Language

Currently AI research on Language is following the same path as Vision. We are extracting handcrafted features to understand language. This is incorrect and as inaccurate as using CV algorithms, compared to using CNNs for feature extraction.

Current Languge methods include: NER, POS tagging, dependency parsing, word vectors and more.

We need a neural network in the form of a hypernetwork, to convert visual and audio information into a formal human language: as text. This text is then connected to its syntax and semantics, weighted by its usage. We have to follow the cognitive approach of a development of a language.