AI TrackerFrontier desk · AEST
Back to news
normalaitech

LitArc-Bench: Disentangling Text-Specific Reader-Response Dynamics from Position and Lexical Affect

arXiv cs.CL · 14 Nov 2026, 02:55 AEST

LitArc-Bench is introduced to analyze text-specific reader-response dynamics using Chinese-language literary excerpts. The study reveals that while position alone is insufficient for alignment, various model runs demonstrate significant advantages in understanding human affect trajectories.

Key facts

Read aloud

LitArc-Bench is introduced to analyze text-specific reader-response dynamics using Chinese-language literary excerpts. The study reveals that while position alone is insufficient for alignment, various model runs demonstrate significant advantages in understanding human affect trajectories. We introduce LitArc-Bench to separate these explanations: thirty nine Chinese-language literary excerpts segmented into seven checkpoints, with four hundred thirty questionnaires from eighty one adults in two classroom groups nine thousand twenty VAD ratings. Position alone reaches r =.four hundred seventy for arousal, making raw alignment insufficient. Cross-fitted models match the correct human text better than thirty eight mismatched texts in all nine model-dimension cells p <.one; a text-invariant negative control returns chance p = one. LitArc-Bench identifies which components of literary reader response are recoverable beyond shallow explanations and why shape, calibration, and human variability require separate evaluation.

Sources in this cluster