thattags

Thattags are a form of metadata used to annotate referential elements in text, focusing on demonstratives about which the referent is the preceding discourse. They are intended to aid natural language processing by linking a pronoun or determiner such as "that" to its antecedent, improving coreference tracking, searchability, and discourse interpretation.

Origin and scope: The concept arose in discussions of language-aware data annotation and knowledge extraction. Thattags are not universally standardized; implementations vary. They are designed to be interoperable with existing semantic formats (for example JSON-LD or RDF) so that they can accompany text in digital libraries, corpora, or educational tools.

Structure and example: A thattag record may include an id, sentence index, token position, refersTo (the antecedent's identifier), relation type (such as antecedentOf or referenceTo), and a confidence score. Example: in "The report was released, and that surprised the team," a thattag could link "that" to "The report" with a confidence of 0.92.

Applications: Thattags support coreference resolution, targeted search, summarization, accessibility tooling, and linguistic research. They can be created by humans or generated by NLP pipelines, and stored alongside the text in compatible data stores.

Limitations and standards: Ambiguity in interpretation, cross-linguistic differences, and annotation costs are challenges. Adoption depends on tooling, governance, and interoperability with broader metadata standards to ensure provenance and reuse.