Hedges and tag questions shorten LLM answers, paper finds

Prompts that lean on hedges, tag questions, and collective reference elicit "shorter, less sophisticated, and less formal responses across three document types and four models," according to a new arXiv preprint from Katherine Van Koevering and Anjalie Field.

The register effect swamps explicit demographic cues. The paper reports that "explicit gender cues like sign-off names are encoded in the same representational space as linguistic dialect" — but it is the dialect that moves the model, not the name attached to it. Two of the researchers we track posted the paper the week it went up.

Mitigation looks hard. The features sit "in early transformer layers and entangled with other features," the authors write, and they warn that "post-hoc mitigation is challenging."

The framing is workplace, not lab curiosity. Van Koevering and Field close by calling for "upstream consideration of the influences of linguistic variation to mitigate disparate impacts of LLM-mediated workplace communication."