Awake fMRI Reveals Mechanisms of Language Comprehension in Dogs

Abstract

To what extent do dogs understand human language? At a basic level, lexical processing would require the differentiation of words from non-words, while semantic processing would further require the recognition of word meaning. To determine the level of language processing, we trained 12 dogs to retrieve two objects based on object names, then probed the neural basis for these auditory representations using awake-fMRI. As a control, we compared the neural response to pseudowords versus trained words, and novel objects versus trained objects during the fMRI scan. If dogs lexically processed the trained words, then a differential auditory response would be predicted to occur to trained words relative to pseudowords. Moreover, if dogs used semantic processing, there should be a difference in activation between the two trained words. In support of lexical processing, we found greater activation for pseudowords relative to trained words bilaterally in the parietotemporal cortex, but no activation difference between the two trained words. However, multivoxel pattern analysis (MVPA) revealed statistically significant clusters of informative voxels for the two trained words in the thalamus, left temporoparietal region, and left caudate nucleus. These results suggest a different mechanism of language comprehension than in humans. The dogs’ greater activation for pseudowords in the parietotemporal cortex indicates an underlying bias for novelty rather than semantic processing of words as object-referents, while the MVPA results suggest a potential link between words/sounds and motor action.