Wed-SS-1-6-9 Multiscale system for Alzheimer's Dementia Recognition through Spontaneous Speech

Abstract: This paper describes the Verisk submission to The ADReSS Challenge. We analyze the text data at both the word level and phoneme level, which leads to our best-performing system in combination with audio features. Thus, the system is both multi-modal (audio and text) and multi-scale (word and phoneme levels). Experiments with larger neural language models did not result in improvement, given the small amount of text data available. By contrast, the phoneme representation has a vocabulary size of only 66 tokens and could be trained from scratch on the present data. Therefore, we believe this method to be useful in cases of limited text data, as in many medical settings.