While model-based verifiers are essential for scaling Reinforcement Learning with Verifiable Rewards (RLVR), current outcome-centric verification paradigms primarily focus on the consistency between the final result and the ground truth, often neglecting potential errors in the derivation process. This leads to assigning positive rewards to correct answers produced from incorrect derivations. To bridge this gap, we introduce PRIME, a benchmark for evaluating verifiers on PRocess-outcome alignment verification In Mathematics and Engineering. Curated from a comprehensive collection of college-level STEM problems, PRIME comprises 2,530 high-difficulty samples through a consistency-based filtering pipeline. Through extensive evaluation, we find that current verifiers frequently fail to detect derivation flaws. Furthermore, we propose a process-aware RLVR training paradigm utilizing verifiers selected via PRIME. This approach substantially outperforms the outcome-only verification baseline, achieving absolute performance gains of 8.29%, 9.12%, and 7.31% on AIME24, AIME25, and Beyond-AIME, respectively, for the Qwen3-14B-Base model. Finally, we demonstrate a strong linear correlation (R² > 0.92) between verifier accuracy on PRIME and RLVR training effectiveness, validating PRIME as a reliable predictor for verifier selection.

Character Coreference Resolution in Movie Screenplays (MovieCoref) is a newly emerging task for understanding complex movie plots and character relationships. This task poses greater challenges than traditional coreference resolution, due to the intricate narrative structures and character interactions unique to screenplays. In light of these challenges, we introduce a novel approach: a Structure-aware Triaffine Parser (STPar) for the MovieCoref task. STPar combines discourse and syntactic structures in the feature encoding process, enabling comprehensive analysis of ternary relationships and complex interactions. During the pairing process, STPar utilizes a triaffine scorer to consider high-order relations between candidate mention pairs, thus enhancing its ability to capture detailed narrative structures. In addition, STPar incorporates multi-task learning, encompassing singleton and span detection tasks, to further improve coreference resolution performance. Our evaluations on the MovieCoref dataset demonstrate that STPar significantly outperforms the best baseline by 7.4%, 21.5%, 7.1%, and 10.2% in F1 scores of B3, CEAFe, LEA, and CoNLL. Further analysis highlights the benefits of integrating structural discourse and syntactic information as well as the combined approaches of triaffine and multi-task learning.1

Long-context modeling is crucial for next-generation language models, yet the high computational cost of standard attention mechanisms poses significant computational challenges. Sparse attention offers a promising direction for improving efficiency while maintaining model capabilities. We present NSA, a Natively trained Sparse Attention mechanism that integrates algorithmic innovations with hardware-aligned optimizations to achieve efficient long-context modeling. NSA employs a dynamic hierarchical sparse strategy, combining coarse-grained token compression with fine-grained token selection to preserve both global context awareness and local precision. Our approach advances sparse attention design with two key innovations: (1) We achieve substantial speedups through arithmetic intensity-balanced algorithm design, with implementation optimizations for modern hardware. (2) We enable end-to-end training, reducing pretraining computation without sacrificing model performance. As shown in Figure 1, experiments show the model pretrained with NSA maintains or exceeds Full Attention models across general benchmarks, long-context tasks, and instruction-based reasoning. Meanwhile, NSA achieves substantial speedups over Full Attention on 64k-length sequences across decoding, forward propagation, and backward propagation, validating its efficiency throughout the model lifecycle.

In-context learning has emerged as a groundbreaking ability of Large Language Models (LLMs) and revolutionized various fields by providing a few task-relevant demonstrations in the prompt. However, trustworthy issues with LLM’s response, such as hallucination, have also been actively discussed. Existing works have been devoted to quantifying the uncertainty in LLM’s response, but they often overlook the complex nature of LLMs and the uniqueness of in-context learning. In this work, we delve into the predictive uncertainty of LLMs associated with in-context learning, highlighting that such uncertainties may stem from both the provided demonstrations (aleatoric uncertainty) and ambiguities tied to the model’s configurations (epistemic uncertainty). We propose a novel formulation and corresponding estimation method to quantify both types of uncertainties. The proposed method offers an unsupervised way to understand the prediction of in-context learning in a plug-and-play fashion. Extensive experiments are conducted to demonstrate the effectiveness of the decomposition. The code and data are available at: https://github.com/lingchen0331/UQ_ICL.

“本技术报告是对2024CCL评测任务(中小学作文修辞识别与理解评测)的一种解决方案。在中小学生的学习过程中,修辞手法不仅是阅读理解和写作技巧的核心组成部分,同时也是塑造优秀文学作品的不可或缺的元素。识别并理解学生作文中的修辞使用,可以帮助学生提高作文表达能力,指导学生更高质量的叙述和描写。对修辞的识别目前属于自然理解领域比较困难的任务,因为需要用到人类领域的大量先验知识,而且很多时候不同的修辞之间的边界还是模糊的。我们通过lora技术直接微调基于qwen-chat-7B的大语言预训练模型,来进行修辞类别的识别。我们的主要创新技术点为:基于相同的输入输出数据来构造多条训练数据提升算法表现;分级分层来进行修辞的判断,先进行大的修辞类别判断,再把大的修辞类别做为输入对修辞的子类别进行判断;针对修辞成分抽取的任务,直接输出对应的结果文本,再对应回原文本进行位置检索,而不是直接输出索引下标。”

Though advanced in understanding visual information with human languages, Large Vision-Language Models (LVLMs) still suffer from multimodal hallucinations. A natural concern is that during multimodal interaction, the generated hallucinations could influence the LVLMs’ subsequent generation. Thus, we raise a question: When presented with a query relevant to the previously generated hallucination, will LVLMs be misled and respond incorrectly, even though the ground visual information exists? To answer this, we propose a framework called \\textitMMHalSnowball to evaluate LVLMs’ behaviors when encountering generated hallucinations, where LVLMs are required to answer specific visual questions within a curated hallucinatory conversation. Crucially, our experiment shows that the performance of open-source LVLMs drops by at least 31\\%, indicating that LVLMs are prone to accept the generated hallucinations and make false claims that they would not have supported without distractions. We term this Multimodal Hallucination Snowballing. To mitigate this issue, we further propose a training-free method called Residual Visual Decoding, where we revise the output distribution of LVLMs with the one derived from the residual visual input, providing models with direct access to the visual information. Experiments show that our method can mitigate more than 24\\% of the snowballed multimodal hallucination while maintaining capabilities.

Open-ended Commonsense Reasoning is defined as solving a commonsense question without providing 1) a short list of answer candidates and 2) a pre-defined answer scope. Conventional ways of formulating the commonsense question into a question-answering form or utilizing external knowledge to learn retrieval-based methods are less applicable in the open-ended setting due to an inherent challenge. Without pre-defining an answer scope or a few candidates, open-ended commonsense reasoning entails predicting answers by searching over an extremely large searching space. Moreover, most questions require implicit multi-hop reasoning, which presents even more challenges to our problem. In this work, we leverage pre-trained language models to iteratively retrieve reasoning paths on the external knowledge base, which does not require task-specific supervision. The reasoning paths can help to identify the most precise answer to the commonsense question. We conduct experiments on two commonsense benchmark datasets. Compared to other approaches, our proposed method achieves better performance both quantitatively and qualitatively.

In the present paper, we introduce the ManDi Corpus, a spoken corpus of regional Mandarin dialects and Standard Mandarin. The corpus currently contains 357 recordings (about 9.6 hours) of monosyllabic words, disyllabic words, short sentences, a short passage and a poem, each produced in Standard Mandarin and in one of six regional Mandarin dialects: Beijing, Chengdu, Jinan, Taiyuan, Wuhan, and Xi’an Mandarin from 36 speakers. The corpus was collected remotely using participant-controlled smartphone recording apps. Word- and phone-level alignments were generated using Praat and the Montreal Forced Aligner. The pilot study of dialect-specific tone systems showed that with practicable design and decent recording quality, remotely collected speech data can be suitable for analysis of relative patterns in acoustic-phonetic realization. The corpus is available on OSF (https://osf.io/fgv4w/) for non-commercial use under a CC BY-NC 3.0 license.

Event extraction (EE) is an essential task of information extraction, which aims to extract structured event information from unstructured text. Most prior work focuses on extracting flat events while neglecting overlapped or nested ones. A few models for overlapped and nested EE includes several successive stages to extract event triggers and arguments,which suffer from error propagation. Therefore, we design a simple yet effective tagging scheme and model to formulate EE as word-word relation recognition, called OneEE. The relations between trigger or argument words are simultaneously recognized in one stage with parallel grid tagging, thus yielding a very fast event extraction speed. The model is equipped with an adaptive event fusion module to generate event-aware representations and a distance-aware predictor to integrate relative distance information for word-word relation recognition, which are empirically demonstrated to be effective mechanisms. Experiments on 3 overlapped and nested EE benchmarks, namely FewFC, Genia11, and Genia13, show that OneEE achieves the state-of-the-art (SOTA) results. Moreover, the inference speed of OneEE is faster than those of baselines in the same condition, and can be further substantially improved since it supports parallel inference.

In this paper, we present an approach for sentiment analysis in code-mixed language on twitter defined in SemEval-2020 Task 9. Our team (referred as LiangZhao) employ different multilingual models with weighted loss focused on complexity of code-mixing in sentence, in which the best model achieved f1-score of 0.806 and ranked 1st of subtask- Sentimix Spanglish. The performance of method is analyzed and each component of our architecture is demonstrated.

Recent work has shown that current text classification models are fragile and sensitive to simple perturbations. In this work, we propose a novel adversarial training approach, LexicalAT, to improve the robustness of current classification models. The proposed approach consists of a generator and a classifier. The generator learns to generate examples to attack the classifier while the classifier learns to defend these attacks. Considering the diversity of attacks, the generator uses a large-scale lexical knowledge base, WordNet, to generate attacking examples by replacing some words in training examples with their synonyms (e.g., sad and unhappy), neighbor words (e.g., fox and wolf), or super-superior words (e.g., chair and armchair). Due to the discrete generation step in the generator, we use policy gradient, a reinforcement learning approach, to train the two modules. Experiments show LexicalAT outperforms strong baselines and reduces test errors on various neural networks, including CNN, RNN, and BERT.

Compositional generalization is a basic mechanism in human language learning, but current neural networks lack such ability. In this paper, we conduct fundamental research for encoding compositionality in neural networks. Conventional methods use a single representation for the input sentence, making it hard to apply prior knowledge of compositionality. In contrast, our approach leverages such knowledge with two representations, one generating attention maps, and the other mapping attended input words to output symbols. We reduce the entropy in each representation to improve generalization. Our experiments demonstrate significant improvements over the conventional methods in five NLP tasks including instruction learning and machine translation. In the SCAN domain, it boosts accuracies from 14.0% to 98.8% in Jump task, and from 92.0% to 99.7% in TurnLeft task. It also beats human performance on a few-shot learning task. We hope the proposed approach can help ease future research towards human-level compositional language learning.

To deploy a spoken language understanding (SLU) model to a new language, language transferring is desired to avoid the trouble of acquiring and labeling a new big SLU corpus. An SLU corpus is a monolingual corpus with domain/intent/slot labels. Translating the original SLU corpus into the target language is an attractive strategy. However, SLU corpora consist of plenty of semantic labels (slots), which general-purpose translators cannot handle well, not to mention additional culture differences. This paper focuses on the language transferring task given a small in-domain parallel SLU corpus. The in-domain parallel corpus can be used as the first adaptation on the general translator. But more importantly, we show how to use reinforcement learning (RL) to further adapt the adapted translator, where translated sentences with more proper slot tags receive higher rewards. Our reward is derived from the source input sentence exclusively, unlike reward via actor-critical methods or computing reward with a ground truth target sentence. Hence we can adapt the translator the second time, using the big monolingual SLU corpus from the source language. We evaluate our approach on Chinese to English language transferring for SLU systems. The experimental results show that the generated English SLU corpus via adaptation and reinforcement learning gives us over 97% in the slot F1 score and over 84% accuracy in domain classification. It demonstrates the effectiveness of the proposed language transferring method. Compared with naive translation, our proposed method improves domain classification accuracy by relatively 22%, and the slot filling F1 score by relatively more than 71%.