Reward models (RMs) are crucial for aligning large language models (LLMs) with human preferences. However, most RM research is centered on English and relies heavily on synthetic resources, which leads to limited and less reliable datasets and benchmarks for Chinese. To address this gap, we introduce CheemsBench, a fully human-annotated RM evaluation benchmark within Chinese contexts, and CheemsPreference, a large-scale and diverse preference dataset annotated through human-machine collaboration to support Chinese RM training. We systematically evaluate open-source discriminative and generative RMs on CheemsBench and observe significant limitations in their ability to capture human preferences in Chinese scenarios. Additionally, based on CheemsPreference, we construct an RM that achieves state-of-the-art performance on CheemsBench, demonstrating the necessity of human supervision in RM training. Our findings reveal that scaled AI-generated data struggles to fully capture human preferences, emphasizing the importance of high-quality human supervision in RM development.

Self-critic has become a crucial mechanism for enhancing the reasoning performance of LLMs. However, current approaches mainly involve basic prompts for intuitive instance-level feedback, which resembles System-1 processes and limits the reasoning capabilities. Moreover, there is a lack of in-depth investigations into the relationship between LLM’s ability to criticize and its task-solving performance. To address these issues, we propose Critic-CoT, a novel framework that pushes LLMs toward System-2-like critic capability. Through a step-wise CoT reasoning paradigm and the automatic construction of weak-supervision data without human annotation, Critic-CoT enables LLMs to engage in slow, analytic self-critique and refinement, thereby improving their reasoning abilities. Experiments on GSM8K and MATH and out-of-domain evaluation demonstrate that our enhanced model significantly boosts task-solving performance by filtering out invalid solutions or iterative refinement. Furthermore, we investigate the intrinsic correlation between critique and task-solving abilities within LLMs, discovering that these abilities can mutually reinforce each other rather than conflict.

Hallucination occurs when large language models exhibit behavior that deviates from the boundaries of their knowledge during response generation. To address this critical issue, previous learning-based methods attempt to finetune models but are limited by off-policy sampling and coarse-grained feedback. In this paper, we present Reinforcement Learning for Hallucination (RLFH), an on-policy self-alignment approach that enables LLMs to actively explore their knowledge boundaries and self-correct generation behavior through fine-grained feedback signals. RLFH introduces a self-assessment framework where the policy serves as its own judge. Through this framework, responses are automatically decomposed into atomic facts and their truthfulness and informativeness are assessed against external knowledge sources. The resulting fine-grained feedback at the statement level are then converted into token-level dense reward signals. This enables online reinforcement learning to achieve precise and timely optimization without human intervention. Comprehensive evaluations on HotpotQA, SQuADv2, and Biography benchmarks validate RLFH’s effectiveness in hallucination mitigation.

Annotation conversion is an effective way to construct datasets under new annotation guidelines based on existing datasets with little human labour. Previous work has been limited in conversion between tree-structured datasets and mainly focused on feature-based models which are not easily applicable to new conversions. In this paper, we propose two simple and effective graph-to-graph annotation conversion approaches, namely Label Switching and Graph2Graph Linear Transformation, which use pseudo data and inherit parameters to guide graph conversions respectively. These methods are able to deal with conversion between graph-structured annotations and require no manually designed features. To verify their effectiveness, we manually construct a graph-structured parallel annotated dataset and evaluate the proposed approaches on it as well as other existing parallel annotated datasets. Experimental results show that the proposed approaches outperform strong baselines with higher conversion score. To further validate the quality of converted graphs, we utilize them to train the target parser and find graphs generated by our approaches lead to higher parsing score than those generated by the baselines.

This paper describes our submission system (HIT-SCIR) for the CoNLL 2020 shared task: Cross-Framework and Cross-Lingual Meaning Representation Parsing. The task includes five frameworks for graph-based meaning representations, i.e., UCCA, EDS, PTG, AMR, and DRG. Our solution consists of two sub-systems: transition-based parser for Flavor (1) frameworks (UCCA, EDS, PTG) and iterative inference parser for Flavor (2) frameworks (DRG, AMR). In the final evaluation, our system is ranked 3rd among the seven team both in Cross-Framework Track and Cross-Lingual Track, with the macro-averaged MRP F1 score of 0.81/0.69.