Kewei Tu

Songlin Yang, Wei Liu, Kewei Tu · mit

Hidden Markov Models (HMMs) and Probabilistic Context-Free Grammars (PCFGs) are widely used structured models, both of which can be represented as factor graph grammars (FGGs), a powerful formalism capable of describing a wide range of models. Recent research found it beneficial to use large state spaces for HMMs and PCFGs. However, inference with large state spaces is computationally demanding, especially for PCFGs. To tackle this challenge, we leverage tensor rank decomposition (aka.\ CPD) to decrease inference computational complexities for a subset of FGGs subsuming HMMs and PCFGs. We apply CPD on the factors of an FGG and then construct a new FGG defined in the rank space. Inference with the new FGG produces the same result but has a lower time complexity when the rank size is smaller than the state size. We conduct experiments on HMM language modeling and unsupervised PCFG parsing, showing better performance than previous work. Our code is publicly available at \url{https://github.com/VPeterV/RankSpace-Models}.

Tianyu Yu, Chengyue Jiang, Chao Lou et al.

Large language models (LLMs) have shown impressive ability for open-domain NLP tasks. However, LLMs are sometimes too footloose for natural language understanding (NLU) tasks which always have restricted output and input format. Their performances on NLU tasks are highly related to prompts or demonstrations and are shown to be poor at performing several representative NLU tasks, such as event extraction and entity typing. To this end, we present SeqGPT, a bilingual (i.e., English and Chinese) open-source autoregressive model specially enhanced for open-domain natural language understanding. We express all NLU tasks with two atomic tasks, which define fixed instructions to restrict the input and output format but still ``open'' for arbitrarily varied label sets. The model is first instruction-tuned with extremely fine-grained labeled data synthesized by ChatGPT and then further fine-tuned by 233 different atomic tasks from 152 datasets across various domains. The experimental results show that SeqGPT has decent classification and extraction ability, and is capable of performing language understanding tasks on unseen domains. We also conduct empirical studies on the scaling of data and model size as well as on the transfer across tasks. Our model is accessible at https://github.com/Alibaba-NLP/SeqGPT.

Chao Lou, Songlin Yang, Kewei Tu · mit

Nested named entity recognition (NER) has been receiving increasing attention. Recently, (Fu et al, 2021) adapt a span-based constituency parser to tackle nested NER. They treat nested entities as partially-observed constituency trees and propose the masked inside algorithm for partial marginalization. However, their method cannot leverage entity heads, which have been shown useful in entity mention detection and entity typing. In this work, we resort to more expressive structures, lexicalized constituency trees in which constituents are annotated by headwords, to model nested entities. We leverage the Eisner-Satta algorithm to perform partial marginalization and inference efficiently. In addition, we propose to use (1) a two-stage strategy (2) a head regularization loss and (3) a head-aware labeling loss in order to enhance the performance. We make a thorough ablation study to investigate the functionality of each component. Experimentally, our method achieves the state-of-the-art performance on ACE2004, ACE2005 and NNE, and competitive performance on GENIA, and meanwhile has a fast inference speed.

Zhaohui Yan, Songlin Yang, Wei Liu et al. · mit

Entity and Relation Extraction (ERE) is an important task in information extraction. Recent marker-based pipeline models achieve state-of-the-art performance, but still suffer from the error propagation issue. Also, most of current ERE models do not take into account higher-order interactions between multiple entities and relations, while higher-order modeling could be beneficial.In this work, we propose HyperGraph neural network for ERE ($\hgnn{}$), which is built upon the PL-marker (a state-of-the-art marker-based pipleline model). To alleviate error propagation,we use a high-recall pruner mechanism to transfer the burden of entity identification and labeling from the NER module to the joint module of our model. For higher-order modeling, we build a hypergraph, where nodes are entities (provided by the span pruner) and relations thereof, and hyperedges encode interactions between two different relations or between a relation and its associated subject and object entities. We then run a hypergraph neural network for higher-order inference by applying message passing over the built hypergraph. Experiments on three widely used benchmarks (\acef{}, \ace{} and \scierc{}) for ERE task show significant improvements over the previous state-of-the-art PL-marker.

Chengyue Jiang, Yong Jiang, Weiqi Wu et al.

Ultra-fine entity typing (UFET) aims to predict a wide range of type phrases that correctly describe the categories of a given entity mention in a sentence. Most recent works infer each entity type independently, ignoring the correlations between types, e.g., when an entity is inferred as a president, it should also be a politician and a leader. To this end, we use an undirected graphical model called pairwise conditional random field (PCRF) to formulate the UFET problem, in which the type variables are not only unarily influenced by the input but also pairwisely relate to all the other type variables. We use various modern backbones for entity typing to compute unary potentials, and derive pairwise potentials from type phrase representations that both capture prior semantic information and facilitate accelerated inference. We use mean-field variational inference for efficient type inference on very large type sets and unfold it as a neural network module to enable end-to-end training. Experiments on UFET show that the Neural-PCRF consistently outperforms its backbones with little cost and results in a competitive performance against cross-encoder based SOTA while being thousands of times faster. We also find Neural- PCRF effective on a widely used fine-grained entity typing dataset with a smaller type set. We pack Neural-PCRF as a network module that can be plugged onto multi-label type classifiers with ease and release it in https://github.com/modelscope/adaseq/tree/master/examples/NPCRF.

Haoyi Wu, Wenyang Hui, Yezeng Chen et al.

Mathematical understanding and reasoning are crucial tasks for assessing the capabilities of artificial intelligence (AI). However, existing benchmarks either require just a few steps of reasoning, or only contain a small amount of data in one specific topic, making it hard to analyse AI's behaviour with reference to different problems within a specific topic in detail. In this work, we propose Conic10K, a challenging math problem dataset on conic sections in Chinese senior high school education. Our dataset contains various problems with different reasoning depths, while only the knowledge from conic sections is required. Since the dataset only involves a narrow range of knowledge, it is easy to separately analyse the knowledge a model possesses and the reasoning ability it has. For each problem, we provide a high-quality formal representation, the reasoning steps, and the final solution. Experiments show that existing large language models, including GPT-4, exhibit weak performance on complex reasoning. We hope that our findings could inspire more advanced techniques for precise natural language understanding and reasoning. Our dataset and codes are available at https://github.com/whyNLP/Conic10K.

Chengyue Jiang, Wenyang Hui, Yong Jiang et al.

Ultra-fine entity typing (UFET) predicts extremely free-formed types (e.g., president, politician) of a given entity mention (e.g., Joe Biden) in context. State-of-the-art (SOTA) methods use the cross-encoder (CE) based architecture. CE concatenates the mention (and its context) with each type and feeds the pairs into a pretrained language model (PLM) to score their relevance. It brings deeper interaction between mention and types to reach better performance but has to perform N (type set size) forward passes to infer types of a single mention. CE is therefore very slow in inference when the type set is large (e.g., N = 10k for UFET). To this end, we propose to perform entity typing in a recall-expand-filter manner. The recall and expand stages prune the large type set and generate K (K is typically less than 256) most relevant type candidates for each mention. At the filter stage, we use a novel model called MCCE to concurrently encode and score these K candidates in only one forward pass to obtain the final type prediction. We investigate different variants of MCCE and extensive experiments show that MCCE under our paradigm reaches SOTA performance on ultra-fine entity typing and is thousands of times faster than the cross-encoder. We also found MCCE is very effective in fine-grained (130 types) and coarse-grained (9 types) entity typing. Our code is available at \url{https://github.com/modelscope/AdaSeq/tree/master/examples/MCCE}.

Wei Liu, Songlin Yang, Yoon Kim et al. · mit

Scaling dense PCFGs to thousands of nonterminals via a low-rank parameterization of the rule probability tensor has been shown to be beneficial for unsupervised parsing. However, PCFGs scaled this way still perform poorly as a language model, and even underperform similarly-sized HMMs. This work introduces \emph{SimplePCFG}, a simple PCFG formalism with independent left and right productions. Despite imposing a stronger independence assumption than the low-rank approach, we find that this formalism scales more effectively both as a language model and as an unsupervised parser. As an unsupervised parser, our simple PCFG obtains an average F1 of 65.1 on the English PTB, and as a language model, it obtains a perplexity of 119.0, outperforming similarly-sized low-rank PCFGs. We further introduce \emph{FlashInside}, a hardware IO-aware implementation of the inside algorithm for efficiently scaling simple PCFGs.

Chengyue Jiang, Yong Jiang, Weiqi Wu et al.

Open knowledge graph (KG) consists of (subject, relation, object) triples extracted from millions of raw text. The subject and object noun phrases and the relation in open KG have severe redundancy and ambiguity and need to be canonicalized. Existing datasets for open KG canonicalization only provide gold entity-level canonicalization for noun phrases. In this paper, we present COMBO, a Complete Benchmark for Open KG canonicalization. Compared with existing datasets, we additionally provide gold canonicalization for relation phrases, gold ontology-level canonicalization for noun phrases, as well as source sentences from which triples are extracted. We also propose metrics for evaluating each type of canonicalization. On the COMBO dataset, we empirically compare previously proposed canonicalization methods as well as a few simple baseline methods based on pretrained language models. We find that properly encoding the phrases in a triple using pretrained language models results in better relation canonicalization and ontology-level canonicalization of the noun phrase. We release our dataset, baselines, and evaluation scripts at https://github.com/jeffchy/COMBO/tree/main.

Xinyu Wang, Jiong Cai, Yong Jiang et al.

Multi-modal named entity recognition (NER) and relation extraction (RE) aim to leverage relevant image information to improve the performance of NER and RE. Most existing efforts largely focused on directly extracting potentially useful information from images (such as pixel-level features, identified objects, and associated captions). However, such extraction processes may not be knowledge aware, resulting in information that may not be highly relevant. In this paper, we propose a novel Multi-modal Retrieval based framework (MoRe). MoRe contains a text retrieval module and an image-based retrieval module, which retrieve related knowledge of the input text and image in the knowledge corpus respectively. Next, the retrieval results are sent to the textual and visual models respectively for predictions. Finally, a Mixture of Experts (MoE) module combines the predictions from the two models to make the final decision. Our experiments show that both our textual model and visual model can achieve state-of-the-art performance on four multi-modal NER datasets and one multi-modal RE dataset. With MoE, the model performance can be further improved and our analysis demonstrates the benefits of integrating both textual and visual cues for such tasks.

Yida Zhao, Chao Lou, Kewei Tu

Syntactic Transformer language models aim to achieve better generalization through simultaneously modeling syntax trees and sentences. While prior work has been focusing on adding constituency-based structures to Transformers, we introduce Dependency Transformer Grammars (DTGs), a new class of Transformer language model with explicit dependency-based inductive bias. DTGs simulate dependency transition systems with constrained attention patterns by modifying attention masks, incorporate the stack information through relative positional encoding, and augment dependency arc representation with a combination of token embeddings and operation embeddings. When trained on a dataset of sentences annotated with dependency trees, DTGs achieve better generalization while maintaining comparable perplexity with Transformer language model baselines. DTGs also outperform recent constituency-based models, showing that dependency can better guide Transformer language models. Our code is released at https://github.com/zhaoyd1/Dep_Transformer_Grammars.

Songlin Yang, Kewei Tu

Second-order semantic parsing with end-to-end mean-field inference has been shown good performance. In this work we aim to improve this method by modeling label correlations between adjacent arcs. However, direct modeling leads to memory explosion because second-order score tensors have sizes of $O(n^3L^2)$ ($n$ is the sentence length and $L$ is the number of labels), which is not affordable. To tackle this computational challenge, we leverage tensor decomposition techniques, and interestingly, we show that the large second-order score tensors have no need to be materialized during mean-field inference, thereby reducing the computational complexity from cubic to quadratic. We conduct experiments on SemEval 2015 Task 18 English datasets, showing the effectiveness of modeling label correlations. Our code is publicly available at https://github.com/sustcsonglin/mean-field-dep-parsing.

Xinyu Wang, Yongliang Shen, Jiong Cai et al.

The MultiCoNER shared task aims at detecting semantically ambiguous and complex named entities in short and low-context settings for multiple languages. The lack of contexts makes the recognition of ambiguous named entities challenging. To alleviate this issue, our team DAMO-NLP proposes a knowledge-based system, where we build a multilingual knowledge base based on Wikipedia to provide related context information to the named entity recognition (NER) model. Given an input sentence, our system effectively retrieves related contexts from the knowledge base. The original input sentences are then augmented with such context information, allowing significantly better contextualized token representations to be captured. Our system wins 10 out of 13 tracks in the MultiCoNER shared task.

Junjie Peng, You Wu, Haoyi Wu et al.

Large language models (LLMs) with extended context lengths rely on the key-value (KV) cache to support attention over prior tokens. However, maintaining the KV cache incurs substantial memory overhead, motivating KV-cache compression methods that enforce a fixed budget through eviction and merging. Modern eviction methods increasingly adopt span-based retention because preserving contiguous spans is empirically effective and better preserves semantic coherence. Yet, when combined with post-eviction merging, span-based retention concentrates merges onto a small set of span-boundary carrier tokens, producing a highly imbalanced merge pattern that exacerbates over-merging and increases information loss. To address this imbalance, we propose GRKV (Global Regression for KV Cache), a training-free KV-cache merging method that directly minimizes the discrepancy between compressed-cache and full-cache attention outputs. GRKV uses ridge-regression-based merge steps to distribute information from evicted tokens across retained tokens, while regularizing the updates to prevent over-smoothing. Across the LongBench and RULER long-context benchmarks, GRKV is the only merging method that improves overall performance with minimal overhead.

Weiqi Wu, Chengyue Jiang, Yong Jiang et al.

Ontological knowledge, which comprises classes and properties and their relationships, is integral to world knowledge. It is significant to explore whether Pretrained Language Models (PLMs) know and understand such knowledge. However, existing PLM-probing studies focus mainly on factual knowledge, lacking a systematic probing of ontological knowledge. In this paper, we focus on probing whether PLMs store ontological knowledge and have a semantic understanding of the knowledge rather than rote memorization of the surface form. To probe whether PLMs know ontological knowledge, we investigate how well PLMs memorize: (1) types of entities; (2) hierarchical relationships among classes and properties, e.g., Person is a subclass of Animal and Member of Sports Team is a subproperty of Member of ; (3) domain and range constraints of properties, e.g., the subject of Member of Sports Team should be a Person and the object should be a Sports Team. To further probe whether PLMs truly understand ontological knowledge beyond memorization, we comprehensively study whether they can reliably perform logical reasoning with given knowledge according to ontological entailment rules. Our probing results show that PLMs can memorize certain ontological knowledge and utilize implicit knowledge in reasoning. However, both the memorizing and reasoning performances are less than perfect, indicating incomplete knowledge and understanding.

Tianyu Huang, Yida Zhao, Chuyan Zhou et al.

Augmenting Transformers with linguistic structures effectively enhances the syntactic generalization performance of language models. Previous work in this direction focuses on syntactic tree structures of languages, in particular constituency tree structures. We propose Graph-Infused Layers Transformer Language Model (GiLT) which leverages dependency graphs for augmenting Transformer language models. Unlike most previous work, GiLT does not insert extra structural tokens in language modeling; instead, it injects structural information into language modeling by modulating attention weights in the Transformer with features extracted from the dependency graph that is incrementally constructed along with token prediction. In our experiments, GiLT with semantic dependency graphs achieves better syntactic generalization while maintaining competitive perplexity in comparison with Transformer language model baselines. In addition, GiLT can be finetuned from a pretrained language model to achieve improved downstream task performance. Our code is released at https://github.com/cookie-pie-oops/GiLT-LM.

Xiang Hu, Qingyang Zhu, Kewei Tu et al.

We present ReCAT, a recursive composition augmented Transformer that is able to explicitly model hierarchical syntactic structures of raw texts without relying on gold trees during both learning and inference. Existing research along this line restricts data to follow a hierarchical tree structure and thus lacks inter-span communications. To overcome the problem, we propose a novel contextual inside-outside (CIO) layer that learns contextualized representations of spans through bottom-up and top-down passes, where a bottom-up pass forms representations of high-level spans by composing low-level spans, while a top-down pass combines information inside and outside a span. By stacking several CIO layers between the embedding layer and the attention layers in Transformer, the ReCAT model can perform both deep intra-span and deep inter-span interactions, and thus generate multi-grained representations fully contextualized with other spans. Moreover, the CIO layers can be jointly pre-trained with Transformers, making ReCAT enjoy scaling ability, strong performance, and interpretability at the same time. We conduct experiments on various sentence-level and span-level tasks. Evaluation results indicate that ReCAT can significantly outperform vanilla Transformer models on all span-level tasks and baselines that combine recursive networks with Transformers on natural language inference tasks. More interestingly, the hierarchical structures induced by ReCAT exhibit strong consistency with human-annotated syntactic trees, indicating good interpretability brought by the CIO layers.

Xiang Hu, Xinyu Kong, Kewei Tu

Deep neural networks based on layer-stacking architectures have historically suffered from poor inherent interpretability. Meanwhile, symbolic probabilistic models function with clear interpretability, but how to combine them with neural networks to enhance their performance remains to be explored. In this paper, we try to marry these two systems for text classification via a structured language model. We propose a Symbolic-Neural model that can learn to explicitly predict class labels of text spans from a constituency tree without requiring any access to span-level gold labels. As the structured language model learns to predict constituency trees in a self-supervised manner, only raw texts and sentence-level labels are required as training data, which makes it essentially a general constituent-level self-interpretable classification model. Our experiments demonstrate that our approach could achieve good prediction accuracy in downstream tasks. Meanwhile, the predicted span labels are consistent with human rationales to a certain degree.

Jiong Cai, Yong Jiang, Yue Zhang et al.

Discovering the intended items of user queries from a massive repository of items is one of the main goals of an e-commerce search system. Relevance prediction is essential to the search system since it helps improve performance. When online serving a relevance model, the model is required to perform fast and accurate inference. Currently, the widely used models such as Bi-encoder and Cross-encoder have their limitations in accuracy or inference speed respectively. In this work, we propose a novel model called the Entity-Based Relevance Model (EBRM). We identify the entities contained in an item and decompose the QI (query-item) relevance problem into multiple QE (query-entity) relevance problems; we then aggregate their results to form the QI prediction using a soft logic formulation. The decomposition allows us to use a Cross-encoder QE relevance module for high accuracy as well as cache QE predictions for fast online inference. Utilizing soft logic makes the prediction procedure interpretable and intervenable. We also show that pretraining the QE module with auto-generated QE data from user logs can further improve the overall performance. The proposed method is evaluated on labeled data from e-commerce websites. Empirical results show that it achieves promising improvements with computation efficiency.

Haoyi Wu, Kewei Tu

Syntactic structures used to play a vital role in natural language processing (NLP), but since the deep learning revolution, NLP has been gradually dominated by neural models that do not consider syntactic structures in their design. One vastly successful class of neural models is transformers. When used as an encoder, a transformer produces contextual representation of words in the input sentence. In this work, we propose a new model of contextual word representation, not from a neural perspective, but from a purely syntactic and probabilistic perspective. Specifically, we design a conditional random field that models discrete latent representations of all words in a sentence as well as dependency arcs between them; and we use mean field variational inference for approximate inference. Strikingly, we find that the computation graph of our model resembles transformers, with correspondences between dependencies and self-attention and between distributions over latent representations and contextual embeddings of words. Experiments show that our model performs competitively to transformers on small to medium sized datasets. We hope that our work could help bridge the gap between traditional syntactic and probabilistic approaches and cutting-edge neural approaches to NLP, and inspire more linguistically-principled neural approaches in the future.

Zixia Jia, Zhaohui Yan, Wenjuan Han et al.

Prior works on joint Information Extraction (IE) typically model instance (e.g., event triggers, entities, roles, relations) interactions by representation enhancement, type dependencies scoring, or global decoding. We find that the previous models generally consider binary type dependency scoring of a pair of instances, and leverage local search such as beam search to approximate global solutions. To better integrate cross-instance interactions, in this work, we introduce a joint IE framework (CRFIE) that formulates joint IE as a high-order Conditional Random Field. Specifically, we design binary factors and ternary factors to directly model interactions between not only a pair of instances but also triplets. Then, these factors are utilized to jointly predict labels of all instances. To address the intractability problem of exact high-order inference, we incorporate a high-order neural decoder that is unfolded from a mean-field variational inference method, which achieves consistent learning and inference. The experimental results show that our approach achieves consistent improvements on three IE tasks compared with our baseline and prior work.

Chao Lou, Kewei Tu

Translation-based AMR parsers have recently gained popularity due to their simplicity and effectiveness. They predict linearized graphs as free texts, avoiding explicit structure modeling. However, this simplicity neglects structural locality in AMR graphs and introduces unnecessary tokens to represent coreferences. In this paper, we introduce new target forms of AMR parsing and a novel model, CHAP, which is equipped with causal hierarchical attention and the pointer mechanism, enabling the integration of structures into the Transformer decoder. We empirically explore various alternative modeling options. Experiments show that our model outperforms baseline models on four out of five benchmarks in the setting of no additional data.

Chaoyi Ai, Yong Jiang, Shen Huang et al.

Named entity recognition (NER) models often struggle with noisy inputs, such as those with spelling mistakes or errors generated by Optical Character Recognition processes, and learning a robust NER model is challenging. Existing robust NER models utilize both noisy text and its corresponding gold text for training, which is infeasible in many real-world applications in which gold text is not available. In this paper, we consider a more realistic setting in which only noisy text and its NER labels are available. We propose to retrieve relevant text of the noisy text from a knowledge corpus and use it to enhance the representation of the original noisy input. We design three retrieval methods: sparse retrieval based on lexicon similarity, dense retrieval based on semantic similarity, and self-retrieval based on task-specific text. After retrieving relevant text, we concatenate the retrieved text with the original noisy text and encode them with a transformer network, utilizing self-attention to enhance the contextual token representations of the noisy text using the retrieved text. We further employ a multi-view training framework that improves robust NER without retrieving text during inference. Experiments show that our retrieval-augmented model achieves significant improvements in various noisy NER settings.

Chao Lou, Kewei Tu

Neural QCFG is a grammar-based sequence-tosequence (seq2seq) model with strong inductive biases on hierarchical structures. It excels in interpretability and generalization but suffers from expensive inference. In this paper, we study two low-rank variants of Neural QCFG for faster inference with different trade-offs between efficiency and expressiveness. Furthermore, utilizing the symbolic interface provided by the grammar, we introduce two soft constraints over tree hierarchy and source coverage. We experiment with various datasets and find that our models outperform vanilla Neural QCFG in most settings.

Penghao Kuang, Haoyi Wu, Kewei Tu

Probabilistic Transformer (PT), a white-box probabilistic model for contextual word representation, has demonstrated substantial similarity to standard Transformers in both computational structure and downstream task performance on small models and small to medium sized datasets. However, PT is less robust to hyperparameter choices than standard Transformers, making it harder to scale efficiently. In this work, we follow Maximal Update Parametrization (muP) to rescale PT's parameters, so that hyperparameters optimized on small models can be transferred to larger models without additional tuning. With this approach, we successfully scale PT to models with up to 0.4B parameters. Experiments show that PT consistently outperforms standard transformer under the same parameter budget on Masked Language Modeling (MLM) tasks. We hope this work will contribute to the practical deployment of probabilistic models at substantially larger scales in the future.

Haoyi Wu, Kewei Tu

Huge memory consumption has been a major bottleneck for deploying high-throughput large language models in real-world applications. In addition to the large number of parameters, the key-value (KV) cache for the attention mechanism in the transformer architecture consumes a significant amount of memory, especially when the number of layers is large for deep language models. In this paper, we propose a novel method that only computes and caches the KVs of a small number of layers, thus significantly saving memory consumption and improving inference throughput. Our experiments on large language models show that our method achieves up to 26$\times$ higher throughput than standard transformers and competitive performance in language modeling and downstream tasks. In addition, our method is orthogonal to existing transformer memory-saving techniques, so it is straightforward to integrate them with our model, achieving further improvement in inference efficiency. Our code is available at https://github.com/whyNLP/LCKV.

Zhuo Chen, Xinyu Geng, Xinyu Wang et al.

Visual Question-Answering (VQA) is a challenging multimodal task that requires integrating visual and textual information to generate accurate responses. While multimodal Retrieval-Augmented Generation (mRAG) has shown promise in enhancing VQA systems by providing more evidence on both image and text sides, the default procedure that addresses VQA queries, especially the knowledge-intensive ones, often relies on multi-stage pipelines of mRAG with inherent dependencies. To mitigate the inefficiency limitations while maintaining VQA task performance, this paper proposes a method that trains a multimodal planning agent, dynamically decomposing the mRAG pipeline to solve the VQA task. Our method optimizes the trade-off between efficiency and effectiveness by training the agent to intelligently determine the necessity of each mRAG step. In our experiments, the agent can help reduce redundant computations, cutting search time by over 60\% compared to existing methods and decreasing costly tool calls. Meanwhile, experiments demonstrate that our method outperforms all baselines, including a Deep Research agent and a carefully designed prompt-based method, on average over six various datasets. Code will be released.

Haoyi Wu, Zhihao Teng, Kewei Tu

Continuous chain-of-thought has been shown to be effective in saving reasoning tokens for large language models. By reasoning with continuous latent thought tokens, continuous CoT is able to perform implicit reasoning in a compact manner. However, the sequential dependencies between latent thought tokens spoil parallel training, leading to long training time. In this paper, we propose Parallel Continuous Chain-of-Thought (PCCoT), which performs Jacobi iteration on the latent thought tokens, updating them iteratively in parallel instead of sequentially and thus improving both training and inference efficiency of continuous CoT. Experiments demonstrate that by choosing the proper number of iterations, we are able to achieve comparable or even better performance while saving nearly 50% of the training and inference time. Moreover, PCCoT shows better stability and robustness in the training process. Our code is available at https://github.com/whyNLP/PCCoT.

Zhuo Chen, Xinyu Wang, Yong Jiang et al.

Despite the advancements made in Vision Large Language Models (VLLMs), like text Large Language Models (LLMs), they have limitations in addressing questions that require real-time information or are knowledge-intensive. Indiscriminately adopting Retrieval Augmented Generation (RAG) techniques is an effective yet expensive way to enable models to answer queries beyond their knowledge scopes. To mitigate the dependence on retrieval and simultaneously maintain, or even improve, the performance benefits provided by retrieval, we propose a method to detect the knowledge boundary of VLLMs, allowing for more efficient use of techniques like RAG. Specifically, we propose a method with two variants that fine-tune a VLLM on an automatically constructed dataset for boundary identification. Experimental results on various types of Visual Question Answering datasets show that our method successfully depicts a VLLM's knowledge boundary, based on which we are able to reduce indiscriminate retrieval while maintaining or improving the performance. In addition, we show that the knowledge boundary identified by our method for one VLLM can be used as a surrogate boundary for other VLLMs. Code will be released at https://github.com/Chord-Chen-30/VLLM-KnowledgeBoundary

Zhuo Chen, Chengyue Jiang, Kewei Tu

In the age of neural natural language processing, there are plenty of works trying to derive interpretations of neural models. Intuitively, when gold rationales exist during training, one can additionally train the model to match its interpretation with the rationales. However, this intuitive idea has not been fully explored. In this paper, we propose a framework of utilizing interpretation methods and gold rationales to enhance models. Our framework is very general in the sense that it can incorporate various interpretation methods. Previously proposed gradient-based methods can be shown as an instance of our framework. We also propose two novel instances utilizing two other types of interpretation methods, erasure/replace-based and extractor-based methods, for model enhancement. We conduct comprehensive experiments on a variety of tasks. Experimental results show that our framework is effective especially in low-resource settings in enhancing models with various interpretation methods, and our two newly-proposed methods outperform gradient-based methods in most settings. Code is available at https://github.com/Chord-Chen-30/UIMER.

You Wu, Ziheng Chen, Yizhen Zhang et al.

Cross-layer key-value (KV) compression has been found to be effective in efficient inference of large language models (LLMs). Although they reduce the memory consumption of the KV cache, such methods usually introduce non-negligible performance degradation. In this work, we aim to enhance the performance of YOCO, a cross-layer KV compression method that shares the KVs of the middle layer with the top-half layers. We propose YOCO++, an enhanced YOCO that incorporates a weighted residual connection between the KVs of each bottom-half layer and the bottom layer. Compared to YOCO, YOCO++ increases model capacity while maintaining the same training and inference efficiency. Our experiments show that YOCO++ achieves state-of-the-art performance among the cross-layer KV compression methods at a 50% KV cache compression rate, outperforming the standard Transformer.

Yida Zhao, Hao Xve, Xiang Hu et al.

Syntactic language models (SLMs) enhance Transformers by incorporating syntactic biases through the modeling of linearized syntactic parse trees alongside surface sentences. This paper focuses on compositional SLMs that are based on constituency parse trees and contain explicit bottom-up composition of constituent representations. We identify key aspects of design choices in existing compositional SLMs and propose a unified framework encompassing both existing models and novel variants. We conduct a comprehensive empirical evaluation of all the variants in our framework across language modeling, syntactic generalization, summarization, dialogue, and inference efficiency. Based on the experimental results, we make multiple recommendations on the design of compositional SLMs. Our code is released at https://github.com/zhaoyd1/compositional_SLMs.

Qingyang Zhu, Xiang Hu, Pengyu Ji et al.

As a cornerstone in language modeling, tokenization involves segmenting text inputs into pre-defined atomic units. Conventional statistical tokenizers often disrupt constituent boundaries within words, thereby corrupting semantic information. To address this drawback, we introduce morphological structure guidance to tokenization and propose a deep model to induce character-level structures of words. Specifically, the deep model jointly encodes internal structures and representations of words with a mechanism named $\textit{MorphOverriding}$ to ensure the indecomposability of morphemes. By training the model with self-supervised objectives, our method is capable of inducing character-level structures that align with morphological rules without annotated training data. Based on the induced structures, our algorithm tokenizes words through vocabulary matching in a top-down manner. Empirical results indicate that the proposed method effectively retains complete morphemes and outperforms widely adopted methods such as BPE and WordPiece on both morphological segmentation tasks and language modeling tasks. Code is available at https://github.com/martianmartina/TreeTokenizer.

Zeqi Tan, Shen Huang, Zixia Jia et al.

The MultiCoNER \RNum{2} shared task aims to tackle multilingual named entity recognition (NER) in fine-grained and noisy scenarios, and it inherits the semantic ambiguity and low-context setting of the MultiCoNER \RNum{1} task. To cope with these problems, the previous top systems in the MultiCoNER \RNum{1} either incorporate the knowledge bases or gazetteers. However, they still suffer from insufficient knowledge, limited context length, single retrieval strategy. In this paper, our team \textbf{DAMO-NLP} proposes a unified retrieval-augmented system (U-RaNER) for fine-grained multilingual NER. We perform error analysis on the previous top systems and reveal that their performance bottleneck lies in insufficient knowledge. Also, we discover that the limited context length causes the retrieval knowledge to be invisible to the model. To enhance the retrieval context, we incorporate the entity-centric Wikidata knowledge base, while utilizing the infusion approach to broaden the contextual scope of the model. Also, we explore various search strategies and refine the quality of retrieval knowledge. Our system\footnote{We will release the dataset, code, and scripts of our system at {\small \url{https://github.com/modelscope/AdaSeq/tree/master/examples/U-RaNER}}.} wins 9 out of 13 tracks in the MultiCoNER \RNum{2} shared task. Additionally, we compared our system with ChatGPT, one of the large language models which have unlocked strong capabilities on many tasks. The results show that there is still much room for improvement for ChatGPT on the extraction task.

Songlin Yang, Kewei Tu

Constituency parsing and nested named entity recognition (NER) are similar tasks since they both aim to predict a collection of nested and non-crossing spans. In this work, we cast nested NER to constituency parsing and propose a novel pointing mechanism for bottom-up parsing to tackle both tasks. The key idea is based on the observation that if we traverse a constituency tree in post-order, i.e., visiting a parent after its children, then two consecutively visited spans would share a boundary. Our model tracks the shared boundaries and predicts the next boundary at each step by leveraging a pointer network. As a result, it needs only linear steps to parse and thus is efficient. It also maintains a parsing configuration for structural consistency, i.e., always outputting valid trees. Experimentally, our model achieves the state-of-the-art performance on PTB among all BERT-based models (96.01 F1 score) and competitive performance on CTB7 in constituency parsing; and it also achieves strong performance on three benchmark datasets of nested NER: ACE2004, ACE2005, and GENIA. Our code is publicly available at \url{https://github.com/sustcsonglin/pointer-net-for-nested}.

Songlin Yang, Kewei Tu

Graph-based methods, which decompose the score of a dependency tree into scores of dependency arcs, are popular in dependency parsing for decades. Recently, \citet{Yang2022Span} propose a headed-span-based method that decomposes the score of a dependency tree into scores of headed spans. They show improvement over first-order graph-based methods. However, their method does not score dependency arcs at all, and dependency arcs are implicitly induced by their cubic-time algorithm, which is possibly sub-optimal since modeling dependency arcs is intuitively useful. In this work, we aim to combine graph-based and headed-span-based methods, incorporating both arc scores and headed span scores into our model. First, we show a direct way to combine with $O(n^4)$ parsing complexity. To decrease complexity, inspired by the classical head-splitting trick, we show two $O(n^3)$ dynamic programming algorithms to combine first- and second-order graph-based and headed-span-based methods. Our experiments on PTB, CTB, and UD show that combining first-order graph-based and headed-span-based methods is effective. We also confirm the effectiveness of second-order graph-based parsing in the deep learning age, however, we observe marginal or no improvement when combining second-order graph-based and headed-span-based methods. Our code is publicly available at \url{https://github.com/sustcsonglin/span-based-dependency-parsing}.

Songlin Yang, Kewei Tu

We propose a new method for projective dependency parsing based on headed spans. In a projective dependency tree, the largest subtree rooted at each word covers a contiguous sequence (i.e., a span) in the surface order. We call such a span marked by a root word \textit{headed span}. A projective dependency tree can be represented as a collection of headed spans. We decompose the score of a dependency tree into the scores of the headed spans and design a novel $O(n^3)$ dynamic programming algorithm to enable global training and exact inference. Our model achieves state-of-the-art or competitive results on PTB, CTB, and UD. Our code is publicly available at \url{https://github.com/sustcsonglin/span-based-dependency-parsing}.

Songlin Yang, Yanpeng Zhao, Kewei Tu

Probabilistic context-free grammars (PCFGs) with neural parameterization have been shown to be effective in unsupervised phrase-structure grammar induction. However, due to the cubic computational complexity of PCFG representation and parsing, previous approaches cannot scale up to a relatively large number of (nonterminal and preterminal) symbols. In this work, we present a new parameterization form of PCFGs based on tensor decomposition, which has at most quadratic computational complexity in the symbol number and therefore allows us to use a much larger number of symbols. We further use neural parameterization for the new form to improve unsupervised parsing performance. We evaluate our model across ten languages and empirically demonstrate the effectiveness of using more symbols. Our code: https://github.com/sustcsonglin/TN-PCFG

Yanpeng Zhao, Liwen Zhang, Kewei Tu

We introduce Latent Vector Grammars (LVeGs), a new framework that extends latent variable grammars such that each nonterminal symbol is associated with a continuous vector space representing the set of (infinitely many) subtypes of the nonterminal. We show that previous models such as latent variable grammars and compositional vector grammars can be interpreted as special cases of LVeGs. We then present Gaussian Mixture LVeGs (GM-LVeGs), a new special case of LVeGs that uses Gaussian mixtures to formulate the weights of production rules over subtypes of nonterminals. A major advantage of using Gaussian mixtures is that the partition function and the expectations of subtype rules can be computed using an extension of the inside-outside algorithm, which enables efficient inference and learning. We apply GM-LVeGs to part-of-speech tagging and constituency parsing and show that GM-LVeGs can achieve competitive accuracies. Our code is available at https://github.com/zhaoyanpeng/lveg.

Chen Zhu, Yanpeng Zhao, Shuaiyi Huang et al.

Visual attention, which assigns weights to image regions according to their relevance to a question, is considered as an indispensable part by most Visual Question Answering models. Although the questions may involve complex relations among multiple regions, few attention models can effectively encode such cross-region relations. In this paper, we demonstrate the importance of encoding such relations by showing the limited effective receptive field of ResNet on two datasets, and propose to model the visual attention as a multivariate distribution over a grid-structured Conditional Random Field on image regions. We demonstrate how to convert the iterative inference algorithms, Mean Field and Loopy Belief Propagation, as recurrent layers of an end-to-end neural network. We empirically evaluated our model on 3 datasets, in which it surpasses the best baseline model of the newly released CLEVR dataset by 9.5%, and the best published model on the VQA dataset by 1.25%. Source code is available at https: //github.com/zhuchen03/vqa-sva.

Dingchu Zhang, Yida Zhao, Jialong Wu et al.

The rapid advancement of large language models (LLMs) has transformed the landscape of agentic information seeking capabilities through the integration of tools such as search engines and web browsers. However, current mainstream approaches for enabling LLM web search proficiency face significant challenges: supervised fine-tuning struggles with data production in open-search domains, while RL converges quickly, limiting their data utilization efficiency. To address these issues, we propose EvolveSearch, a novel iterative self-evolution framework that combines SFT and RL to enhance agentic web search capabilities without any external human-annotated reasoning data. Extensive experiments on seven multi-hop question-answering (MHQA) benchmarks demonstrate that EvolveSearch consistently improves performance across iterations, ultimately achieving an average improvement of 4.7\% over the current state-of-the-art across seven benchmarks, opening the door to self-evolution agentic capabilities in open web search domains.

Minshen Zhang, Xiang Hu, Jianguo Li et al.

We explore Multi-Head FFN (MH-FFN) as a replacement of FFN in the Transformer architecture, motivated by the structural similarity between single-head attention and FFN. While multi-head mechanisms enhance expressivity in attention, naively applying them to FFNs faces two challenges: memory consumption scaling with the head count, and an imbalanced ratio between the growing intermediate size and the fixed head dimension as models scale, which degrades scalability and expressive power. To address these challenges, we propose Flash Multi-Head FFN (FlashMHF), with two key innovations: an I/O-aware fused kernel computing outputs online in SRAM akin to FlashAttention, and a design using dynamically weighted parallel sub-networks to maintain a balanced ratio between intermediate and head dimensions. Validated on models from 128M to 1.3B parameters, FlashMHF consistently improves perplexity and downstream task accuracy over SwiGLU FFNs, while reducing peak memory usage by 3-5x and accelerating inference by up to 1.08x. Our work establishes the multi-head design as a superior architectural principle for FFNs, presenting FlashMHF as a powerful, efficient, and scalable alternative to FFNs in Transformers.

Zhangzhi Xiong, Haoyi Wu, You Wu et al.

The Probabilistic Transformer (PT) establishes that the Transformer's self-attention plus its feed-forward block is mathematically equivalent to Mean-Field Variational Inference (MFVI) on a Conditional Random Field (CRF). Under this equivalence the Transformer ceases to be a black-box neural network and becomes a programmable factor graph: graph topology, factor potentials, and the message-passing schedule are all explicit and inspectable primitives that can be engineered. PT was originally developed for natural language and in this report we investigate its potential for time series. We first lift PT into the Spatial-Temporal Probabilistic Transformer (ST-PT) to repair PT's missing channel axis and weak per-step semantics, and adopt ST-PT as a shared cornerstone backbone. We then identify three distinct properties that PT/ST-PT offers as a factor-graph model and derive three Research Questions, one per property, that probe how each property can be exploited in time series: RQ1. The graph topology and potentials are direct programmable primitives. Can this be used to inject symbolic time-series priors into ST-PT through structural graph modifications, especially under data scarcity and noise? RQ2. The CRF's factor matrices are the operator's potentials. Can an external condition program these factor matrices on a per-sample basis, so that conditional generation becomes structural rather than feature-level modulation of a fixed one? RQ3. Each MFVI iteration is a Bayesian posterior update on the factor graph. Can this turn the latent transition of latent-space AutoRegressive (AR) forecasting from an opaque MLP into a principled posterior update, and can a CRF teacher distill its latents into the AR student to counter cumulative error? We give one empirical study per question. Together, these three studies position ST-PT as a programmable framework for time-series modeling.

Ning Cheng, Zhaohui Yan, Ziming Wang et al.

Large Language Models (LLMs) play a crucial role in capturing structured semantics to enhance language understanding, improve interpretability, and reduce bias. Nevertheless, an ongoing controversy exists over the extent to which LLMs can grasp structured semantics. To assess this, we propose using Semantic Role Labeling (SRL) as a fundamental task to explore LLMs' ability to extract structured semantics. In our assessment, we employ the prompting approach, which leads to the creation of our few-shot SRL parser, called PromptSRL. PromptSRL enables LLMs to map natural languages to explicit semantic structures, which provides an interpretable window into the properties of LLMs. We find interesting potential: LLMs can indeed capture semantic structures, and scaling-up doesn't always mirror potential. Additionally, limitations of LLMs are observed in C-arguments, etc. Lastly, we are surprised to discover that significant overlap in the errors is made by both LLMs and untrained humans, accounting for almost 30% of all errors.

Wenyang Hui, Kewei Tu

Large language models (LLMs) have demonstrated impressive capability in reasoning and planning when integrated with tree-search-based prompting methods. However, since these methods ignore the previous search experiences, they often make the same mistakes in the search process. To address this issue, we introduce Reflection on search Trees (RoT), an LLM reflection framework designed to improve the performance of tree-search-based prompting methods. It uses a strong LLM to summarize guidelines from previous tree search experiences to enhance the ability of a weak LLM. The guidelines are instructions about solving this task through tree search which can prevent the weak LLMs from making similar mistakes in the past search process. In addition, we proposed a novel state selection method, which identifies the critical information from historical search processes to help RoT generate more specific and meaningful guidelines. In our extensive experiments, we find that RoT significantly improves the performance of LLMs in reasoning or planning tasks with various tree-search-based prompting methods (e.g., BFS and MCTS). Non-tree-search-based prompting methods such as Chain-of-Thought (CoT) can also benefit from RoT guidelines since RoT can provide task-specific knowledge collected from the search experience.

Xiang Hu, Pengyu Ji, Qingyang Zhu et al.

A syntactic language model (SLM) incrementally generates a sentence with its syntactic tree in a left-to-right manner. We present Generative Pretrained Structured Transformers (GPST), an unsupervised SLM at scale capable of being pre-trained from scratch on raw texts with high parallelism. GPST circumvents the limitations of previous SLMs such as relying on gold trees and sequential training. It consists of two components, a usual SLM supervised by a uni-directional language modeling loss, and an additional composition model, which induces syntactic parse trees and computes constituent representations, supervised by a bi-directional language modeling loss. We propose a representation surrogate to enable joint parallel training of the two models in a hard-EM fashion. We pre-train GPST on OpenWebText, a corpus with $9$ billion tokens, and demonstrate the superiority of GPST over GPT-2 with a comparable size in numerous tasks covering both language understanding and language generation. Meanwhile, GPST also significantly outperforms existing unsupervised SLMs on left-to-right grammar induction, while holding a substantial acceleration on training.

You Wu, Haoyi Wu, Kewei Tu

Recently, sharing key-value (KV) cache across layers has been found effective in efficient inference of large language models (LLMs). To systematically investigate different techniques of cross-layer KV sharing, we propose a unified framework that covers several recent methods and their novel variants. We conduct comprehensive experiments on all the configurations of the framework, evaluating their generation throughput and performance in language modeling and downstream tasks. We find that when reducing the size of the KV cache by 2$\times$, most configurations can achieve higher throughput than standard transformers while maintaining competitive performance. When further reducing the size of the KV cache, however, pairing queries of all layers with KVs of upper layers performs better, at the expense of additional training cost and prefilling latency. We hope that this work will help users make more informed choices of cross-layer KV sharing approaches and facilitate future research on efficient LLM inference.

Zhuo Chen, Xinyu Wang, Yong Jiang et al.

In the era of large language models, applying techniques such as Retrieval Augmented Generation can better address Open-Domain Question-Answering problems. Due to constraints including model sizes and computing resources, the length of context is often limited, and it becomes challenging to empower the model to cover overlong contexts while answering questions from open domains. This paper proposes a general and convenient method to covering longer contexts in Open-Domain Question-Answering tasks. It leverages a small encoder language model that effectively encodes contexts, and the encoding applies cross-attention with origin inputs. With our method, the origin language models can cover several times longer contexts while keeping the computing requirements close to the baseline. Our experiments demonstrate that after fine-tuning, there is improved performance across two held-in datasets, four held-out datasets, and also in two In Context Learning settings.

Xiang Hu, Jiaqi Leng, Jun Zhao et al.

A key advantage of Recurrent Neural Networks (RNNs) over Transformers is their linear computational and space complexity enables faster training and inference for long sequences. However, RNNs are fundamentally unable to randomly access historical context, and simply integrating attention mechanisms may undermine their efficiency advantages. To overcome this limitation, we propose Hierarchical Sparse Attention (HSA), a novel attention mechanism that enhances RNNs with long-range random access flexibility while preserving their merits in efficiency and length generalization. HSA divides inputs into chunks, selects the top-$k$ chunks and hierarchically aggregates information. The core innovation lies in learning token-to-chunk relevance based on fine-grained token-level information inside each chunk. This approach enhances the precision of chunk selection across both in-domain and out-of-domain context lengths. To make HSA efficient, we further introduce a hardware-aligned kernel design. By combining HSA with Mamba, we introduce RAMba, which achieves perfect accuracy in passkey retrieval across 64 million contexts despite pre-training on only 4K-length contexts, and significant improvements on various downstream tasks, with nearly constant memory footprint. These results show RAMba's huge potential in long-context modeling.

Yida Zhao, Kuan Li, Xixi Wu et al.

LLM-based search agents are increasingly trained on entity-centric synthetic data to solve complex, knowledge-intensive tasks. However, prevailing training methods like Group Relative Policy Optimization (GRPO) discard this rich entity information, relying instead on sparse, outcome-based rewards. This critical limitation renders them unable to distinguish informative "near-miss" samples-those with substantially correct reasoning but a flawed final answer-from complete failures, thus discarding valuable learning signals. We address this by leveraging the very entities discarded during training. Our empirical analysis reveals a strong positive correlation between the number of ground-truth entities identified during an agent's reasoning process and final answer accuracy. Building on this insight, we introduce Entity-aware Group Relative Policy Optimization (E-GRPO), a novel framework that formulates a dense entity-aware reward function. E-GRPO assigns partial rewards to incorrect samples proportional to their entity match rate, enabling the model to effectively learn from these "near-misses". Experiments on diverse question-answering (QA) and deep research benchmarks show that E-GRPO consistently and significantly outperforms the GRPO baseline. Furthermore, our analysis reveals that E-GRPO not only achieves superior accuracy but also induces more efficient reasoning policies that require fewer tool calls, demonstrating a more effective and sample-efficient approach to aligning search agents.