Papers by Zijian Wang
ReCode: Robustness Evaluation of Code Generation Models (2023.acl-long)
Copied to clipboard
Shiqi Wang, Zheng Li, Haifeng Qian, Chenghao Yang, Zijian Wang, Mingyue Shang, Varun Kumar, Samson Tan, Baishakhi Ray, Parminder Bhatia, Ramesh Nallapati, Murali Krishna Ramanathan, Dan Roth, Bing Xiang
| Challenge: | Existing work on robustness in text or code tasks has focused on classification, while robustness for code generation tasks is an uncharted area. |
| Approach: | They propose a robustness evaluation benchmark for code generation models that customizes over 30 transformations specifically for code on docstrings, function and variable names, code syntax, and code format. |
| Outcome: | The proposed model performs better on human annotators and on SOTA models with human annnotators. |
From Charts to Code: A Hierarchical Benchmark for Multimodal Models (2026.acl-long)
Copied to clipboard
Jiahao Tang, Henry Hengyuan Zhao, Lijian Wu, Zijian Zhang, Yifei Tao, Dongxing Mao, Yang Wan, Jingru Tan, Min Zeng, Min Li, Alex Jinpeng Wang
| Challenge: | Chart2Code is a new benchmark for evaluating the natural language to chart code generation capabilities of large multimodal models. |
| Approach: | They introduce Chart2Code, a new benchmark for evaluating the natural language to chart code generation capabilities of large multimodal models. |
| Outcome: | The proposed benchmark is the first to scale task complexity while capturing diverse scenarios. |
Uncovering Scaling Laws for Large Language Models via Inverse Problems (2025.findings-emnlp)
Copied to clipboard
Arun Verma, Zhaoxuan Wu, Zijian Zhou, Xiaoqiang Lin, Zhiliang Chen, Rachael Hwee Ling Sim, Rui Qiao, Jingtan Wang, Nhung Bui, Xinyuan Niu, Wenyang Hu, Gregory Kang Ruey Lau, Zi-Yu Khoo, Zitong Zhao, Xinyi Xu, Apivich Hemachandra, See-Kiong Ng, Bryan Kian Hsiang Low
| Challenge: | Large Language Models (LLMs) have achieved remarkable success across diverse domains. |
| Approach: | inverse problems can efficiently uncover scaling laws that guide the building of LLMs, authors argue . authors propose brute-force approaches to improve LLM training costs due to high costs . |
| Outcome: | This paper advocates that inverse problems can efficiently uncover scaling laws that guide the building of LLMs to achieve the desirable performance with significantly better cost-effectiveness. |
A Static Evaluation of Code Completion by Large Language Models (2023.acl-industry)
Copied to clipboard
Hantian Ding, Varun Kumar, Yuchen Tian, Zijian Wang, Rob Kwiatkowski, Xiaopeng Li, Murali Krishna Ramanathan, Baishakhi Ray, Parminder Bhatia, Sudipta Sengupta
| Challenge: | Large language models trained on code have shown great potential to increase productivity of software developers. |
| Approach: | They propose a static evaluation framework to quantify static errors in Python code completions by leveraging Abstract Syntax Trees. |
| Outcome: | The proposed framework is more efficient and applicable to code in the wild. |
Locating and Extracting Relational Concepts in Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing knowledge recall models lack interpretability for relational concepts . a hidden state expresses causal effects of relational concept in input prompts . |
| Approach: | They propose to use causal mediation analysis to find hidden states that express relational concepts in LLMs. |
| Outcome: | The proposed representations exhibit high credibility and can be flexibly transplanted into other recall processes. |
ThoughtProbe: Classifier-Guided LLM Thought Space Exploration via Probing Representations (2025.emnlp-main)
Copied to clipboard
| Challenge: | Unlike previous works that manipulate representations to steer LLM generation, ThoughtProbe harnesses them as discriminative signals to guide the tree-structured response space exploration. |
| Approach: | They propose a tree-structured inference-time framework that leverages the hidden reasoning features of Large Language Models to improve their reasoning performance. |
| Outcome: | The proposed framework improves reasoning performance across multiple arithmetic reasoning benchmarks and covers valid reasoning chains and identifies optimal answers. |
LibEvolutionEval: A Benchmark and Study for Version-Specific Code Generation (2025.naacl-long)
Copied to clipboard
Sachit Kuhar, Wasi Uddin Ahmad, Zijian Wang, Nihal Jain, Haifeng Qian, Baishakhi Ray, Murali Krishna Ramanathan, Xiaofei Ma, Anoop Deoras
| Challenge: | Recent code completion models focus on local file contexts, but do not fully capture the complexities of real-world software development. |
| Approach: | They propose a version-specific code-completion task across eight libraries as they evolve over the years and an in-depth analysis of two widely used public libraries: PyTorch and Matplotlib. |
| Outcome: | The proposed model improves performance with public libraries, compared with existing models. |
ChunQiuTR: Time-Keyed Temporal Retrieval in Classical Chinese Annals (2026.findings-acl)
Copied to clipboard
| Challenge: | Historical research often focuses on finding exact record for a specific regnal month . classical Chinese sources are a canonical example of evidence-centric retrieval . |
| Approach: | They propose a time-keyed retrieval benchmark that organizes records by month-level reign keys . they propose 'CTD', a dual-encoder that combines absolute context with offset biasing . |
| Outcome: | The proposed benchmark organizes records by month-level reign keys and includes chrono-near confounders that mimic real retrieval failures. |
A Survey of Pun Generation: Datasets, Evaluations and Methodologies (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Pun generation aims to modify linguistic elements in text to produce humour or evoke double meanings. |
| Approach: | They propose to review pun generation datasets and methods across different stages . pun generation aims to produce humour or evoke double meanings . |
| Outcome: | This paper summarises both automated and human evaluation metrics used to assess the quality of pun generation. |
TalkDown: A Corpus for Condescension Detection in Context (D19-1)
Copied to clipboard
| Challenge: | condescending language use can bring dialogues to an end and disrupt healthy communities. |
| Approach: | They propose a model that uses a language-only model to model condescending linguistic acts in context. |
| Outcome: | a new model of condescending language use improves performance and motivates techniques . the model can estimate condescension rates in various online communities and relate these differences to community norms . |
Towards a Better Understanding of Label Smoothing in Neural Machine Translation (2020.aacl-main)
Copied to clipboard
| Challenge: | In recent years, Neural Network (NN) models bring steady and concrete improvements on the task of Machine Translation (MT). |
| Approach: | They propose to penalize over-confident outputs and regularize the model so that its outputs do not diverge too much from some prior distribution. |
| Outcome: | The proposed method is well-motivated and can improve the performance of strong neural machine translation systems. |
Predicting and Using Target Length in Neural Machine Translation (2020.aacl-main)
Copied to clipboard
| Challenge: | Current NMT systems do not model the length of the output explicitly . length normalization is a common technique used in the beam search of NMT to enable a fair comparison of partial hypotheses with different lengths. |
| Approach: | They propose to use length prediction as an auxiliary task to obtain length information from the encoder. |
| Outcome: | The proposed sub-network improves over the baseline system and the predicted length can be used as an alternative to length normalization during decoding. |
COMPKE: Complex Question Answering under Knowledge Editing (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing benchmarks for knowledge editing do not accurately evaluate how well models apply knowledge in real-life situations. |
| Approach: | They propose a benchmark to evaluate how well updated models apply new knowledge in real-life situations. |
| Outcome: | The proposed method achieves 39.47 accuracy on GPT-4o-mini but drops significantly to 3.83 on Qwen2.5-3B. |
Dual-Feedback Knowledge Retrieval for Task-Oriented Dialogue Systems (2023.emnlp-main)
Copied to clipboard
| Challenge: | Current approaches to task-oriented dialogue systems integrate knowledge retrieval and response generation, which poses scalability challenges when dealing with extensive knowledge bases. |
| Approach: | They propose a retriever-generator architecture that harnesses a retrieval and a generator to generate system responses by using feedback from the generator as pseudo-labels. |
| Outcome: | The proposed architecture shows superior performance on three benchmark datasets. |
Token Alignment via Character Matching for Subword Completion (2024.findings-acl)
Copied to clipboard
Ben Athiwaratkun, Shiqi Wang, Mingyue Shang, Yuchen Tian, Zijian Wang, Sujan Kumar Gonugondla, Sanjay Krishna Gouda, Robert Kwiatkowski, Ramesh Nallapati, Parminder Bhatia, Bing Xiang
| Challenge: | Generative models struggle with prompts corresponding to partial tokens due to tokenization, where partial token is out-of-distribution during inference. |
| Approach: | They propose a method to alleviate tokenization artifact on text completion by backtracking to the last complete tokens and aligning subsequent generations to match with the prompt. |
| Outcome: | The proposed method shows that it improves on partial token scenarios with only a minor time increase. |
HSGraphAgent: Knowledge-Graph-Guided Large Language Models for Harmonized System Code Classification (2026.acl-long)
Copied to clipboard
| Challenge: | Harmonized System (HS) code classification is a hierarchically structured and regulation-constrained task, often complicated by short and noisy product descriptions. |
| Approach: | They propose a knowledge-graph-guided LLM framework that formulates HS classification as a stepwise, regulation-aware reasoning process over an explicit HS knowledge graph. |
| Outcome: | The proposed framework constrains inference to legally valid paths while producing explicit and traceable reasoning trajectories. |
Mirror: A Universal Framework for Various Information Extraction Tasks (2023.emnlp-main)
Copied to clipboard
Tong Zhu, Junfei Ren, Zijian Yu, Mengsong Wu, Guoliang Zhang, Xiaoye Qu, Wenliang Chen, Zhefeng Wang, Baoxing Huai, Min Zhang
| Challenge: | Recent studies often formulate IE tasks as a triplet extraction problem, but this paradigm does not support multi-span and n-ary extraction, leading to weak versatility. |
| Approach: | They propose a multi-span cyclic graph extraction problem and a non-autoregressive graph decoding algorithm to extract all spans in a single step. |
| Outcome: | The proposed model outperforms or reaches competitive performance with SOTA systems under few-shot and zero-shot settings and it is compatible with 57 datasets. |
Graph Neural Network Enhanced Retrieval for Question Answering of Large Language Models (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing retrieval methods divide reference documents into passages, treating them in isolation. Existing methods only use contiguous passages or keywords. |
| Approach: | They propose a retrieval method that leverages graph neural networks to exploit relatedness between passages to enhance retrieval. |
| Outcome: | The proposed method improves retrieval by exploiting the relatedness between passages. |
Transformer-Based Direct Hidden Markov Model for Machine Translation (2021.acl-srw)
Copied to clipboard
| Challenge: | Recent studies have found that word alignments produced by the multi-head cross-attention weights are poor. |
| Approach: | They propose to introduce the hidden Markov model to the transformer architecture and introduce alignment components while keeping the system monolithic. |
| Outcome: | The proposed model outperforms the baseline model but is slower in training and decoding. |
LLM-FK: Multi-Agent LLM Reasoning for Foreign Key Detection in Large-Scale Complex Databases (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for detecting missing foreign keys are limited in capturing semantic dependencies across schemas. |
| Approach: | They propose a framework that integrates four agents to detect missing foreign keys . they propose combinatorial search space explosion, ambiguous inference and global inconsistency . |
| Outcome: | The proposed framework achieves F1-scores above 93% on large-scale MusicBrainz database . it reduces candidate search space by two to three orders of magnitude without losing true FKs . |
Empowering Multi-Turn Tool-Integrated Agentic Reasoning with Group Turn Policy Optimization (2026.acl-long)
Copied to clipboard
Yifeng Ding, Hung Le, Songyang Han, Kangrui Ruan, Zhenghui Jin, Varun Kumar, Zijian Wang, Anoop Deoras
| Challenge: | Current reinforcement learning methods suffer from coarse-grained, trajectory-level rewards that provide insufficient learning signals for complex multi-turn interactions, leading to training stagnation. |
| Approach: | They propose a novel RL algorithm for training large language models for multi-turn tool-integrated reasoning (TIR) that incorporates three innovations: turn-level reward assignment that provides fine-grained feedback for individual turns, return-based advantage estimation where normalized discounted returns are calculated as advantages, and self-supervised reward shaping that exploits self-supervision signals from generated code to densify sparse binary outcome-based rewards. |
| Outcome: | The proposed algorithm outperforms GRPO by 3.0% across diverse math reasoning benchmarks and improves grepo by 3.9% on commonsense reasoning and program synthesis tasks. |
CodeFort: Robust Training for Code Generation Models (2024.findings-emnlp)
Copied to clipboard
Yuhao Zhang, Shiqi Wang, Haifeng Qian, Zijian Wang, Mingyue Shang, Linbo Liu, Sanjay Krishna Gouda, Baishakhi Ray, Murali Krishna Ramanathan, Xiaofei Ma, Anoop Deoras
| Challenge: | Existing research efforts to improve code generation models are inadequate . code generation model performance is degraded under small perturbations . |
| Approach: | They propose a framework to improve the robustness of code generation models by generalizing code perturbations to enrich training data and enabling various robust training strategies. |
| Outcome: | The proposed framework increases pass rates and robustness drop rate against code-syntax perturbations. |
Personal Travel Solver: A Preference-Driven LLM-Solver System for Travel Planning (2025.acl-long)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models (LLMs) have demonstrated remarkable capabilities in understanding complex instructions and reasoning across diverse domains. |
| Approach: | They propose to integrate user’s implicit preference into the progress of travel planning by integrating real user reviews and point-of-interest metadata from Google Local into RealTravel. |
| Outcome: | The proposed system achieves better performance than baseline methods and improves the level of personalization. |
DQ-BART: Efficient Sequence-to-Sequence Model via Joint Distillation and Quantization (2022.acl-short)
Copied to clipboard
Zheng Li, Zijian Wang, Ming Tan, Ramesh Nallapati, Parminder Bhatia, Andrew Arnold, Bing Xiang, Dan Roth
| Challenge: | Empirical analyses show that pre-trained sequence-to-sequence models can achieve a 16.5x model footprint compression ratio with little performance drop relative to full-precision counterparts. |
| Approach: | They propose to distill and quantize pre-trained sequence-to-sequence models to reduce memory and latency requirements. |
| Outcome: | Empirical results show that the proposed model achieves 16.5x model footprint compression ratio with little performance drop relative to full-precision counterparts on multiple summarization and QA datasets. |
It’s going to be okay: Measuring Access to Support in Online Communities (D18-1)
Copied to clipboard
| Challenge: | Despite substantial efforts to reduce gender disparities in online social contexts, gender gaps persist and negatively affect women through online harassment. |
| Approach: | They propose a new dataset and method for identifying supportive replies and new methods for inferring gender from text and name to examine the disparity in support across millions of online interactions. |
| Outcome: | The proposed model shows that identifying as a woman is associated with higher rates of support, but also higher rates disparagement. |
ContraCLM: Contrastive Learning For Causal Language Model (2023.acl-long)
Copied to clipboard
Nihal Jain, Dejiao Zhang, Wasi Uddin Ahmad, Zijian Wang, Feng Nan, Xiaopeng Li, Ming Tan, Ramesh Nallapati, Baishakhi Ray, Parminder Bhatia, Xiaofei Ma, Bing Xiang
| Challenge: | Existing studies show that causal language models lack expressiveness due to poor discrimination ability. |
| Approach: | They propose a contrastive learning framework that enhances discrimination of representations and bridges the gap with encoder-only models. |
| Outcome: | The proposed framework improves discrimination and source code generation capabilities on a variety of downstream tasks. |
SKGSum: Structured Knowledge-Guided Document Summarization (2024.findings-acl)
Copied to clipboard
Qiqi Wang, Ruofan Wang, Kaiqi Zhao, Robert Amor, Benjamin Liu, Jiamou Liu, Xianda Zheng, Zijian Huang
| Challenge: | Existing summarization methods ignore the importance of summary structure, resulting in summaries that emphasize the most prominent information while omitting essential details from other sections. |
| Approach: | They propose a method that uses automatically extracted summary points to generate summaries. |
| Outcome: | The proposed methods improve quality and BERTScore of summaries and broaden the types of documents that can be effectively summarized. |
CoCoMIC: Code Completion by Jointly Modeling In-file and Cross-file Context (2024.lrec-main)
Copied to clipboard
Yangruibo Ding, Zijian Wang, Wasi U. Ahmad, Murali Krishna Ramanathan, Ramesh Nallapati, Parminder Bhatia, Dan Roth, Bing Xiang
| Challenge: | Pre-trained language models (LMs) for code have shown promising performance in code completion tasks but ignore the rich semantics in other files within the same project. |
| Approach: | They propose a framework that jointly learns the in-file and cross-file context on top of code LMs and a static-analysis-based tool that locates and retrieves the most relevant project-level cross- file context for code completion. |
| Outcome: | The proposed framework improves existing code LMs with a 33.94% relative increase in exact match and 28.69% in identifier matching when the cross-file context is provided. |
NLP+Code: Code Intelligence in Language Models (2025.emnlp-tutorials)
Copied to clipboard
| Challenge: | Language models have shown impressive abilities in a range of natural language processing tasks. |
| Approach: | This tutorial will provide an overview of the latest advances in natural language processing . it will provide preliminaries of training foundation models on code and their common practices . |
| Outcome: | This tutorial aims to provide an overview of recent advances in code modeling . it provides preliminaries of training foundation models on code and their common practices . |
Planning-Aware Code Infilling via Horizon-Length Prediction (2025.emnlp-main)
Copied to clipboard
| Challenge: | Current approaches to fill-in-the-middle (FIM) often fail to generate content that aligns well with the surrounding context. |
| Approach: | They propose a training objective that teaches models to predict the number of remaining middle tokens at each step. |
| Outcome: | The proposed training objective improves FIM performance by up to 24% on diverse benchmarks across file-level and repository-level. |
Position Paper: Data-Centric AI in the Age of Large Language Models (2024.findings-emnlp)
Copied to clipboard
Xinyi Xu, Zhaoxuan Wu, Rui Qiao, Arun Verma, Yao Shu, Jingtan Wang, Xinyuan Niu, Zhenfeng He, Jiangwei Chen, Zijian Zhou, Gregory Kang Ruey Lau, Hieu Dao, Lucas Agussurja, Rachael Hwee Ling Sim, Xiaoqiang Lin, Wenyang Hu, Zhongxiang Dai, Pang Wei Koh, Bryan Kian Hsiang Low
| Challenge: | a paper proposes a data-centric perspective of AI research, focusing on large language models. |
| Approach: | They propose a data-centric viewpoint of AI research, focusing on large language models . they propose four scenarios centered around data, including data curation, attribution, knowledge transfer . |
| Outcome: | The proposed research focuses on large language models with data centric benchmarks . the proposed benchmarks can be used to develop new data curation methods . |
Answering Complex Open-domain Questions Through Iterative Query Generation (D19-1)
Copied to clipboard
| Challenge: | Currently, one-step retrieve-and-read question answering systems cannot answer such questions because they rarely contain retrievable clues about the missing entity. |
| Approach: | They propose a multi-step approach to retrieve relevant content with the question, then reading the paragraphs returned by the information retrieval component to arrive at the final answer. |
| Outcome: | The proposed model outperforms the best previously published model despite not using pretrained language models such as BERT. |