NarraSpace ties picture-description speech to dementia scores
Original title: NarraSpace: Multimodal Scene-Grounded Evidence for Picture-Description Dementia Classification
On the 48-participant ADReSS test set, a fixed SVM reached 80.88% mean accuracy over fifty seeds.
NarraSpace links words and speech times in picture descriptions to named scene elements and retrieves utterances for predefined events. A frozen language model contrasts event text from Alzheimer's disease and healthy-control examples, with transcript surprisal and coverage measures. On the official Alzheimer's Dementia Recognition through Spontaneous Speech test set, averaged probabilities reached 81.25% accuracy and 91.32% area under the ROC curve. Source records connect scores to utterances, timing and lexical matches for review.
- Mean accuracy was 80.88% plus or minus 2.56% over fifty seeds.
- Narrative-spatial descriptors summarize coverage and temporal organization.
- The paper says the measurements could support clinician review of original speech.
From the original article
1 min read at arxiv.orgPicture-description dementia classification uses a shared visual scene, but a transcript-level prediction alone does not identify the scene evidence underlying its inputs. NarraSpace links words and speech times to named picture elements and retrieves utterances for predefined scene events. Narrative-spatial descriptors summarize coverage and temporal organization.
By Qingming Li · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.