Are distributional representations ready for the real world? Evaluating word vectors for grounded perceptual meaning
This work highlights a limitation in word vectors for real-world applications, motivating grounded language learning approaches.
The paper evaluated whether standard distributional word representations accurately encode perceptual features of concrete concepts using human semantic norm datasets, finding that several representations fail to capture many salient perceptual features and that these deficits correlate with similarity prediction errors.
Distributional word representation methods exploit word co-occurrences to build compact vector encodings of words. While these representations enjoy widespread use in modern natural language processing, it is unclear whether they accurately encode all necessary facets of conceptual meaning. In this paper, we evaluate how well these representations can predict perceptual and conceptual features of concrete concepts, drawing on two semantic norm datasets sourced from human participants. We find that several standard word representations fail to encode many salient perceptual features of concepts, and show that these deficits correlate with word-word similarity prediction errors. Our analyses provide motivation for grounded and embodied language learning approaches, which may help to remedy these deficits.