Last week, Aaron Earle-Richardson (Research Tech) presented Christian Herff’s new conference paper on semantic representations of speech production. This 🧵 explores our thoughts (🤍 & ❔)
Thank you to the authors at Maastricht University for your work, and we look forward to following its progress!
❔3️⃣: Are semantically correlated channels spatially distributed within patient? If not, is it possible that semantic processing is highly localized, but the locality varies widely from patient to patient?
❔2️⃣: Given that semantic representations as a whole were widely distributed, is it possible to correlate elements of the embeddings vector with specific brain areas? Would that then support the spatial organization of semantic processing (such as in work by Huth et al.)?
❔1️⃣: How do you know that ‘production’ responses aren’t just auditory responses to the participants hearing themselves speak?
🤍3️⃣: The focus on semantic processing and public availability of the data allows other neuroprosthetics researchers to take advantage of your findings.
🤍2️⃣: The wide distribution of semantic encoding channels is interesting because supports previous work that shows that semantic processing is not localized to one area of the brain.
🤍1️⃣: Novel use of word2vec embeddings + regression instead of tokens + classification in a word repetition task allows for fine-tuned modeling and reduces need for trial repetitions required for classification.
ieeexplore.ieee.org/abstract/doc...
Speech neurprostheses have the potential to provide severely paralyzed patients with a means of communication. To enable the best possible decoding of speech processes from neural data, it is importan...