In this work, we establish a novel theoretical connection between supervised fine-tuning and offline reinforcement learning under the token-level Markov decision process, revealing that large language models indeed learn an implicit Q-function for inference.Through this theoretical lens, we demonstrate that the widely used beam search method suffers from unacceptable over-optimism, where inference errors are inevitably amplified due to inflated Q-value estimations of suboptimal steps. To address this limitation, we propose **S**upervised **O**ptimism **C**orrection (SOC), which introduces a simple yet effective auxiliary loss for token-level Q-value estimations during supervised fine-tuning. Specifically, the auxiliary loss employs implicit value regularizationto boost model confidence in expert-demonstrated responses, thereby suppressing over-optimism toward insufficiently supervised responses.Extensive experiments on mathematical reasoning benchmarks, including GSM8K, MATH, and GAOKAO, showcase the superiority of the proposed SOC with beam search across a series of open-source models.

The growing interest in Large Language Models (LLMs) for specialized applications has revealed a significant challenge: when tailored to specific domains, LLMs tend to experience catastrophic forgetting, compromising their general capabilities and leading to a suboptimal user experience. Additionally, crafting a versatile model for multiple domains simultaneously often results in a decline in overall performance due to confusion between domains. In response to these issues, we present the RolE Prompting Guided Multi-Domain Adaptation (REGA) strategy. This novel approach effectively manages multi-domain LLM adaptation through three key components: 1) Self-Distillation constructs and replays general-domain exemplars to alleviate catastrophic forgetting. 2) Role Prompting assigns a central prompt to the general domain and a unique role prompt to each specific domain to minimize inter-domain confusion during training. 3) Role Integration reuses and integrates a small portion of domain-specific data to the general-domain data, which are trained under the guidance of the central prompt. The central prompt is used for a streamlined inference process, removing the necessity to switch prompts for different domains.Empirical results demonstrate that REGA effectively alleviates catastrophic forgetting and inter-domain confusion. This leads to improved domain-specific performance compared to standard fine-tuned models, while still preserving robust general capabilities.

Numerous works are proposed to align large language models (LLMs) with human intents to better fulfill instructions, ensuring they are trustful and helpful.Nevertheless, some human instructions are often malicious or misleading and following them will lead to untruthful and unsafe responses.Previous work rarely focused on understanding how LLMs manage instructions based on counterfactual premises, referred to here as inductive instructions, which may stem from users’ false beliefs or malicious intents.In this paper, we aim to reveal the behaviors of LLMs towards inductive instructions and enhance their truthfulness and helpfulness accordingly. Specifically, we first introduce a benchmark of Inductive Instructions (INDust), where the false knowledge is incorporated into instructions in multiple different styles. After extensive human and automatic evaluations, we uncovered a universal vulnerability among LLMs in processing inductive instructions.Additionally, we identified that different inductive styles affect the models’ ability to identify the same underlying errors,and the complexity of the underlying assumptions also influences the model’s performance.Motivated by these results, we propose Dual-critique prompting to improve LLM robustness against inductive instructions.Our experiments demonstrate that Dual-critique prompting significantly bolsters the robustness of a diverse array of LLMs, even when confronted with varying degrees of inductive instruction complexity and differing inductive styles.

Dialogue models are often enriched with extensive external knowledge to provide informative responses through a retrieval-augmented pipeline. Nevertheless, retrieval-augmented approaches rely on finely annotated retrieval training data and knowledge-grounded response generation data, making it costly to transfer. To tackle this challenge, this paper proposed a retrieval-free approach, KiDG, by automatically turning knowledge documents into simulated multi-turn dialogues through a Multi-Document Traversal algorithm. The simulated knowledge-intensive dialogues constructed by KiDG in one domain can be easily used to train and enhance pre-trained dialogue models’ knowledge w.r.t. this domain without costly annotation. We conduct extensive experiments comparing retrieval-augmented models and a variety of retrieval-free models. We found that dialogue models enhanced with data simulated with KiDG largely outperform state-of-the-art retrieval-free methods, and it achieves comparable performance compared to retrieval-augmented methods while being better, and cheaper at domain transfer.

Continual learning for named entity recognition (CL-NER) aims to enable models to continuously learn new entity types while retaining the ability to recognize previously learned ones. However, the current strategies fall short of effectively addressing the catastrophic forgetting of previously learned entity types. To tackle this issue, we propose the SKD-NER model, an efficient continual learning NER model based on the span-based approach, which innovatively incorporates reinforcement learning strategies to enhance the model’s ability against catastrophic forgetting. Specifically, we leverage knowledge distillation (KD) to retain memory and employ reinforcement learning strategies during the KD process to optimize the soft labeling and distillation losses generated by the teacher model to effectively prevent catastrophic forgetting during continual learning. This approach effectively prevents or mitigates catastrophic forgetting during continuous learning, allowing the model to retain previously learned knowledge while acquiring new knowledge. Our experiments on two benchmark datasets demonstrate that our model significantly improves the performance of the CL-NER task, outperforming state-of-the-art methods.

In this paper, we firstly empirically find that existing models struggle to handle hard mentions due to their insufficient contexts, which consequently limits their overall typing performance. To this end, we propose to exploit sibling mentions for enhancing the mention representations. Specifically, we present two different metrics for sibling selection and employ an attentive graph neural network to aggregate information from sibling mentions. The proposed graph model is scalable in that unseen test mentions are allowed to be added as new nodes for inference. Exhaustive experiments demonstrate the effectiveness of our sibling learning strategy, where our model outperforms ten strong baselines. Moreover, our experiments indeed prove the superiority of sibling mentions in helping clarify the types for hard mentions.

We present MCPG: a simple and effectiveapproach for controllable unsupervised paraphrase generation, which is also flexible toadapt to specific domains without extra training. MCPG is controllable in different levels: local lexicons, global semantics, and universal styles. The unsupervised paradigm ofMCPG combines factual keywords and diversified semantic embeddings as local lexical andglobal semantic constraints. The semantic embeddings are diversified by standard dropout,which is exploited for the first time to increaseinference diversity by us. Moreover, MCPGis qualified with good domain adaptability byadding a transfer vector as a universal style constraint, which is refined from the exemplars retrieved from the corpus of the target domain in atraining-free way. Extensive experiments showthat MCPG outperforms state-of-the-art unsupervised baselines by a margin. Meanwhile,our domain-adapted MCPG also achieves competitive performance with strong supervisedbaselines even without training.

Auxiliary information from multiple sources has been demonstrated to be effective in zero-shot fine-grained entity typing (ZFET). However, there lacks a comprehensive understanding about how to make better use of the existing information sources and how they affect the performance of ZFET. In this paper, we empirically study three kinds of auxiliary information: context consistency, type hierarchy and background knowledge (e.g., prototypes and descriptions) of types, and propose a multi-source fusion model (MSF) targeting these sources. The performance obtains up to 11.42% and 22.84% absolute gains over state-of-the-art baselines on BBN and Wiki respectively with regard to macro F1 scores. More importantly, we further discuss the characteristics, merits and demerits of each information source and provide an intuitive understanding of the complementarity among them.

针对在金融领域实体级情感分析任务中,往往缺乏足够的标注语料,以及通用的情感分析模型难以有效处理金融文本等问题。本文构建一个百万级别的金融领域实体情感分析语料库,并标注五千余个金融领域情感词作为金融领域情感词典。同时,基于该金融领域数据集,提出一种结合金融领域情感词典和注意力机制的金融文本细粒度情感分析模型。该模型使用两个LSTM网络分别提取词级别的语义信息和基于情感词典分类后的词类级别信息,能有效获取金融领域词语的特征信息。此外,为了让文本中金融领域情感词获得更多关注,提出一种基于金融领域情感词典的注意力机制来为不同实体获取重要的情感信息。最终在构建的金融领域实体级语料库上进行实验,取得了比对比模型更好的效果。