This model seeks to answer the question "What if we trained a Word2Vec model to map semantic relationships between words then drove it insane by adding James Joyce's Finnegan's Wake to its training corpus?"
The original objective in training this model was to see if it was possible to find semantic relationships between Joyce's neologisms and common english words. However the low frequency of the neologisms with only one or two examples of each in Joyce's work effectively precludes building meanigful embeddings for them. Nevertheless it remains a fun experiment. A Word2Vec model was trained on a 670MB subset of the Gutenberg text corpus augmented with a copy of Finnegan's Wake using the Gensim library. Words are represented as 250-dimension embedding vectors. Semantic relationships between words can be determined by examining the proximity of words in embedding space.
