Automatic evaluation of retrieval augmented generation (RAG) systems relies on fine-grained dimensions like faithfulness and relevance, as judged by expert human annotators. Meta-evaluation benchmarks support the development of automatic evaluators that correlate well with human judgement. However, existing benchmarks predominantly focus on English or use translated data, which fails to capture cultural nuances. A native approach provides a better representation of the end user experience.In this work, we develop a Multilingual End-to-end Meta-Evaluation RAG benchmark MEMERAG. Our benchmark builds on the popular MIRACL dataset, using native-language questions and generating responses with diverse large language models (LLMs), which are then assessed by expert annotators for faithfulness and relevance. We describe our annotation process and show that it achieves high inter-annotator agreement. We then analyse the performance of the answer-generating LLMs across languages as per the human evaluators. Finally we apply the dataset to our main use-case which is to benchmark multilingual automatic evaluators (LLM-as-a-judge). We show that our benchmark can reliably identify improvements offered by advanced prompting techniques and LLMs. We release our benchmark to support the community developing accurate evaluation methods for multilingual RAG systems.

Automatic open-domain dialogue evaluation has attracted increasing attention, yet remains challenging due to the complexity of assessing response appropriateness. Traditional evaluation metrics, typically trained with true positive and randomly selected negative responses, tend to assign higher scores to responses that share greater content similarity with contexts. However, adversarial negative responses, despite possessing high lexical overlap with contexts, can be semantically incongruous. Consequently, existing metrics struggle to evaluate such responses effectively, resulting in low correlations with human judgments. While recent studies have demonstrated the effectiveness of Large Language Models (LLMs) for open-domain dialogue evaluation, they still face challenges in handling adversarial negative examples. We propose a novel evaluation framework that integrates Abstract Meaning Representation (AMR) enhanced domain-specific language models (SLMs) with LLMs. Our SLMs explicitly incorporate AMR graph information through a gating mechanism for enhanced semantic representation learning, while both SLM predictions and AMR knowledge are integrated into LLM prompts for robust evaluation. Extensive experiments on open-domain dialogue evaluation tasks demonstrate the superiority of our method compared to state-of-the-art baselines, particularly in discriminating adversarial negative responses. Our framework achieves strong correlations with human judgments across multiple datasets, establishing a new benchmark for dialogue evaluation. Our code and data are publicly available at https://github.com/Bernard-Yang/SIMAMR.

Large Language Models (LLMs) demonstrate remarkable ability to comprehend instructions and generate human-like text, enabling sophisticated agent simulation beyond basic behavior replication. However, the potential for creating freely customisable characters remains underexplored. We introduce the Customisable Conversation Agent Framework, which employs LLMs to simulate real-world characters through personalised characteristic feature injection, enabling diverse character creation according to user preferences.We propose the SimsConv dataset, comprising 68 customised characters and 13,971 multi-turn role-playing dialogues across 1,360 real-world scenes. Characters are initially customised using pre-defined elements (career, aspiration, traits, skills), then expanded through personal and social profiles. Building on this, we present SimsChat, a freely customisable role-playing agent incorporating various realistic settings and topic-specified character interactions.Experimental results on both SimsConv and WikiRoleEval datasets demonstrate SimsChat’s superior performance in maintaining character consistency, knowledge accuracy, and appropriate question rejection compared to existing models. Comprehensive ablation studies validate each component’s contribution to overall performance, with the pre-defined aspects framework and scene construction showing particularly significant impact. Our framework provides valuable insights for developing more accurate and customisable human simulacra.Our data and code are publicly available at https://github.com/Bernard-Yang/SimsChat.

We propose MT2ST, a general and efficient framework for accelerating multi-task training by progressively transitioning to single-task optimization. Unlike conventional multi-task learning (MTL) or single-task fine-tuning (STL), MT2ST dynamically adjusts the training focus via two complementary strategies: Diminish, which gradually down-weights auxiliary losses, and Switch, which explicitly switches to the primary task at a scheduled point. We demonstrate the effectiveness of MT2ST across three key paradigms: representation learning, transformers, and diffusion models, covering both unimodal (text/image) and multimodal (vision-language) tasks. Extensive experiments show that MT2ST significantly improves training efficiency—achieving up to 56% FLOPs compression—while maintaining or surpassing task performance. These results suggest MT2ST as a general-purpose solution for scalable and adaptive multi-task training. Although this work is general-purpose, it is especially suitable for multimodal settings such as VQA or vision-language retrieval, where auxiliary pretraining (e.g., masked language modeling or contrastive learning) often diverges from final objectives. We include a VQA case study and outline its efficiency for multimodal retrieval.

Semantic parsing of long documents remains challenging due to quadratic growth in pairwise composition and memory requirements. We introduce Hierarchical Segment-Graph Memory (HSGM), a novel framework that decomposes an input of length N into M meaningful segments, constructs Local Semantic Graphs on each segment, and extracts compact summary nodes to form a Global Graph Memory. HSGM supports incremental updates—only newly arrived segments incur local graph construction and summary-node integration—while Hierarchical Query Processing locates relevant segments via top-K retrieval over summary nodes and then performs fine-grained reasoning within their local graphs.Theoretically, HSGM reduces worst-case complexity from O(N²) to O\bigl(N\,k + (N/k)²\bigr),with segment size k ≪ N, and we derive Frobenius-norm bounds on the approximation error introduced by node summarization and sparsification thresholds. Empirically, on three benchmarks—long-document AMR parsing, segment-level semantic role labeling (OntoNotes), and legal event extraction—HSGM achieves 2–4× inference speedup, >60% reduction in peak memory, and ≥95% of baseline accuracy. Our approach unlocks scalable, accurate semantic modeling for ultra-long texts, enabling real-time and resource-constrained NLP applications.

Oracle bone script (OBS) is the earliest writing system in China, which is of great value in the improvement of archaeology and Chinese cultural history. However, there are some problems such as the lack of labels and the difficulty to distinguish the glyphs from the background of OBS, which makes the automatic recognition of OBS in the real world not achieve the satisfactory effect. In this paper, we propose a character recognition method based on an unsupervised domain adaptive network (UFCNet). Firstly, a convolutional attention fusion module (CAFM) is designed in the encoder to obtain more global features through multi-layer feature fusion. Second, we construct a Fourier transform (FT) module that focuses on the differences between glyphs and backgrounds. Finally, to further improve the network’s ability to recognize character edges, we introduce a kernel norm-constrained loss function. Extensive experiments perform on the Oracle-241 dataset show that the proposed method is superior to other adaptive methods. The code will be available at https://github.com/zhouynan/UFCNet.

Due to ancient origin, there are many incomplete characters in the unearthed Oracle Bone Inscriptions(OBI), which brings the great challenges to recognition and research. In recent years, image inpainting techniques have made remarkable progress. However, these models are unable to adapt to the unique font shape and complex text background of OBI. To meet these aforementioned challenges, we propose a two-stage method for restoring damaged OBI using Generative Adversarial Networks (GAN), which incorporates a dual discriminator structure to capture both global and local image information. In order to accurately restore the image structure and details, the spatial attention mechanism and a novel loss function are proposed. By feeding clear copies of existing OBI and various types of masks into the network, it learns to generate content for the missing regions. Experimental results demonstrate the effectiveness of our proposed method in completing OBI compared to several state-of-the-art techniques.

Complex named entities (NE), like the titles of creative works, are not simple nouns and pose challenges for NER systems. In the SemEval 2023, Task 2: MultiCoNER II was proposed, whose goal is to recognize complex entities against out of knowledge-base entities and noisy scenarios. To address the challenges posed by MultiCoNER II, our team NetEase.AI proposed an entity recognition system that integrates text error correction system and external knowledge, which can recognize entities in scenes that contain entities out of knowledge base and text with noise. Upon receiving an input sentence, our systems will correct the sentence, extract the entities in the sentence as candidate set using the entity recognition model that incorporates the gazetteer information, and then use the external knowledge to classify the candidate entities to obtain entity type features. Finally, our system fused the multi-dimensional features of the candidate entities into a stacking model, which was used to select the correct entities from the candidate set as the final output. Our system exhibited good noise resistance and excellent entity recognition performance, resulting in our team’s first place victory in the Chinese track of MultiCoNER II.