, the manipulation of the data available is difficult with limited annotation, making semantic and syntactic predictions and classification a challenge in multilingual situations. Harmonization between symbol sets has been hard to achieve; this paper aims to illustrate how AI can be used to improve the situation. The goal is to provide an improved automated mapping system between various symbol sets, with the potential to enhance access to more culturally sensitive multilingual symbols. Ultimately, it is hoped that the results can be used for better context sensitive symbol to text or text to symbol translations for speech generating devices and web content.