Select a kernel to display the spelling tree for that kernel. The tree represents just the section of word that matches the two letter element. The root node is the first letter of the two letter element. Then, recursively, if the next letter in the word matches the first letter of the two letter pair, then it branches left, represented by a lighter gray edge, and if it matches the second letter of the pair then it branches right, represented by a darker gray edge. This branching continues until the word is finished, and ends at some node. The first few nodes are highlighted with the letter corresponding to that point of the tree. The edge weights are logarithmically related to the number of tokens flowing through them. These trees are trimmed by only including words that have a token count of at least the fourth root of the total token count for the stretched tokens. See Sec. IIID in the paper for more information.