AI Paper Atlas

Research map / Large language models / Alignment and preferences

Align to Structure: Aligning Large Language Models with Structural Information

Zae Myung Kim, Anand Ramachandran, Farideh Tavazoee, Joo-Kyung Kim, Oleg Rokhlenko, Dongyeop Kang

AAAI 2026 · Large language models › Alignment and preferences · approach: Steering and correcting LLM behavior · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
align with graph-level discourse reward (authorship classifier on discourse motifs)Qwen2-1.5B-Instruct · GovReportROUGE-1: 53.21 -> 55.86Improved
align with surface-level text structure rewardQwen2-1.5B-Instruct · GovReportROUGE-1: 53.21 -> 55.45Improved
align with off-the-shelf OpenAssistant RLHF reward modelQwen2-1.5B-Instruct · GovReportROUGE-1: 53.21 -> 53.25No clear effect
align with graph-level discourse rewardQwen2-1.5B-Instruct · essay quality, GPT-4o pairwisewin rateImproved
align with surface-level rewardQwen2-1.5B-Instruct · essay quality, GPT-4o pairwisewin rateImproved
apply second-stage alignment with the other structural reward (two-step) (ablation)SA_S / SA_G · essay quality, GPT-4o pairwisewin rateNo clear effect
use dense discourse-motif token rewards with graph reward vs standard PPO (ablation)Qwen2-1.5B-Instruct · human-distinctive motif proportion in training batchesmotif proportionImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy