Siyang Liu, Chujie Zheng, Orianna Demasi, Sahand Sabour, Yu Li, Zhou Yu, Yong Jiang, Minlie Huang. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Xinyu Wang, Yong Jiang, Nguyen Bach, Tao Zhongqiang Huang, Fei Kewei Tu. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Xinyu Wang, Yong Jiang, Nguyen Bach, Tao Zhongqiang Huang, Fei Kewei Tu. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Deep learning (e.g., Transformer) has been widely and successfully used in multivariate time series forecasting (MTSF). Unlike existing methods that focus on training models from a single modal of input, large language (LLMs) based MTSF with cross-modal text input have recently shown great superiority, especially limited temporal data. However, current LLM-based usually adapting fine-tuning LLMs, while neglecting the distribution discrepancy between textual tokens, thus leading to...

Xinyu Wang, Yongliang Shen, Jiong Cai, Tao Xiaobin Pengjun Xie, Fei Huang, Weiming Lu, Yueting Zhuang, Kewei Tu, Wei Yong Jiang. Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022). 2022.

Xinyu Wang, Min Gui, Yong Jiang, Zixia Jia, Nguyen Bach, Tao Zhongqiang Huang, Kewei Tu. Proceedings of the 2022 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies. 2022.

Multimodal named entity recognition (MNER) aims to detect and classify entities in multimodal scenarios. It requires bridging the gap between natural language visual context, which presents two-fold challenges: cross-modal alignment is diversified, interaction sometimes implicit. Existing MNER methods are vulnerable some implicit interactions prone overlook involved significant features. To tackle this problem, we novelly propose refine attention by identifying highlighting task-salient The...

Recent advances in large language models (LLMs) have demonstrated remarkable potential the field of natural processing. Unfortunately, LLMs face significant security and ethical risks. Although techniques such as safety alignment are developed for defense, prior researches reveal possibility bypassing defenses through well-designed jailbreak attacks. In this paper, we propose QueryAttack, a novel framework to systematically examine generalizability alignment. By treating knowledge databases,...

Multilingual sequence labeling is a task of predicting label sequences using single unified model for multiple languages. Compared with relying on monolingual models, multilingual has the benefit smaller size, easier in online serving, and generalizability to low-resource However, current models still underperform individual significantly due capacity limitations. In this paper, we propose reduce gap between by distilling structural knowledge several (teachers) (student). We two novel KD...

In this paper we propose an end-to-end neural CRF autoencoder (NCRF-AE) model for semi-supervised learning of sequential structured prediction problems. Our NCRF-AE consists two parts: encoder which is a enhanced by deep networks, and decoder generative trying to reconstruct the input. has unified structure with different loss functions labeled unlabeled data shared parameters. We developed variation EM algorithm optimizing both simultaneously decoupling their Experimental results over...

Unsupervised dependency parsing, which tries to discover linguistic structures from unannotated data, is a very challenging task. Almost all previous work on this task focuses learning generative models. In paper, we develop an unsupervised parsing model based the CRF autoencoder. The encoder part of our discriminative and globally normalized allows us use rich features as well universal priors. We propose exact algorithm for tractable algorithm. evaluated performance eight multilingual...

Building an effective adversarial attacker and elaborating on countermeasures for attacks natural language processing (NLP) have attracted a lot of research in recent years. However, most the existing approaches focus classification problems. In this paper, we investigate defenses structured prediction tasks NLP. Besides difficulty perturbing discrete words sentence fluency problem faced by attackers any NLP tasks, there is specific challenge to models: output models sensitive small...

Most of the unsupervised dependency parsers are based on first-order probabilistic generative models that only consider local parent-child information. Inspired by second-order supervised parsing, we proposed a extension neural incorporate grandparent-child or sibling We also propose novel design parameterization and optimization methods models. In models, number grammar rules grows cubically with increase vocabulary size, making it difficult to train lexicalized may contain thousands words....

Most of the unsupervised dependency parsers are based on probabilistic generative models that learn joint distribution given sentence and its parse. Probabilistic usually explicit decompose desired tree into factorized grammar rules, which lack global features entire sentence. In this paper, we propose a novel model called discriminative neural with valence (D-NDMV) generates parse from continuous latent representation, encodes contextual information generated We two approaches to...

Multivariate time series forecasting has recently gained great success with the rapid growth of deep learning models. However, existing approaches usually train models from scratch using limited temporal data, preventing their generalization. Recently, surge Large Language Models (LLMs), several works have attempted to introduce LLMs into forecasting. Despite promising results, these methods directly take as input LLMs, ignoring inherent modality gap between and text data. In this work, we...

This paper focuses on the task of cross domain few-shot named entity recognition (NER), which aims to adapt knowledge learned from source recognize entities in target with only a few labeled examples. To address this challenging task, we propose MANNER, variational memory-augmented NER model. Specifically, MANNER uses memory module store information and then retrieve relevant augment domain. In order effectively utilize memory, optimal transport process can explicitly retrieved improve...

Sum-product networks (SPNs) are a class of probabilistic graphical models that allow tractable marginal inference. However, the maximum posteriori (MAP) inference in SPNs is NP-hard. We investigate MAP from both theoretical and algorithmic perspectives. For part, we reduce general to its special case without evidence hidden variables; also show it NP-hard approximate problem 2nε for fixed 0 ≤ ε < 1, where n input size. first present an exact solver runs reasonably fast could handle with...

Unsupervised constituency parsing aims to learn a parser from training corpus without parse tree annotations. While many methods have been proposed tackle the problem, including statistical and neural methods, their experimental results are often not directly comparable due discrepancies in datasets, data preprocessing, lexicalization, evaluation metrics. In this paper, we first examine settings used previous work propose standardize for better comparability between methods. We then...

Despite their impressive generative capabilities, LLMs are hindered by fact-conflicting hallucinations in real-world applications. The accurate identification of texts generated LLMs, especially complex inferential scenarios, is a relatively unexplored area. To address this gap, we present FactCHD, dedicated benchmark designed for the detection from LLMs. FactCHD features diverse dataset that spans various factuality patterns, including vanilla, multi-hop, comparison, and set operation. A...

This paper presents the system used in our submission to IWPT 2020 Shared Task. Our is a graph-based parser with second-order inference. For low-resource Tamil corpora, we specially mixed training data of other languages and significantly improved performance Tamil. Due misunderstanding requirements, submitted graphs that are not connected, which makes only rank 6th over 10 teams. However, after fixed this problem, 0.6 ELAS higher than team ranked 1st official results.

In open-domain dialogue systems, generative approaches have attracted much attention for response generation. However, existing methods are heavily plagued by generating safe responses and unnatural responses. To alleviate these two problems, we propose a novel framework named Dual Adversarial Learning(DAL) high-quality DAL innovatively utilizes the duality between query generation to avoid increase diversity of generated Additionally, uses adversarial learning mimic human judges guides...

Transition-based dependency parsing is a fast and effective approach for parsing. Traditionally, transitionbased parser processes an input sentence predicts sequence of actions in left-to-right manner. During this process, early prediction error may negatively impact the subsequent actions. In paper, we propose simple framework bidirectional training, learn right-to-left separately. To parse sentence, perform joint decoding with two parsers. We three algorithms that are based on scoring,...

Entity retrieval, which aims at disambiguating mentions to canonical entities from massive KBs, is essential for many tasks in natural language processing. Recent progress entity retrieval shows that the dual-encoder structure a powerful and efficient framework nominate candidates if are only identified by descriptions. However, they ignore property meanings of diverge different contexts related various portions descriptions, treated equally previous works. In this work, we propose...

Xinyu Wang, Yong Jiang, Zhaohui Yan, Zixia Jia, Nguyen Bach, Tao Zhongqiang Huang, Fei Kewei Tu. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.