Balisage Paper: Declarative markup in the time of “AI” Controlling the semantics of tokenized strings

Can large language models, working with tokenized text, effectively compare strings and accurately model the comparison in structured forms like alignment tables and the TEI critical apparatus? The answer is no, or not very well, but that answer helps to illuminate a remarkable strength and versatility of declarative markup, when deliberately introduced to imperative programs. We can see how declarative methods control the imperative machinery of a complex collation process in the Frankenstein Variorum project. The key question or provocation is whether “artificial intelligence” in our time can be improved by returning to its declarative origins, to make room for the authority of domain experts to control an otherwise unwieldy stochastic process that poses as a voice of reason.