Search papers, labs, and topics across Lattice.
This paper introduces D2C-Routing, a novel approach for detecting mixed-origin AI-generated text by framing the problem as dimension-to-composition source attribution. The method effectively separates content and expression origins, allowing for a nuanced classification of text into four collaboration types. On the MixD2C dataset, the D2C-Routing system achieved a four-way average true positive rate of 0.8603 at 1% false positive rate, significantly outperforming previous methods by 6.5 points.
Distinguishing between AI-generated content and human expression can improve detection accuracy by over 6% in mixed-origin text scenarios.
AI-generated text detection is commonly framed as a binary document-level judgment about whether a text is human-written or machine-generated. This framing breaks down for mixed-origin writing, where content origin and expression origin may differ. We cast mixed-origin detection as dimension-to-composition source attribution, inferring content origin and expression origin before composing them into four collaboration types. We propose Dimension-to-Composition Routing (D2C-Routing), which routes content-side and expression-side evidence to supervised dimension heads before a learned gated composition layer predicts the final label. On MixD2C, a reconstructed split derived from the HART mixed-origin benchmark, our disclosed D2C-Routing-based detector system reaches 0.8603 four-way Avg TPR@1%FPR, 6.5 points above the same-split RACE-local rerun. Core ablations support the routing design, while error analysis shows that distinguishing AI-content/human-expression from fully AI-generated text remains the hardest boundary. Code is available at https://github.com/bystander563/d2c-routing-artifact.
