CS Paper Radar

Why Summaries Turn Neutral: Policy Attribution for Sentiment Drift in Reinforcement Learning from Human Feedback

Mikhail Krasitskii, Alexander Gelbukh, Olga Kolesnikova, Grigori Sidorov

cs.CL ·submitted 2026-08-16 04:56 UTC · v1

Abstract

Reinforcement learning with human feedback (RLHF) aligns LLMs with human preferences, improving summarization fluency and safety, but causes sentiment drift: overly neutral summaries stripped of emotional nuance. We diagnose why RL acts as a sentiment neutralizer and present Policy Attribution, a framework using gradient and logit decomposition to trace drift to reward model (RM) signals and KL (Kullback-Leibler) penalty. Sentiment drift reflects a strategic bias toward "low-risk" tokens maximizing expected rewards under preference uncertainty (Stiennon et al., 2020; Gao, Schulman, and Hilton, 2023). On Reddit TL;DR and CNN/DailyMail, RLHF summaries get higher rewards but show 30-40% lower sentiment variance. Cross-lingual analysis across eight languages shows language-independent drift, with morphologically richer languages more suppressed (Krasitskii et al., 2026). We propose and validate a sentiment-aware regularization technique reducing drift by 18-22% without harming summary quality. The code and toolkit will be public.

Closest papers

Nearest neighbours by meaning across the whole index.

Whose Voice Survives the Summary? A Voice-Retention Audit of LLM Employee Listening

cs.AI cs.CLcs.CYcs.HC ·2026-09-30 ·Thilo Tamme, Anton Hantel, Bijan Khosrawi-Rad
published similarity 0.63
Abstract · 10 pages, 3 figures, 3 tables. Accepted at the 60th Hawaii International Conference on System Sciences (HICSS 2027)

Organizations increasingly route employee feedback to leaders through large language model (LLM) summaries, an unaudited layer that silences already-spoken voice. We introduce a Voice Retention / Representation Ratio metric for representational bias in summarization and apply it to a bilingual (English/German) corpus of 2,586 free-text responses from a global professional service company. First, employees supply criticism more reliably than praise (withholding praise is 82 times more common). Second, across 45 leader-summaries the pipeline filters by popularity, not sentiment: criticism survives, yet a concern voiced once is dropped 86% of the time, with short and German-only content lost on the same axis (theme retention 0.14 vs 0.74; German directional). Controlling for frequency, sentiment has no independent effect; the harm is prevalence-driven, which sentiment-only audits miss. A targeted prompt recovers only named themes. We contribute the metric, field evidence, and a disaggregated voice-retention card.

0.31

RMB: Reward Model Boosting Mitigates Reward Hacking

cs.LG ·2026-09-27 ·Jiabin Fan, Dezhi Ye, Yongchang Hao, Lili Mou
published similarity 0.63
Abstract · Published in Transactions on Machine Learning Research (TMLR), September 2026

Reinforcement Learning from Human Feedback (RLHF) is a powerful technique for aligning large language models (LLMs) with human preference. However, it often suffers from the reward hacking issue, where policy optimization improves the proxy reward model while actually degrading performance with respect to the true human preference, due to the imperfection of the proxy. To address this, we propose Reward Model Boosting (RMB), a novel approach that enhances the robustness and reliability of the reward signal for RLHF. RMB first trains a set of reward models with a diversity-promoting regularizer. This encourages each model to learn complementary aspects of the reward landscape. Then, RMB learns a lightweight aggregator in the principle of boosting to aggregate the outputs of the diverse reward models into a more accurate and robust reward signal. Our extensive experiments demonstrate that RMB significantly improves reward accuracy on both in-distribution and out-of-distribution datasets, substantially mitigating the reward hacking issue and ultimately improving RLHF performance.

0.31

Concept Direction Reliability Across Languages with Different Tokenizer Fertility

cs.CL ·2026-09-28 ·Muhammad Abdullahi Said, Abass Oguntade, Elisha Komolafe, Babangida Sani, Fatima Muhammad Adam +1 more
similarity 0.63
Abstract

Extracted sentiment directions can vary across samples even when downstream sentiment classification remains accurate. To evaluate direction reproducibility, we measure split-half agreement in English, Hausa, and Yoruba representations across four language models using both native and translated texts. We identify layers selected for agreement using ten topics and evaluate direction agreement across separate groups of fifteen topics. Using the final token, split-half agreement ranges from 0.737 to 0.870 for English, 0.589 to 0.762 for Hausa, and 0.101 to 0.399 for Yoruba, maintaining this language rank order across all 77 complete model comparisons. Classifiers trained on these same layers consistently predict sentiment above chance, demonstrating that predictive accuracy does not imply directional consistency. Furthermore, averaging token representations yields less consistent agreement, and high agreement can partially reflect sentence length. Ultimately, our findings highlight the need to measure vector direction reproducibility independently of classification performance, though they do not establish that tokenizer fertility which is the average number of tokens per whitespace separated word causes cross-lingual differences.

0.17

TIAO: Token Importance-Aware Policy Optimization for Text Summarization

cs.CL ·2026-09-15 ·Qixiu Li, Chenlong Bao, Xiang Zhu, Xiaoyong Li, Ruixin Cao +2 more
code/data similarity 0.62
Abstract

Text summarization requires models to condense content while preserving key qualities such as consistency and coherence. Large language models (LLMs) have shown strong performance on this task and can be further improved through reinforcement learning (RL). However, most existing methods apply reward signals directly to undifferentiated token sequences, overlooking the varying importance of individual tokens to word and sentence level quality in summarization. In this paper, we propose Token Importance-Aware Policy Optimization (TIAO), a novel reinforcement learning strategy that explicitly leverages token-importance awareness. Specifically, TIAO identifies core tokens based on token dependency and reweights a trajectory's advantage according to its overall dependencies. Experiments on the real world dataset show that our TIAO achieves highly competitive results, and that a 7B foundation model enhanced by TIAO performs comparably to GPT-4 and GPT-5-nano. Code is available at https://github.com/TechCloud-x/TIAO

0.31

Usage-Modulated Sentiment Representations in Large Language Models

cs.CL ·2026-10-04 ·Hongfei Du, Jiacheng Shi, Yanfu Zhang, Gang Zhou, Ye Gao
top venuedataset similarity 0.61
Abstract · Accepted to EMNLP 2026 (main conference). 18 pages, 3 figures, including appendices

Prior work suggests that sentiment can often be captured by approximately linear directions in LLM activation spaces, but a single direction may not fully capture sentiment representations. In natural communication, sentiment is shaped not only by polarity but also by usage factors, such as tone and audience adaptation. We test whether these factors systematically modulate sentiment representations beyond a shared sentiment direction. We construct a controlled paired dataset that holds event content fixed while varying sentiment polarity and usage factors, and analyze Llama, Mistral, and Gemma. We identify a shared sentiment direction, remove it, and test the residual structure through erasure and generation-time tone steering. Across models, the shared direction is robust (median cosine 0.953-0.975), yet removing it leaves 0.833-0.909 of the original positive-negative representation-difference norm. The residuals contain compact, reproducible usage-conditioned structure. Targeted erasure weakens held-out usage metrics more than random and label-shuffled controls. On Llama, outputs steered along residualized tone components are preferred in 92.8% of blind target-tone comparisons while preserving the requested sentiment polarity in 98.7% of evaluated outputs.

0.69

Steering with Multilingual Lexicons for Sentiment Control in Low-Resource Languages

cs.CL ·2026-11-02 ·Lucinda Laurent, Muhammad Dehan Al Kautsar, Alfan Farizki Wicaksono, Fajri Koto
published similarity 0.61
Abstract · 30 pages, 12 figures, 11 tables

Activation steering offers a lightweight mechanism for controlling model behavior, but existing methods depend on sentence-level contrastive data that is unavailable for most of the world's languages. We show that multilingual sentiment lexicons are a viable substitute across three model families: aggregated multilingual lexicons outperform English-derived steering vectors for mid- and low-resource languages, while monolingual lexicons remain stronger for languages a model has specialized in during pre-training. The steering efficacy is mechanistically dictated by both representational density and pre-training priors. Large-magnitude English vectors overwhelm the compact representations of low-resource languages, causing "linguistic collapse." This fragility is mitigated by the norm-dampening of multilingual aggregation. Architectural biases, including bilingual specialization penalty and latent numerical rigidity, also heavily constrain cross-lingual steering responsiveness. Ultimately, our work establishes lexicons as a practical steering signal and offers a comprehensive mechanistic account of cross-lingual steering dynamics.

0.32

EasyPPO: Stabilizing the Critic Is Key

cs.LG ·2026-09-29 ·Xuanyi Zhou, Qiuyang Mang, Huanzhi Mao, Dacheng Li, Wenhao Chai +5 more
similarity 0.60
Abstract

A key strength of Proximal Policy Optimization (PPO) is its learned critic, which uses historical trajectories collected during reinforcement learning to estimate expected returns and reduce policy-gradient variance. However, we find that the critic is also a major source of instability in reinforcement learning for large language models (LLMs). We identify two critic failure modes that destabilize PPO. First, filtering truncated rollouts from both actor and critic shifts the policy objective to reward conditioned on completion, allowing truncation to increase even as conditional reward improves. Second, heterogeneous return noise can cause high-variance prompts to dominate critic updates in finite batches. We introduce EasyPPO to address these failures. Actor-only overlong filtering trains the critic on returns from both completed and truncated rollouts. Noise-normalized critic regression weights each prompt's critic loss by the inverse standard deviation of its sampled returns, balancing noise contributions across prompts. Moderately smaller critic mini-batches confine outlier influence to fewer rollouts during gradient clipping. Across continuous-reward coding on FrontierCS, binary-reward mathematical reasoning on AIME24, and multi-turn search on Search-R1, EasyPPO remains stable throughout the full training horizon and consistently outperforms vanilla PPO, VAPO, and HL-Gauss PPO. Its best validation scores show relative gains of 14.89%, 2.28%, and 9.47% over PPO, respectively.

0.21

Trust the Critic More

cs.LG cs.AI ·2026-09-30 ·Kaiyue Wen, Luke Bailey, Arvind Mahankali, Tengyu Ma
similarity 0.59
Abstract

Standard language model RL algorithms credit every token of a long rollout with the same advantage determined by the terminal reward. Actor-critic methods can provide finer-grained credit assignment, but learned critics are generally considered too inaccurate to trust when training LLMs with RL. In recent works, even when a critic is present, it is used only for baseline estimation, so every trajectory must be rolled out to its terminal reward. We introduce Actor-Critic with Action Chunking (AC2) that removes the need to roll every trajectory to completion. AC2 instead assigns credit to action chunks: short continuations of prefixes of past trajectories. A learned critic scores the state reached at the end of each action chunk, allowing the policy to update without observing a terminal reward. We make critic-based credit assignment reliable through three design choices. First, we introduce local readiness which uses critic-based updates on a problem only when the critic is sufficiently accurate on that particular problem. Second, when available, we provide the critic with a reference solution from a previous successful rollout. Third, we assign credit over action chunks of 10k tokens rather than individual tokens, giving the critic a more meaningful portion of the trajectory to evaluate. We train Qwen3-4B on FineProofs-RL using AC2 and evaluate on IMO-ProofBench. AC2 exceeds GRPO's peak validation score of 18.5% using 2.5x fewer decoding FLOPs. This gain comes from two sources, (1) AC2 requires 25% fewer training steps to reach this score, and (2) each step generates fewer tokens because the policy does not need to continue every trajectory to completion. Conceptually, we demonstrate that we can remove the need to roll out every trajectory to completion, opening up a large previously unexplored design space for LLM RL algorithms.

0.43