A GADDAG is a data structure presented by Steven Gordon in 1994, for use in generating moves for Scrabble and other word-generation games where such moves require words that "hook into" existing words.
It is often in contrast to move-generation algorithms using a directed acyclic word graph (DAWG) such as the one used by Maven.
It is generally twice as fast as the traditional DAWG algorithms, but take about 5 times as much space for regulation Scrabble dictionaries.
This means every word has as many representations as it does letters; since the average word in most Scrabble regulation dictionaries is 5 letters long, this makes the GADDAG about 5 times as big as a simple DAWG.
For any word in a dictionary that is formed by a non-empty prefix x and a suffix y, a GADDAG contains a direct, deterministic path for any string REV(x)+y, where + is a concatenation operator.