Given a tax term or a citation, this shows a list of terms or citations that are similar to it, i.e. nearby in the vector space. In addition to terms like gross income or binding commitment test, this takes a variety of citations, including 348 US 426, 61, 1.1001-3, Rev Rul 99-5, Rev Proc 77-37. But it does not handle subsections (so 704(c) will confuse it).
This demo computes word analogy: the first word is to the second word like the third word is to which word? Try for example corporation, section 351, and partnership. Or, try sham transaction, 364 US 361, and step transaction doctrine. As with any artificial intelligence tool, it sometimes gives nonsensical results.
This demo uses 500-dimensional collocation embeddings induced using the word2vec method, trained on 226M words of U.S. federal tax law, some from Harvard's Caselaw Access Project Federal cases corpus and others from the corpus of PLRs and Tax Court Memorandum decisions available here. The vectors themselves are available under TaxVectors here. Please email Andrew Blair-Stanek with questions or comments.