Comparing Sentence-Based and Word-Based Semantic Space Representations to Brain Responses

EasyChair Preprint no. 3877

Abstract

Computational semantic space models have now been applied to sentences, but it is unclear whether they capture how the human brain represents sentences. Using fMRI we scanned adult readers reading expository texts and compared their brain responses to 3 semantic space vectors that modeled sentences either as combinations of words or as single units. We observe that computational semantic representations that are specifically designed to capture sentence content share information content with brain responses.

Keyphrases: fMRI, RSA, Sentence embeddings, sentence processing

@Booklet{EasyChair:3877,
  author = {Friederike Seyfried and Ping Li},
  title = {Comparing Sentence-Based and Word-Based Semantic Space Representations to Brain Responses},
  howpublished = {EasyChair Preprint no. 3877},

  year = {EasyChair, 2020}}