From Binary Groundedness to Support Relations: Towards a Reader-Centred Taxonomy for Comprehension of AI Output

ACM CHI 2026 Workshop on Science and Technology for Augmenting Reading (CHI '26 STAR) |

Download BibTex
Left: standard citation-enabled responses from a language model. Right: a hypothetical interface that distinguishes between different types of syntactic manipulation (e.g. direct quote vs. paraphrase) and interpretation (e.g. induction, deduction, deduction subject to assumptions) involved in the production of language model output. We propose the development of a taxonomy of reader-centric \emph{support relations} that would enable such interfaces, thereby leading to better critical engagement of readers with language model output and understanding of how it relates to the sources.

Generative AI tools often answer questions using source documents, e.g., through retrieval augmented generation. Current groundedness and hallucination evaluations largely frame the relationship between an answer and its sources as binary (the answer is either supported or unsupported). However, this obscures both the syntactic moves (e.g., direct quotation vs. paraphrase) and the interpretive moves (e.g., induction vs. deduction) performed when models reformulate evidence into an answer. This limits both benchmarking and user-facing provenance interfaces.

We propose the development of a reader-centred taxonomy of grounding as a set of support relations between generated statements and source documents. We explain how this might be synthesised from prior research in linguistics and philosophy of language, and evaluated through a benchmark and human annotation protocol. Such a framework would enable interfaces that communicate not just whether a claim is grounded, but how.