Has anyone else created a multi-layered lexicon with a curated corpus, training on a six layer bidirectional GRU? I could really use some help if possible.