About the Project

Neologinator is a generator of made-up words (i.e., pseudowords). Pseudowords are generated using n-gram markov models, which have been trained on a dataset of words represented as grapheme-phoneme alignments. That is, the words are generated based on the probability of one grapheme-phoneme pair at a time, given the previous n grapheme-phoneme pairs. For example, given an initial grapheme-phoneme pair of c_k, a bigram model (n = 2) will generate the next grapheme-phoneme pair based on the probability of it occurring after c_k in the training dataset. The words generated by the n-gram model are checked against a dictionary of English words, as well as a list of English proper nouns, to ensure the generated word is not a real word. As these look-ups are not exhaustive, it is possible that some real words may still be generated.

Technical Details

This web application was created using Flask and basic CSS and HTML for the front-end. It is hosted on Render using Docker.

Prototype

The front-end was designed on Figma before being implemented using vanilla HTML, CSS, and JavaScript.

Demo