Speculative decoding accelerates autoregressive LLM generation with a lightweight draft model, yet existing distillation methods treat all historical tokens equally despite their divergent contributions to next-token prediction. We discover that attention in draft models is extremely concentrated: the top-3 historical tokens capture 80% of attention weight, while over 90% of tokens contribute negligibly. This uniform training paradigm forces the draft model to learn from noisy, low-importance tokens, diluting the learning signal from informative positions and limiting the alignment between draft and target distributions. To address this, we propose DarterSD, a dynamic token-weighted distillation framework for efficient speculative decoding. DarterSD employs a learnable sigmoid gate in self-attention to compute per-token importance scores, and designs a multi-objective distillation loss comprising distribution alignment, logit matching, and gate regularization to focus training on high-informative tokens. We further introduce a lightweight Boolean-mask denoising strategy at inference to suppress contextual noise, without modifying the target model’s verification pipeline. Experiments show DarterSD delivers a lossless speedup up to 5.52x , outperforming EAGLE3 (4.67x), and boosts the average accepted token length by 12.9%.

The rapid spread of misinformation has made fact verification a critical necessity. LLM-based methods have become the mainstream approach for fact verification, but they still suffer from hallucinations. To mitigate hallucinations, researchers incorporate KG on LLM reasoning with a graph-based constrained decoding. However, these methods with constrained decoding suffer from error cascading due to the inherent knowledge gaps of the LLM, which cause it to select wrong branches during beam search in the constrained token space. To tackle error cascading, we think that the decoding process should be dynamically guided by step-level plausibility assessment. In light of this, we propose TrustFV (https://anonymous.4open.science/r/Trust_FV-6239), a fact verification framework that step-wise beam search in token-level constrained decoding with risk-averse pruning for trustworthy verification. Specifically, TrustFV integrates (1) token-level constrained decoding on a Knowledge Graph Tries (KG-Tries) to ground generation in retrieved evidence from KG, (2) an Uncertainty-aware Process Reward Model (UPRM) trained via Monte Carlo rollouts and ordinal regression to estimate both expected success and uncertainty per step, and (3) a step-wise beam search that actively prunes high-risk branches via the step-level signal of UPRM, producing faithful and plausible reasoning paths. Experiments show that TrustFV achieves state-of-the-art performance on FactKG and CWQ datasets.

LLM-based fact-checking verifies complex claims through decomposition, retrieval, and reasoning aggregation. Existing methods expand verification inputs to cover more factual conditions, but added fragments can be redundant or misleading, compete with decisive evidence, and bias predictions toward surface relevance. Debiasing methods mitigate this bias by estimating shortcut effects from claim-side cues or biased signals in constructed inputs. However, because they calibrate verifier predictions after input construction, they are not designed to repair missing or noisy verification states. Thus, complex fact-checking should balance factual coverage and noise control during verification-state construction. We propose ERGA, an Explorer–Refiner Game with Backdoor-Adjusted Verdict Aggregation. ERGA formulates verification as editable state construction, initializing the state from decomposed verification units, retrieved evidence, and local analyses. It then runs an adversarial Explorer–Refiner game over this state: the Explorer strengthens under-covered factual signals, while the Refiner merges redundancy and separates weak or misleading fragments. Through this game interaction, ERGA obtains a refined state with a better coverage–noise balance and a removed fragment record. For verdict aggregation, ERGA uses the removed-fragment record to condition verifier predictions and aggregates them in a backdoor-adjusted form to reduce residual shortcut effects. Experiments on HoVer and FEVEROUS-S show that ERGA outperforms strong baselines.

Emotional support conversation (ESC) aims to alleviate users’ psychological stress. Selecting the appropriate strategy is crucial for effective emotional support. Current strategy planner-based methods prioritize immediate responses while neglecting users’ future reactions. Some studies retrieve historical examples with similar emotions to the current utterance, then anticipating future emotions based on next-turn emotions of historical examples. However, their retrievals focus on the current emotion (i.e. a single-turn emotion state), while they ignore the evolution of user’s emotion before the current state. We argue that retrievals considering the whole emotional trajectories enables models to capture the dynamic emotional needs, thereby enhancing the anticipation of future emotions. To this end, we propose Markov-driven emotion anticipation framework with emotion trajectory-aware retrieval for LLM-based ESC, which anticipates future emotion states to guide strategy planning and achieve sustained emotional support. First, we construct a dynamic emotion memory and perform hierarchical retrieval that combines semantic matching and emotion trajectory alignment. Then, we model emotional transitions as Markov chains, leveraging trajectory-aware retrieval to estimate future emotion. Finally, we use the anticipated emotion to steer LLMs in generating candidate strategies and introduce active online learning to optimize the planner, boosting its robustness on diverse users. Experiments on two datasets with two models shows that our method excels all baselines.

Large language model agents in social deduction games must reason and speak under hidden roles, shifting public beliefs, and social pressure. Existing methods mainly target final win rate through strategic prompting, belief modeling, or reinforcement learning, but they leave the utterance-level trade-off between belief correction and social risk largely implicit. In false-consensus intervention, useful private information can move the group toward the truth, while poorly timed or overly direct speech can trigger suspicion or rejection. We propose PACR-Pro, a risk-aware rhetorical decision framework for false-consensus intervention. PACR-Pro maps each speaking step to consensus-conflict encoding, structured candidate-action generation, Pareto-based selection, and risk-adaptive utterance realization. A dual-branch Epistemic-Social Critic scores candidate actions, and a state-conditioned 𝜆 policy adjusts the trade-off between epistemic value and social safety. We train these decision modules offline from historical games and evaluate PACR-Pro on One Night Ultimate Werewolf (ONUW) as the test platform. Experiments on standard ONUW games and the Wolf-Pack Induction Benchmark show that PACR-Pro achieves the best mixed-agent win rate, the highest attack win rate, the lowest attack vote rate, and the smallest attack-control win-rate gap among all baselines.

Incomplete-Modality Multimodal Sentiment Analysis extends multimodal sentiment analysis (MSA) to infer sentiment from language, audio, and visual streams when one or more modalities are incomplete at inference time. Existing methods improve incomplete-modality MSA through feature reconstruction, distillation, or fusion adjustment, but usually assign similar correction strength to all samples with incomplete inputs. This overlooks the uneven emotional impact of modality missingness: mild missingness often leaves sentiment recognizable, whereas severe missingness can distort the expressed emotion and therefore requires stronger correction. We propose HyDRA (Hyperbolic Damage-aware Recovery and Adaptation), a damage-aware and text-guided hyperbolic framework that aligns correction strength with missingness-induced sentiment damage. HyDRA first recovers lost emotional cues from available modalities, then places each sample in the Poincaré ball according to estimated damage severity, and finally applies damage-weighted hyperbolic correction toward a complete-modality teacher anchor. Since hyperbolic distances expand near the boundary, heavily incomplete samples receive stronger correction while lightly incomplete samples are gently adjusted. Experiments on MOSI, MOSEI, and SIMS show SOTA robustness across incomplete-modality conditions.

Deploying large language models (LLMs) on sensitive data via Retrieval-Augmented Generation (RAG) introduces severe privacy risks. Recent studies apply Differential Privacy (DP) to LLMs with RAG for formal privacy guarantees. However, existing DP-RAG frameworks rapidly exhaust the privacy budget. Although recent efforts attempt to save the budget by narrowing the retrieval scope or sparsifying private generation, these methods themselves cumulatively consume the budget, whereas they could actually rely merely on public information or at a negligible one-time privacy cost. This mismatch fails to align budget expenditure with the model’s actual reliance on private data, causing substantial waste on operations that require no private access. To address this, we propose SparsePay-RAG, adopting "only pay what you must spend" as its core principle. Using public information as a zero-privacy prior, it charges the privacy budget only for the private increment. Specifically, SparsePay-RAG narrows the retrieval scope via public topic-guided clustering, adaptively controls private access frequency without privacy cost through isotonic cross-layer trajectory fitting, and compresses per-access budget via DP contrastive decoding. Under strong privacy constraints, experiments show SparsePay-RAG achieves superior privacy-utility trade-offs over baselines.

Multimodal fake news detection predicts post veracity from text, images, and inter-post context. Existing content-centric and graph-propagation methods model post evidence and inter-post context, but graph propagation still assumes connected posts provide compatible veracity evidence. In multimodal fake news graphs, edges often represent post relatedness rather than veracity consistency, so propagation can pass misleading evidence across conflicting posts. In this paper, we propose Semantic-Structural Propagation Calibration (SSPC), which uses noisy LLM pseudo-labels to calibrate where and how much propagation should be trusted, rather than using them as direct labels. SSPC refines graph edges from high-confidence pseudo-label agreement, computes a graph-encoded base prediction and an explicitly graph-propagated prediction, and learns node-level propagation reliance with a cross-fitted utility gate. Experiments on Twitter, PHEME, and Weibo show SOTA performance over strong multimodal and graph-propagation fake news detection baselines.

Text understanding application often suffers from domain shifts. To handle testing domains, domain adaptation (DA) is trained to adapt to a fixed and observed testing domain; a more challenging paradigm, test-time adaptation (TTA), cannot access the testing domain during training and online adapts to the testing samples during testing, where the samples are from a fixed domain. We aim to explore a more practical and underexplored scenario, continual test-time adaptation (CTTA) for text understanding, which involves a sequence of testing (unobserved) domains in testing. Current CTTA methods struggle in reducing error accumulation over domains and enhancing generalization to handle unobserved domains: 1) Noise-filtering reduces accumulated errors but discards useful information, and 2) accumulating historical domains enhances generalization, but it is hard to achieve adaptive accumulation. In this paper, we propose a CTTA-T (continual test-time adaptation for text understanding) framework adaptable to evolving target domains: CTTA-T adopts a teacher-student framework, where the teacher is equipped with domain awareness and generalization for evolving domains. To improve teacher predictions, we propose a refine-then-filter based on dropout-driven consistency, which calibrates predictions and removes unreliable guidance. For the adaptation–generalization trade-off, we construct a domain-aware teacher by dynamically accumulating cross-domain semantics via incremental PCA, which continuously tracks domain shifts. Experiments show CTTA-T excels baselines.

Existing LLM hallucination mitigation methods, including prompt engineering and model optimization, either hardly alter models’ internal knowledge or have poor cross-domain generalization. Contrastive decoding mitigates hallucinations by using layer-wise differences in LLMs. However, prior studies only explore transformer-based models (e.g., GPT), ignoring other effective frameworks like mixture-of-experts (MoE) models. Since MoE alters the traditional transformer architecture, we conduct empirical studies to investigate whether similar layer-wise differences exist in MoEs. Our results show that they do not exist in MoE with shared experts; nevertheless, across different MoEs, higher layers exhibit distinct expert activation patterns between factual and non-factual outputs. Building on these, we propose EAACD, an expert-aware adaptive contrast decoding that uses expert differences in MoE’s higher layers to mitigate hallucinations on QA tasks. EAACD splits high-layer experts into a higher-reliability group and several lower-reliability groups based on their confidence and consistency. It contrasts the higher-reliability group’s prediction with each lower-reliability group’s prediction to calibrate the model’s original predictions. To strengthen this contrast, EAACD amplifies hallucinations from lower-reliability experts via attention and masking to provide stronger negative references. EAACD outperforms all baselines on four datasets

As the text generated by large language models (LLMs) increasingly resembles human-written text (HWT), detecting LLM-generated text (LGT) is crucial to avoid malicious use of LGT. Recent research treats LGT detection as an out-of-distribution (OOD) detection problem and views HWT as the OOD. However, existing OOD detection methods assume that LGT is a single homogeneous distribution. In practice, LGT exhibits different characteristics under different generation conditions. Text from weaker LLMs tends to form distinct clusters and is easy to detect, whereas text from stronger models significantly overlaps with HWTs and is hard to detect. To address the issue, in this paper, we propose an LGT detection framework based on density-aware manifold learning and the construction of hybrid Mahalanobis energy. We apply density-aware manifold learning with Laplacian smoothness and density regularization in embedding space, amplifying differences between LGT and HWT. We further propose a density-adaptive hybrid Mahalanobis metric that combines global and local covariance via density weighting, enabling adaptation to the manifold-aware embedding space. Finally, based on the metric, we define the distribution energy as a measure of distribution discrepancy, and we employ energy learning and contrastive learning to separate distributions hierarchically, establishing a clear OOD decision boundary. Experiments show that our method outperforms strong baselines.

Large language models often lose track of persistent instructions and relevant information as multi-turn conversations grow. We study this cumulative contextual decay through three related failure modes: attention pollution, dilution, and drift. We propose **REA** (Role-aware Heuristic Episodic Attention), a context-management framework that assigns different persistence and representation policies to instructions and episodic interactions. Instructional Memory retains identified global constraints in a dedicated prefix. Episodic Memory preserves user inputs and compresses model replies, while heuristic retrieval selects raw text, compressed representations, or omission for each historical turn. On Long-MT-Bench+, REA improves the judge score from 6.32 to 7.36 on a 10-point scale, a 16.5% relative gain over the Vanilla baseline, and reduces average latency by 2.91×. Additional evaluations show aggregate gains on three backbones spanning 1.7B–7B parameters and on Chinese and English role-playing tasks. These results support role-aware context management as a practical approach to maintaining conversational continuity and instruction adherence.

Using large language models (LLMs) has a potential risk of privacy leakage since the data with sensitive information may be used for fine-tuning the LLMs. Differential privacy (DP) provides theoretical guarantees of privacy protection, but its practical application in LLMs still has the problem of privacy-utility trade-off. Researchers synthesized data with strong generation capabilities closed-source LLMs (i.e., GPT-4) under DP to alleviate this problem, but this method is not so flexible in fitting the given privacy distributions without fine-tuning. Besides, such methods can hardly balance the diversity of synthetic data and its relevance to target privacy data without accessing so much private data. To this end, this paper proposes DPGA-TextSyn, combining general LLMs with genetic algorithm (GA) to produce relevant and diverse synthetic text under DP constraints. First, we integrate the privacy gene (i.e., metadata) to generate better initial samples. Then, to achieve survival of the fittest and avoid homogeneity, we use privacy nearest neighbor voting and similarity suppression to select elite samples. In addition, we expand elite samples via genetic strategies such as mutation, crossover, and generation to expand the search scope of GA. Experiments show that this method significantly improves the performance of the model in downstream tasks while ensuring privacy.

LLMs face privacy risks when handling sensitive data. To ensure privacy, researchers use differential privacy (DP) to provide protection by adding noise during LLM training. However, users may be hesitant to share complete data with LLMs. Researchers follow local DP to sanitize the text on the user side and feed non-sensitive text to LLMs. The sanitization usually uses a fixed non-sensitive token list or a fixed noise distribution, which induces the risk of being attacked or semantic distortion. We argue that the token’s protection level should be adaptively adjusted according to its semantic-based information to balance the privacy-utility trade-off. In this paper, we propose DYNTEXT, an LDP-based Dynamic Text sanitization for privacy-preserving LLM inference, which dynamically constructs semantic-aware adjacency lists of sensitive tokens to sample non-sensitive tokens for perturbation. Specifically, DYNTEXT first develops a semantic-based density modeling under DP to extract each token’s density information. We propose token-level smoothing sensitivity by combining the idea of global sensitivity (GS) and local sensitivity (LS), which dynamically adjusts the noise scale to avoid excessive noise in GS and privacy leakage in LS. Then, we dynamically construct an adjacency list for each sensitive token based on its semantic density information. Finally, we apply the replacement mechanism to sample non-sensitive, semantically similar tokens from the adjacency list to replace sensitive tokens. Experiments show that DYNTEXT excels strong baselines on three datasets.

With the emergence of new topics on social media as sources of rumor propagation, addressing the domain shift between the source and target domain and the target domain samples scarcity remains a crucial task in cross-domain rumor detection. Traditional deep learning-based methods and LLM-based methods are mostly focused on the in-domain condition, thus having poor performance in cross-domain setting. Existing domain adaptation rumor detection approaches ignore the data generalization differences and rely on a large amount of unlabeled target domain samples to achieve domain adaptation, resulting in less effective on emerging topic rumor detection. In this paper, we propose a Gradient Coherence guided Meta-Learning approach (GCML) for emerging topics rumor detection. Firstly, we calculate the task generalization score of each source task (sampled from source domain) from a gradient coherence perspective, and selectively learn more “generalizable” tasks that are more beneficial in adapting to the target domain. Secondly, we leverage meta-learning to alleviate the target domain samples scarcity, which utilizes task generalization scores to re-weight meta-test gradients and adaptively updates learning rate. Extensive experimental results on real-world datasets show that our method substantially outperforms SOTA baselines.

With the continuous advancement in the performance of large language models (LLMs), their demand for computational resources and memory has significantly increased, which poses major challenges for efficient inference on consumer-grade devices and legacy servers. These devices typically feature relatively weaker GPUs and stronger CPUs. Although techniques such as parameter offloading and partial offloading can alleviate GPU memory pressure to some extent, their effectiveness is limited due to communication latency and suboptimal hardware resource utilization. To address this issue, we propose Dovetail—a lossless inference acceleration method that leverages the complementary characteristics of heterogeneous devices and the advantages of speculative decoding. Dovetail deploys a draft model on the GPU to perform preliminary predictions, while a target model running on the CPU validates these outputs. By reducing the granularity of data transfer, Dovetail significantly minimizes communication overhead. To further improve efficiency, we optimize the draft model specifically for heterogeneous hardware environments by reducing the number of draft tokens to lower parallel verification latency, increasing model depth to enhance predictive capabilities, and introducing a Dynamic Gating Fusion (DGF) mechanism to improve the integration of feature and embedding information. We conduct comprehensive evaluations of Dovetail across various consumer-grade GPUs, covering multiple tasks and mainstream models. Experimental results on 13B models demonstrate that Dovetail achieves inference speedups ranging from 1.79× to 10.1× across different devices, while maintaining consistency and stability in the distribution of generated texts.

LLMs with in-context learning (ICL) obtain remarkable performance but are sensitive to the quality of ICL examples. Prior works on ICL example selection explored unsupervised heuristic methods and supervised LLM-based methods, but they typically focus on the selection of individual examples and ignore correlations among examples. Researchers use the determinantal point process (DPP) to model negative correlations among examples to select diverse examples. However, the DPP fails to model positive correlations among examples, while ICL still requires the positive correlations of examples to ensure the consistency of examples, which provides a clear instruction for LLMs. In this paper, we propose an ICL example selection method based on the nonsymmetric determinantal point process (NDPP) to capture positive and negative correlations, considering both the diversity and the relevance among ICL examples. Specifically, we optimize NDPP via kernel decomposition-based MLE to fit a constructed pseudo-labeled dataset, where we also propose a low-rank decomposition to reduce the computational cost. Further, we perform query-aware kernel adaptation on our NDPP to customize the input query, and we select examples via a MAP inference based on the adapted NDPP. Experimental results show our model outperforms strong baselines in ICL example selection.

LLMs demonstrate remarkable utility but remain vulnerable to jailbreak attacks that aim to elicit harmful responses. Existing defenses, including post-training alignment and prompt engineering, rely on training on safety-annotated datasets and safe prompt templates, struggling with adaptability to out-of-distribution (OOD) attacks. Steering internal representations of LLMs provides real-time adjustments to defend against OOD attacks. However, it struggles with maintaining model utility, since modifying the representation disrupts the forward pass of inference. It barely considers the competitive objectives of helpfulness and harmlessness in LLMs. We argue that adversarial game-based approaches promise a solution for conflicts between the two objectives. In this paper, we propose Adversarial Game Defense (AGD), an adversarial game-based defense method that dynamically adjusts LLMs’ internal representations to achieve a balanced trade-off between helpfulness and harmlessness. AGD first proposes an interquartile range (IQR) method to detect abnormal attention weights and correct the abnormal weights via adversarial training. AGD adopts a bi-level optimization to play a two-player variable-sum game to approach Nash Equilibrium (NE), where the two players adversarially refine head activations for helpfulness and harmlessness respectively. Furthermore, AGD applies an expert model to next-token sampling to generate safer responses. Experiments show that AGD significantly improves LLMs’ safety over all baselines.

Non-collaborative dialogue involves two participants with conflicting interests engaging in a multi-round dialogue to achieve their own goals. Strategy planning is the key to guiding both participants towards a consensus. Most LLMs-based methods use stimulus prompts or external strategy planners for strategy planning. However, stimulus prompts fail to teach LLMs to plan dialogue strategies explicitly. Moreover, training external strategy planners doesn’t fully account for adversarial interactions, thereby limiting their effectiveness against tough resisters. In this paper, to mitigate the above issues, we propose GAIA, a Game-based Adversarial self-play InterActive training paradigm, which constructs an adversarial two-player (a persuader and a resister) zero-sum game and guides the game to approximate Nash Equilibrium (NE) via reinforcement learning (RL) for the non-collaborative dialogues. First, we design a Chain-of-Mind prompt to reason the resister’s dialogue act step-by-step to plan the persuasive strategies. Secondly, to adversarially improve the persuader, we construct diverse resistant planners and theoretically improve the persuader’s optimal lower bound. Finally, we iteratively optimise their policies via adversarial self-play interactive RL and design an 𝜖-NE verification algorithm to approximate the game’s NE. Experiments on three datasets show that our model obtains state-of-the-art performance.

Rumor detection on social media has become an emerging topic. Traditional deep learning-based methods model rumors based on content, propagation structure, or user behavior, but these approaches are constrained by limited modeling capacity and insufficient training corpora. Recent studies have explored using LLMs for rumor detection through supervised fine-tuning (SFT), but face two issues: 1) unreliable samples sometimes mislead the model learning; 2) the model only learns the most salient input-output mapping and skips in-depth analyses of the rumored content for convenience. To address these issues, we propose an SFT-based LLM rumor detection model with Influence guided Sample selection and Game-based multi-perspective Analysis (ISGA). Specifically, we first introduce the Influence Score (IS) to assess the impact of samples on model predictions and select samples for SFT. We also approximate IS via Taylor expansion to reduce computational complexity. Next, we use LLMs to generate in-depth analyses of news content from multiple perspectives and model their collaborative process for prediction as a cooperative game. Then we utilize the Shapley value to quantify the contribution of each perspective for selecting informative perspective analyses. Experiments show that ISGA excels existing SOTA on three datasets.

Multi-agent collaborative tasks exhibit exceptional capabilities in natural language applications and generation. By prompting agents to assign clear roles, it is possible to facilitate cooperation and achieve complementary capabilities among LLMs. A common strategy involves adopting a relatively general role assignment mechanism, such as introducing a “judge” or a “summarizer”. However, these approaches lack task-specific role customization based on task characteristics. Another strategy involves decomposing the task based on domain knowledge and task characteristics, followed by assigning appropriate roles according to LLMs’ respective strengths, such as programmers and testers. However, in some given tasks, obtaining domain knowledge related to task characteristics and getting the strengths of different LLMs is hard. To solve these problems, we propose a Multi-LLM Cooperation (MLC) framework with automatic role assignment capabilities. The core idea of the MLC is to initialize role assignments randomly and then allow the role embeddings to be learned jointly with the downstream task. To capture the state transitions of multiple LLMs during turn-based speaking, the role embedding is sequence-aware. At the same time, to avoid role convergence, the role differentiation module in MLC encourages behavioral differentiation between LLMs while ensuring the LLM team consistency, guiding different LLMs to develop complementary strengths from the optimization level. Our experiments on seven datasets demonstrate that MLC significantly enhances collaboration and expertise, which collaboratively addresses multi-agent tasks.

Counterspeech is an effective way to combat online hate speech. Considering the multifaceted nature of online hate speech, counterspeech with varying intents (e.g., denouncing or empathy) has significant potential to mitigate hate speech effectively. Recently, controlled approaches based on large language models (LLMs) have been explored to generate intent-specific counterspeech. Due to the lack of attention to intent-specific information by LLMs during the decoding process, those methods cater more to the semantic information rather than matching with the desired intents. Further, there are still limitations in quantitatively evaluating the effectiveness of counterspeech with different intents in mitigating hate speech. In this paper, to address the above issues, we propose DART, an LLMs-based DuAl-discRiminaTor guided framework for counterspeech generation. We employ an intent-aware discriminator and hate-mitigating discriminator to jointly guide the decoding preferences of LLMs, which facilitates the model towards generating counterspeech catering to specific intent and hate mitigation. We apply a maximum-margin relative objective for training discriminators. This objective leverages the distance between counterspeech aligned with the desired target (such as specific intent or effectiveness in hate mitigation) and undesired as an effective learning signal. Extensive experiments show that DART achieves excellent performances in matching the desired intent and mitigating hate.

Unsupervised text style transfer aims to modify the style of a sentence while preserving its content without parallel corpora. Existing approaches attempt to separate content from style, but some words contain both content and style information. It makes them difficult to disentangle, where unsatisfactory disentanglement results in the loss of the content information or the target style. To address this issue, researchers adopted a “cycle reconstruction” mechanism to maintain content information, but it is still hard to achieve satisfactory content preservation due to incomplete disentanglement. In this paper, we propose a new disentanglement-based method, StyleFlow, which effectively avoids the loss of contents through a better cycle reconstruction via a reversible encoder. The reversible encoder is a normalizing flow that can not only produce output given input but also infer the exact input given the output reversely. We design a stack of attention-aware coupling layers, where each layer is reversible and adopts the attention mechanism to improve the content-style disentanglement. Moreover, we propose a data augmentation method based on normalizing flow to enhance the training data. Our experiments on sentiment transfer and formality transfer tasks show that StyleFlow outperforms strong baselines on both content preservation and style transfer.

Watermarking enables people to determine whether the text is generated by a specific model. It injects a unique signature based on the “green-red” list that can be tracked during detection, where the words in green lists are encouraged to be generated. Recent researchers propose to fix the green/red lists or increase the proportion of green tokens to defend against paraphrasing attacks. However, these methods cause degradation of text quality due to semantic disparities between the watermarked text and the unwatermarked text. In this paper, we propose a semantic-aware watermark method that considers contexts to generate a semantic-aware key to split a semantically balanced green/red list for watermark injection. The semantic balanced list reduces the performance drop due to adding bias on green lists. To defend against paraphrasing attacks, we generate the watermark key considering the semantics of contexts via locally sensitive hashing. To improve the text quality, we propose to split green/red lists considering semantics to enable the green list to cover almost all semantics. We also dynamically adapt the bias to balance text quality and robustness. The experiments show our advantages in both robustness and text quality comparable to existing baselines.

Low-resource languages (LRLs) face challenges in supervised neural machine translation (NMT) due to limited parallel data, prompting research in unsupervised NMT.Unsupervised NMT (UNMT), without requiring ground truth, provides solutions for LRL translations using synthetic pseudo-parallel data and parallel data from auxiliary language pairs. However, they usually encounter translation errors, including errors from synthetic data and from auxiliary language pairs with linguistic biases.We argue that large language models (LLMs) mitigate UNMT’s translation errors by dynamically organizing auxiliary languages in prompts to improve LRL translations. In this paper, we propose PrObability-driven Meta-graph Prompter (POMP), an approach employing a dynamic graph to organize multiple auxiliary languages, to prompt LLMs in LRL translations. POMP proposes a language-specific meta-graph that dynamically samples multiple translation paths to organize auxiliary languages in constructing prompts. Following the path, POMP prompts LLMs to translate with a mixture of auxiliary languages. We achieve the meta-graph’s evolution by back-propagating evaluation scores to update probabilities on the graph.Our experimental improvements show POMP’s effectiveness on LRLs’ translation.

Conditional story generation is significant in human-machine interaction, particularly in producing stories with complex plots. While Large language models (LLMs) perform well on multiple NLP tasks, including story generation, it is challenging to generate stories with both complex and creative plots. Existing methods often rely on detailed prompts to guide LLMs to meet target conditions, which inadvertently restrict the creative potential of the generated stories. We argue that leveraging information from exemplary human-written stories facilitates generating more diverse plotlines. Delving deeper into story details helps build complex and credible plots. In this paper, we propose a retrieval-auGmented stoRy generation framework with a fOrest of eVidEnce (GROVE) to enhance stories’ complexity. We build a retrieval repository for target conditions to produce few-shot examples to prompt LLMs. Additionally, we design an “asking-why” prompting scheme that extracts a forest of evidence, providing compensation for the ambiguities that may occur in the generated story. This iterative process uncovers underlying story backgrounds. Finally, we select the most fitting chains of evidence from the evidence forest and integrate them into the generated story, thereby enhancing the narrative’s complexity and credibility. Experimental results and numerous examples verify the effectiveness of our method.

Attribute-based generation methods are of growing significance in controlling the generation of large pre-trained language models (PLMs). Existing studies control the generation by (1) finetuning the model with attributes or (2) guiding the inference processing toward control signals while freezing the PLM. However, finetuning approaches infuse domain bias into generation, making it hard to generate out-of-domain texts. Besides, many methods guide the inference in its word-by-word generation, pushing the word probability to the target attributes, resulting in less fluent sentences. We argue that distilling controlling information from natural texts can produce fluent sentences while maintaining high controllability. In this paper, we propose GRAdient-guided Controllable rEtrieval (GRACE), a retrieval-augmented generation framework to facilitate the generation of fluent sentences with high attribute relevance. GRACE memorizes the semantic and attribute information from unlabeled corpora and applies a controllable retrieval to obtain desired information. For the generation, we design techniques to eliminate the domain bias from the retrieval results and integrate it into the generation model. Additionally, we propose a gradient-guided generation scheme that iteratively steers generation toward higher attribute relevance. Experimental results and quantities of examples verify the effectiveness of our method.

Text classification tasks often encounter few-shot scenarios with limited labeled data, and addressing data scarcity is crucial. Data augmentation with mixup merges sample pairs to generate new pseudos, which can relieve the data deficiency issue in text classification. However, the quality of pseudo-samples generated by mixup exhibits significant variations. Most of the mixup methods fail to consider the varying degree of learning difficulty in different stages of training. And mixup generates new samples with one-hot labels, which encourages the model to produce a high prediction score for the correct class that is much larger than other classes, resulting in the model’s over-confidence. In this paper, we propose a self-evolution learning (SE) based mixup approach for data augmentation in text classification, which can generate more adaptive and model-friendly pseudo samples for the model training. SE caters to the growth of the model learning ability and adapts to the ability when generating training samples. To alleviate the model over-confidence, we introduce an instance-specific label smoothing regularization approach, which linearly interpolates the model’s output and one-hot labels of the original samples to generate new soft labels for label mixing up. Through experimental analysis, experiments show that our SE brings consistent and significant improvements upon different mixup methods. In-depth analyses demonstrate that SE enhances the model’s generalization ability.

Self-training emerges as an important research line on domain adaptation. By taking the model’s prediction as the pseudo labels of the unlabeled data, self-training bootstraps the model with pseudo instances in the target domain. However, the prediction errors of pseudo labels (label noise) challenge the performance of self-training. To address this problem, previous approaches only use reliable pseudo instances, i.e., pseudo instances with high prediction confidence, to retrain the model. Although these strategies effectively reduce the label noise, they are prone to miss the hard examples. In this paper, we propose a new self-training framework for domain adaptation, namely Domain adversarial learning enhanced Self-Training Framework (DaMSTF). Firstly, DaMSTF involves meta-learning to estimate the importance of each pseudo instance, so as to simultaneously reduce the label noise and preserve hard examples. Secondly, we design a meta constructor for constructing the meta-validation set, which guarantees the effectiveness of the meta-learning module by improving the quality of the meta-validation set. Thirdly, we find that the meta-learning module suffers from the training guidance vanish- ment and tends to converge to an inferior optimal. To this end, we employ domain adversarial learning as a heuristic neural network initialization method, which can help the meta-learning module converge to a better optimal. Theoretically and experimentally, we demonstrate the effectiveness of the proposed DaMSTF. On the cross-domain sentiment classification task, DaMSTF improves the performance of BERT with an average of nearly 4%.

Emotional conversation systems generate responses for the input queries considering the speaker’s emotions in a conversation. Existing emotional conversation systems output emotional responses according to either a given emotion or the user’s emotion reflected in the input queries. Following a given emotion may lead to an emotional drift between the given emotion and the conversation state, and following only the user’s emotion may aggravate the user’s negative feelings if users suffer from a negative mood. In this paper, we propose to generate empathetic responses catering to the user’s emotions while leading the conversation to be emotionally positive. Particularly, by abstracting the conversation corpus, we extract and store the different responding strategies for different users’ emotions and conversational topics into a memory. We encourage positive emotions in conversation via a sentiment evaluator. We model the memory outputs with a Gaussian mixture distribution and sample a final responding strategy from the distribution. The strategy acts as a condition to a transformer model to generate responses. The experiments verify our model surpasses the baseline methods in appropriateness, diversity, and generating emotionally positive responses.

Lifelong learning aims to accumulate knowledge and alleviate catastrophic forgetting when learning tasks sequentially. However, existing lifelong language learning methods only focus on the supervised learning setting. Unlabeled data, which can be easily accessed in real-world scenarios, are underexplored. In this paper, we explore a novel setting, semi-supervised lifelong language learning (SSLL), where a model learns sequentially arriving language tasks with both labeled and unlabeled data. We propose an unlabeled data enhanced lifelong learner to explore SSLL. Specially, we dedicate task-specific modules to alleviate catastrophic forgetting and design two modules to exploit unlabeled data: (1) a virtual supervision enhanced task solver is constructed on a teacher-student framework to mine the underlying knowledge from unlabeled data; and (2) a backward augmented learner is built to encourage knowledge transfer from newly arrived unlabeled data to previous tasks. Experimental results on various language tasks demonstrate our model’s effectiveness and superiority over competitive baselines under the new setting SSLL.

Lifelong learning (LL) is vital for advanced task-oriented dialogue (ToD) systems. To address the catastrophic forgetting issue of LL, generative replay methods are widely employed to consolidate past knowledge with generated pseudo samples. However, most existing generative replay methods use only a single task-specific token to control their models. This scheme is usually not strong enough to constrain the generative model due to insufficient information involved. In this paper, we propose a novel method, prompt conditioned VAE for lifelong learning (PCLL), to enhance generative replay by incorporating tasks’ statistics. PCLL captures task-specific distributions with a conditional variational autoencoder, conditioned on natural language prompts to guide the pseudo-sample generation. Moreover, it leverages a distillation process to further consolidate past knowledge by alleviating the noise in pseudo samples. Experiments on natural language understanding tasks of ToD systems demonstrate that PCLL significantly outperforms competitive baselines in building lifelong learning models.

In knowledge distillation, a student model is trained with supervisions from both knowledge from a teacher and observations drawn from a training data distribution. Knowledge of a teacher is considered a subject that holds inter-class relations which send a meaningful supervision to a student; hence, much effort has been put to find such knowledge to be distilled. In this paper, we explore a question that has been given little attention: “when to distill such knowledge.” The question is answered in our work with the concept of model calibration; we view a teacher model not only as a source of knowledge but also as a gauge to detect miscalibration of a student. This simple and yet novel view leads to a hard gate knowledge distillation scheme that switches between learning from a teacher model and training data. We verify the gating mechanism in the context of natural language generation at both the token-level and the sentence-level. Empirical comparisons with strong baselines show that hard gate knowledge distillation not only improves model generalization, but also significantly lowers model calibration error.

Building models of natural language processing (NLP) is challenging in low-resource scenarios where limited data are available. Optimization-based meta-learning algorithms achieve promising results in low-resource scenarios by adapting a well-generalized model initialization to handle new tasks. Nonetheless, these approaches suffer from the memorization overfitting issue, where the model tends to memorize the meta-training tasks while ignoring support sets when adapting to new tasks. To address this issue, we propose a memory imitation meta-learning (MemIML) method that enhances the model’s reliance on support sets for task adaptation. Specifically, we introduce a task-specific memory module to store support set information and construct an imitation module to force query sets to imitate the behaviors of support sets stored in the memory. A theoretical analysis is provided to prove the effectiveness of our method, and empirical results also demonstrate that our method outperforms competitive baselines on both text classification and generation tasks.

Text style transfer aims to alter the style (e.g., sentiment) of a sentence while preserving its content. A common approach is to map a given sentence to content representation that is free of style, and the content representation is fed to a decoder with a target style. Previous methods in filtering style completely remove tokens with style at the token level, which incurs the loss of content information. In this paper, we propose to enhance content preservation by implicitly removing the style information of each token with reverse attention, and thereby retain the content. Furthermore, we fuse content information when building the target style representation, making it dynamic with respect to the content. Our method creates not only style-independent content representation, but also content-dependent style representation in transferring style. Empirical results show that our method outperforms the state-of-the-art baselines by a large margin in terms of content preservation. In addition, it is also competitive in terms of style transfer accuracy and fluency.

Neural conversation models are known to generate appropriate but non-informative responses in general. A scenario where informativeness can be significantly enhanced is Conversing by Reading (CbR), where conversations take place with respect to a given external document. In previous work, the external document is utilized by (1) creating a context-aware document memory that integrates information from the document and the conversational context, and then (2) generating responses referring to the memory. In this paper, we propose to create the document memory with some anticipated responses in mind. This is achieved using a teacher-student framework. The teacher is given the external document, the context, and the ground-truth response, and learns how to build a response-aware document memory from three sources of information. The student learns to construct a response-anticipated document memory from the first two sources, and teacher’s insight on memory creation. Empirical results show that our model outperforms the previous state-of-the-art for the CbR task.

Neural generative models for open-domain chit-chat conversations have become an active area of research in recent years. A critical issue with most existing generative models is that the generated responses lack informativeness and diversity. A few researchers attempt to leverage the results of retrieval models to strengthen the generative models, but these models are limited by the quality of the retrieval results. In this work, we propose a memory-augmented generative model, which learns to abstract from the training corpus and saves the useful information to the memory to assist the response generation. Our model clusters query-response samples, extracts characteristics of each cluster, and learns to utilize these characteristics for response generation. Experimental results show that our model outperforms other competitive baselines.

Generative conversational systems are attracting increasing attention in natural language processing (NLP). Recently, researchers have noticed the importance of context information in dialog processing, and built various models to utilize context. However, there is no systematic comparison to analyze how to use context effectively. In this paper, we conduct an empirical study to compare various models and investigate the effect of context information in dialog systems. We also propose a variant that explicitly weights context vectors by context-query relevance, outperforming the other baselines.

Neural conversation systems, typically using sequence-to-sequence (seq2seq) models, are showing promising progress recently. However, traditional seq2seq suffer from a severe weakness: during beam search decoding, they tend to rank universal replies at the top of the candidate list, resulting in the lack of diversity among candidate replies. Maximum Marginal Relevance (MMR) is a ranking algorithm that has been widely used for subset selection. In this paper, we propose the MMR-BS decoding method, which incorporates MMR into the beam search (BS) process of seq2seq. The MMR-BS method improves the diversity of generated replies without sacrificing their high relevance with the user-issued query. Experiments show that our proposed model achieves the best performance among other comparison methods.