This application searches for and scores clusters of words shared across two groups of texts, allowing you to look for quotations, paraphrases, or shared topics. When configured correctly, Tangram can also find idioms and collocations. Each input file, which may come in a variety of formats (TAN, TEI, other XML formats, plain text, Word documents) must be assigned to one or both of two groups, each group representing a work. Members of a work-group can be from different languages. Users can specify how many ngrams ("words") should be found, and how far apart they can be from each other. Ngram order is disregarded (e.g., ngram "shear", "blue", "sheep" would match ngram "sheep", "blue", "shear"). Tangram first normalizes and tokenizes each text according to language rules. Each token is converted to one or more aliases. If lexico-morphological data is available through a TAN-A-lm file, or if there is a TAN-A-lm language library for the language of the text being processed, a token may be replaced by multiple lexemes (e.g., "rung" would attract aliases "ring" and "rung"); otherwise, a case-insensitive generic form of the word is used. Then each text in group 1 is compared to each text in group 2 that shares the same language. For each pair of texts, Tangram identifies clusters of tokens that share the same alias. It then consolidates adjacent clusters of ngrams, and scores the results based upon several criteria. Grouped clusters are then converted into a primitive TAN-A file consisting of claims that identify parallel passages of each pair of texts, and the output is rendered as sortable HTML, to facilitate better study of the results. Tangram was written primarily to support quotation detection in ancient Greek and Latin texts, which has rather demanding requirements. Because of these objectives, Tangram frequently operates in quadratic or cubic time, so can be quite time-consuming to run. A feature allows the user to save intermediate stages as temporary files, to reduce processing time. any XML file, including this one (input is ignored) one or more files allocated to two groups; perhaps temporary files; perhaps TAN-A-lm files, either associated with secondary input, or part of a language catalog perhaps diagnostics (1) an XML file with TAN-A claims identifying quotations or parallels, with the most likely at the top; (2) an HTML file that renders #1 in a more legible format. Support the method pioneered by Shmidman, Koppel, and Porat: https://arxiv.org/abs/1602.08715v2 Make sure texts run against themselves work. Incorporate simpler tablesorter javascript Edited, prepared for TAN 2021 release.