Traditional research on Continual Machine Translation (CMT) has long been predicated on the assumption that catastrophic forgetting stems from the irreversible erasure of underlying linguistic knowledge. In this work, we challenge this conventional wisdom within the context of Large Language Models (LLMs). Through systematic behavioral and representational analyses, we demonstrate that the model’s latent knowledge often remains intact, yet its ability to align this knowledge with instructions becomes compromised. We attribute this " expressive drift" to the ubiquitous Supervised Fine-Tuning (SFT) paradigm. By minimizing token-level loss against fixed references, SFT neglects the inherent plurality of valid translations for a single source, inadvertently coercing the model toward narrow external distributions that conflict with its internal linguistic priors. To rectify this, we advocate for a transition from SFT to a Reinforcement Learning (RL) paradigm, which optimizes performance directly within the model’s own generative distribution. Our results show that RL significantly mitigates forgetting in both translation-specific and general tasks by preserving the model’s expressive stability. This work redefines the fundamental challenge of CMT and positions RL as a more principled approach for sustainable multilingual alignment in the era of LLMs.

Recent advances in Reinforcement Post-Training (RPT) have substantially improved the mathematical reasoning capabilities of large language models (LLMs), but it remains unclear how well such gains transfer across languages. In this work, we study cross-lingual mathematical reasoning transfer in English-centric LLMs under controlled multilingual evaluation. We first evaluate a range of English-centric reasoning models on multilingual mathematical reasoning benchmarks and find that transfer is highly variable across initial model choice, target language, and training paradigm. Through controlled comparative experiments, we further find that stronger English-centric initialization does not necessarily lead to stronger relative transfer efficiency, even when it yields higher multilingual reasoning accuracy. Finally, we show that the largest improvement comes from moving beyond English-only supervision: adding a single parallel language yields a substantial gain, while further gains from additional languages are smaller but consistent. Overall, our results highlight the need to evaluate multilingual mathematical reasoning directly, using both absolute performance and relative transfer-efficiency metrics.

Latent reasoning offers a more efficient alternative to explicit reasoning by compressing intermediate reasoning into continuous representations and substantially shortening reasoning chains. However, existing latent reasoning methods mainly focus on supervised learning, and reinforcement learning in latent space remains highly unstable. We study this problem through the lens of Group Relative Policy Optimization (GRPO), and show that directly adapting GRPO to latent reasoning is fundamentally non-trivial: latent reasoning changes both the probability density and the sampling mechanism, causing three coupled bottlenecks: absence of intrinsic latent manifolds, where unconstrained exploration pushes rollouts off the valid latent manifold; exploration-optimization misalignment, where trajectory-level rewards can induce incorrect token-level updates; and latent mixture non-closure, where jointly reinforcing multiple correct latent paths can produce an invalid averaged state. To address them, we propose Latent-GRPO, which combines invalid-sample advantage masking, one-sided noise sampling, and optimal correct-path first-token selection. Across four low-difficulty benchmarks (e.g., GSM8K-Aug) and four high-difficulty benchmarks (e.g., AIME), Latent-GRPO improves over its latent initialization by 7.86 Pass@1 points on low-difficulty tasks and surpasses explicit GRPO by 4.27 points on high-difficulty tasks while using 3–4 shorter reasoning chains. It also achieves stronger pass@k performance under Gumbel sampling. These results establish Latent-GRPO as an effective approach for stable and efficient latent reasoning.

Large Language Models (LLMs) have achieved impressive results across numerous NLP tasks, and fine-tuning them for Machine Translation (MT) has improved their performance. However, vanilla fine-tuning often leads to catastrophic forgetting, compromising the broad general abilities of LLMs and introducing potential security risks. These abilities, which are developed using proprietary and unavailable training data, make simple data replay methods ineffective. To overcome this issue, we propose a novel approach called Rationale Distillation. RaDis harnesses the strong generative capabilities of LLMs to create rationales for training data, which are then “replayed” to prevent forgetting. These rationales connect prior knowledge with new tasks, acting as self-distillation targets to regulate the training process. By jointly training on reference translations and self-generated rationales, the model can learn new translation skills while preserving its general abilities across other tasks. Additionally, RaDis provides a fresh perspective on using rationales in the CL field and has the potential to serve as a general continual learning method for a variety of tasks.

Direct Preference Optimization (DPO) has become a prominent method for aligning Large Language Models (LLMs) with human preferences. While DPO has enabled significant progress in aligning English LLMs, multilingual preference alignment is hampered by data scarcity. To address this, we propose a novel approach that captures learned preferences from well-aligned English models by implicit rewards and transfers them to other languages through iterative training. Specifically, we derive an implicit reward model from the logits of an English DPO-aligned model and its corresponding reference model. This reward model is then leveraged to annotate preference relations in cross-lingual instruction-following pairs, using English instructions to evaluate multilingual responses. The annotated data is subsequently used for multilingual DPO fine-tuning, facilitating preference knowledge transfer from English to other languages. Fine-tuning Llama3 for two iterations resulted in a 12.72% average improvement in Win Rate and a 5.97% increase in Length Control Win Rate across all training languages on the X-AlpacaEval leaderboard. Our findings demonstrate that leveraging existing English-aligned models can enable efficient and effective multilingual preference alignment, significantly reducing the need for extensive multilingual preference data.

Recent advances in text-only “slow-thinking” reasoning have prompted efforts to transfer this capability to vision-language models (VLMs), for training visual reasoning models (VRMs). However, such transfer faces critical challenges: Effective “slow thinking” in VRMs requires visual reflection, the ability to check the reasoning process based on visual information. Through quantitative analysis, we observe that current VRMs exhibit limited visual reflection, as their attention to visual information diminishes rapidly with longer generated responses. To address this challenge, we propose a new VRM Reflection-V, which enhances visual reflection based on reasoning data construction for cold-start and reward design for reinforcement learning (RL). Firstly, we construct vision-centered reasoning data by leveraging an agent that interacts between VLMs and reasoning LLMs, enabling cold-start learning of visual reflection patterns. Secondly, a visual attention based reward model is employed during RL to encourage reasoning based on visual information. Therefore, Reflection-V demonstrates significant improvements across multiple visual reasoning benchmarks. Furthermore, Reflection-V maintains a stronger and more consistent reliance on visual information during visual reasoning, indicating effective enhancement in visual reflection capabilities.

Ensembling various LLMs to unlock their complementary potential and leverage their individual strengths is highly valuable. Previous studies typically focus on two main paradigms: sample-level and token-level ensembles. Sample-level ensemble methods either select or blend fully generated outputs, which hinders dynamic correction and enhancement of outputs during the generation process. On the other hand, token-level ensemble methods enable real-time correction through fine-grained ensemble at each generation step. However, the information carried by an individual token is quite limited, leading to suboptimal decisions at each step. To address these issues, we propose SweetSpan, a span-level ensemble method that effectively balances the need for real-time adjustments and the information required for accurate ensemble decisions. Our approach involves two key steps: First, we have each candidate model independently generate candidate spans based on the shared prefix. Second, we calculate perplexity scores to facilitate mutual evaluation among the candidate models and achieve robust span selection by filtering out unfaithful scores. To comprehensively evaluate ensemble methods, we propose a new challenging setting (ensemble models with significant performance gaps) in addition to the standard setting (ensemble the best-performing models) to assess the performance of model ensembles in more realistic scenarios. Experimental results in both standard and challenging settings across various language generation tasks demonstrate the effectiveness, robustness, and versatility of our approach compared with previous ensemble methods.

Long-context modeling has drawn more and more attention in the area of Large Language Models (LLMs). Continual training with long-context data becomes the de-facto method to equip LLMs with the ability to process long inputs. However, it still remains an open challenge to measure the quality of long-context training data. To address this issue, we propose a Long-context data selection framework with Attention-based Dependency Measurement (LADM), which can efficiently identify high-quality long-context data from a large-scale, multi-domain pre-training corpus. LADM leverages the retrieval capabilities of the attention mechanism to capture contextual dependencies, ensuring a comprehensive quality measurement of long-context data. Experimental results show that our LADM framework significantly boosts the performance of LLMs on multiple long-context tasks with only 1B tokens for continual training.

In the evolving landscape of Neural Machine Translation (NMT), the pretrain-then-finetune paradigm has yielded impressive results. However, the persistent challenge of Catastrophic Forgetting (CF) remains a hurdle. While previous work has introduced Continual Learning (CL) methods to address CF, these approaches grapple with the delicate balance between avoiding forgetting and maintaining system extensibility. To address this, we propose a CL method, named F-MALLOC (Feed-forward Memory ALLOCation). F-MALLOC is inspired by recent insights highlighting that feed-forward layers emulate neural memories and encapsulate crucial translation knowledge. It decomposes feed-forward layers into discrete memory cells and allocates these memories to different tasks. By learning to allocate and safeguard these memories, our method effectively alleviates CF while ensuring robust extendability. Besides, we propose a comprehensive assessment protocol for multi-stage CL of NMT systems. Experiments conducted following this new protocol showcase the superior performance of F-MALLOC, evidenced by higher BLEU scores and almost zero forgetting.

The recent release of GPT-4o showcased the potential of end-to-end multimodal models, not just in terms of low latency but also in their ability to understand and generate expressive speech with rich emotions. While the details are unknown to the open research community, it likely involves significant amounts of curated data and compute, neither of which is readily accessible. In this paper, we present BLSP-Emo (Bootstrapped Language-Speech Pretraining with Emotion support), a novel approach to developing an end-to-end speech-language model capable of understanding both semantics and emotions in speech and generate empathetic responses. BLSP-Emo utilizes existing speech recognition (ASR) and speech emotion recognition (SER) datasets through a two-stage process. The first stage focuses on semantic alignment, following recent work on pretraining speech-language models using ASR data. The second stage performs emotion alignment with the pretrained speech-language model on an emotion-aware continuation task constructed from SER data. Our experiments demonstrate that the BLSP-Emo model excels in comprehending speech and delivering empathetic responses, both in instruction-following tasks and conversations.