A tensor-competition based architecture: to capture the influence of word sense

Studies done in the field of word meaning, and in particular the Latent Semantic Analysis (LSA) project, found out the representation of words as vectors in a high dimensional semantic space. However, the system completely neglects the effect of word order on this representation. This results in a semantically viable model that neglects the differences in sentence structure. If the sense a word implies in a sentence is different, as with "the letter is sealed" versus "the letter is N" then LSA assumes them to still be similar. The following model takes LSA data and processes them in tensor space to further incorporate the aspect of word sense. The architecture is amazingly robust and is sensitive to how a word is used in different settings. Ten rules are tested to show how it can add to the representation of meaning offered by LSA.