Papers by Xi Sun

28 papers
FOLIO: Natural Language Reasoning with First-Order Logic (2024.emnlp-main)

Copied to clipboard

Challenge: Existing benchmarks for logical reasoning in large language models lack language naturalness or limited complexity.
Approach: They propose to use first-order logic annotations to evaluate logical reasoning capabilities of large language models.
Outcome: The proposed dataset evaluates the FOL reasoning ability of supervised fine-tuning on medium-sized language models.
AgentGym2: Benchmarking Large Language Model Agents in De-Idealized Real-World Environments (2026.acl-long)

Copied to clipboard

Challenge: Existing benchmarks evaluate agents in simplified, idealized settings, relying on pre-packaged tool interfaces, overlooking critical steps, and assume inputs are clean and fully specified.
Approach: They propose a framework that evaluates language agents in simplified, idealized settings . they show that even SOTA systems like Gemini and GPT-5 struggle on AgentGym2 .
Outcome: Experiments on 15 proprietary and open-source models show that even SOTA systems like Gemini and GPT-5 struggle on AgentGym2 .
Long-tail Relation Extraction via Knowledge Graph Embeddings and Graph Convolution Networks (N19-1)

Copied to clipboard

Challenge: Existing distance supervised relation extraction models for long-tail data are inadequate for many applications.
Approach: They propose to leverage implicit relational knowledge among class labels and learn explicit relational knowing using graph convolution networks.
Outcome: The proposed approach outperforms baselines for long-tail relations on a large-scale dataset.
Effective Large Language Model Adaptation for Improved Grounding and Citation Generation (2024.naacl-long)

Copied to clipboard

Challenge: Large language models generate "hallucinated" answers that are not factual . despite their widespread adoption, they can generate plausiblesounding but nonfactual information.
Approach: They propose a framework that tunes large language models to self-ground claims and provide citations to retrieved documents.
Outcome: The proposed framework generates superior grounded responses with more accurate citations compared to prompting-based approaches and post-hoc citing-based methods.
ChildMandarin: A Comprehensive Mandarin Speech Dataset for Young Children Aged 3-5 (2025.acl-long)

Copied to clipboard

Challenge: Automatic speech recognition systems have advanced significantly with models like Whisper, Conformer, and self-supervised frameworks such as Wav2vec 2.0.
Approach: They propose to use Mandarin speech datasets to analyze pronunciation and tone of children aged 3 to 5 and evaluate their models on speaker verification (SV) They find that the datasets are more robust than those used by adult speech recognition systems and are open-source and available for all academic purposes.
Outcome: The proposed dataset includes 41.25 hours of speech with carefully crafted manual transcriptions, collected from 397 speakers across various provinces in China, with balanced gender representation.
Time-for-Accuracy: Formalizing Chain-of-Thought as an Expansion of Logical Depth (2026.findings-acl)

Copied to clipboard

Challenge: Chain-of-thought (CoT) prompting can improve multi-step reasoning, but it is unclear what kind of additional sequential computation longer traces actually enable.
Approach: They propose a deletion-based measure of step necessity under a specified inference interface to operationalize realized depth beyond raw length.
Outcome: The proposed method combines effective logical depth with Bennett's logical depth to show that it is more efficient than a linear model.
SudokuFill: A Multi-Agent Progressive Filling Framework for Document-Level Scientific Information Extraction (2026.findings-acl)

Copied to clipboard

Challenge: Scientific information extraction (SciIE) is a key bottleneck for turning unstructured papers into computable knowledge bases.
Approach: They propose a scientific information extraction framework that solves a Sudoku problem as a progressive filling problem.
Outcome: The proposed framework outperforms the GPT-4o model on a document-level adjuvant dataset.
STELLA: A Multimodal LLM for Protein Functional Annotation via Unified Sequence-Structure Encoding (2026.findings-acl)

Copied to clipboard

Challenge: a multimodal protein language model (LLM) integrates sequence, structure, and function into functional annotation.
Approach: They propose a multimodal protein language model that synergistically aligns bimodal representations with the textual modality to advance protein functional annotation.
Outcome: The proposed model synergizes bimodal representations with the textual modality to advance protein functional annotation.
Datasets for Scientific Literature Understanding: A Survey (2026.findings-acl)

Copied to clipboard

Challenge: Empowering machines to understand scientific literature is crucial for accelerating scientific discovery and advancing the AI for Science paradigm.
Approach: They propose a systematic taxonomy that organizes resources spanning structural understanding, text understanding, multimodal understanding and pre-training/instruction fine-tuning.
Outcome: The proposed taxonomy organizes resources spanning structural understanding, text understanding, multimodal understanding and pre-training/instruction fine-tuning.
Relation-aware Video Reading Comprehension for Temporal Language Grounding (2021.emnlp-main)

Copied to clipboard

Challenge: Existing methods for temporal language grounding in videos are boundary regression and span extraction tasks.
Approach: They propose a Relation-aware Network to localize a temporal span relevant to a given query sentence.
Outcome: The proposed framework selects a video moment choice from the predefined answer set with the aid of coarse-and-fine choice-query interaction and choice-choice relation construction.
Connectivity Patterns are Task Embeddings (2023.findings-acl)

Copied to clipboard

Challenge: Existing methods for predicting inter-task transferability are sparse and task-specific.
Approach: They propose a method that uses connectivity patterns of neurons as a unique identifier associated with a task.
Outcome: The proposed method outperforms baselines in predicting inter-task transferability across data regimes and transfer settings while keeping high efficiency in computation and storage.
Inconsistency Matters: A Knowledge-guided Dual-inconsistency Network for Multi-modal Rumor Detection (2021.findings-emnlp)

Copied to clipboard

Challenge: Existing rumor detection models focus on textual data to extract distinctive features, but they fail to capture the inconsistency information among the content and background knowledge.
Approach: They propose to capture inconsistency semantics and content-knowledge level in a unified framework to detect rumors with multimedia content.
Outcome: Extensive experiments on two public real-world datasets show that the proposed network outperforms the state-of-the-art models.
CogniBench: A Legal-inspired Framework and Dataset for Assessing Cognitive Faithfulness of Large Language Models (2025.acl-long)

Copied to clipboard

Challenge: Existing benchmarks focus on “factual statements” that rephrase source materials, but ignore “cognitive statements” . evaluating and detecting "faithfulness hallucinations" remains challenging .
Approach: They propose a framework to assess faithfulness of cognitive statements and introduce a dataset to scale easily across models.
Outcome: The proposed framework assesses faithfulness of cognitive statements and scales easily across models.
Sequential-NIAH: A Needle-In-A-Haystack Benchmark for Extracting Sequential Needles from Long Contexts (2025.emnlp-main)

Copied to clipboard

Challenge: Recent models have extended Corresponding Author. context lengths to millions of tokens while maintaining reasoning and comprehension capabilities.
Approach: They propose a benchmark to evaluate the ability of large language models to extract sequential information items from long contexts.
Outcome: The proposed model achieves maximum accuracy of 63.50% on six well-known LLMs.
BFS-Prover: Scalable Best-First Tree Search for LLM-based Automatic Theorem Proving (2025.acl-long)

Copied to clipboard

Challenge: Existing approaches to theorem proving in large language models rely on value functions and/or Monte Carlo Tree Search (MCTS), but the potential of simpler methods like Best-First Tree Search remains underexplored.
Approach: They propose a scalable expert iteration framework that implements strategic data filtering at each expert iteration round, excluding problems solvable via beam search node expansion to focus on harder cases.
Outcome: The proposed framework achieves a state-of-the-art score of 72.95 on the MiniF2F test set and challenges the perceived necessity of complex tree search methods.
Instruction-tuned Language Models are Better Knowledge Learners (2024.acl-long)

Copied to clipboard

Challenge: Large language models store factual knowledge in parameters, but it can become outdated as the work evolves . pre-instruction-tuning improves ability of LLMs to absorb knowledge from new documents .
Approach: They propose a method that instruction-tunes on questions prior to training on documents . they propose to use QA pairs to update factual knowledge of large language models .
Outcome: The proposed method outperforms instruction-tuning on documents by 17.8%.
Discourse Structure-Aware Prefix for Generation-Based End-to-End Argumentation Mining (2024.findings-acl)

Copied to clipboard

Challenge: Recent advances in AM models overlook the integration of supplementary discourse structure information, resulting in suboptimal outcomes.
Approach: They propose a framework which generates discourse structure-aware prefixes for each layer of the generation model.
Outcome: The proposed framework achieves state-of-the-art performance on two AM benchmarks.
EmotionTalk: An Interactive Chinese Multimodal Emotion Dataset With Rich Annotations (2026.findings-acl)

Copied to clipboard

Challenge: Existing datasets face issues such as low quality, limited scale, and incomplete modalities, hindering model performance.
Approach: They propose to use Chinese multimodal datasets to capture authentic emotional interplay from 19 professional actors.
Outcome: The EmotionTalk dataset spans 23.6 hours of dyadic conversations across diverse scenarios.
Parrot: A Training Pipeline Enhances Both Program CoT and Natural Language CoT for Reasoning (2025.emnlp-main)

Copied to clipboard

Challenge: Existing work focuses on enabling models to generate natural language chain-of-thought rationales or leverage executable and verifiable code, such as Python.
Approach: They propose a novel training pipeline that integrates sequential P-CoT and N-Co T generation and a subtask hybrid training strategy to facilitate natural language transferability.
Outcome: The proposed training pipeline improves both N-CoT and P-Co T performance over the RL baseline.
Diversity in Unity, Theory in Practice: Hierarchical Multitask Benchmarks for Chinese Minority Languages (2026.acl-long)

Copied to clipboard

Challenge: CMiLBench is a framework to evaluate linguistically and culturally diverse minority languages . rapid evolution of LLMs has revolutionized NLP, but progress is unevenly distributed .
Approach: They propose a framework to translate a theoretical notion of "diversity in unity" into practical evaluation for three minority languages . CMiLBench comprises 24,663 instances across 5 difficulty levels and 17 tasks .
Outcome: The proposed framework evaluates 14 state-of-the-art LLMs with a hybrid framework . it integrates automatic metrics and LLM-as-a-Judge scoring .
AEGIS: A Holistic Benchmark for Evaluating Forensic Analysis of AI-Generated Academic Images (2026.acl-long)

Copied to clipboard

Challenge: AEGIS examines whether current models can effectively audit AI-generated images in academic papers.
Approach: They propose a holistic benchmark for forensic analysis of AI-Generated academic ImageS that reveals limitations in academic image forensics.
Outcome: AEGIS compared with existing benchmarks on seven academic categories and features key advances in forensic analysis.
Decoding Scientific Experimental Images: The SPUR Benchmark for Perception, Understanding, and Reasoning (2026.acl-long)

Copied to clipboard

Challenge: Xu and Peng, 2025) . . SPUR is a comprehensive benchmark for scientific experimental image perception, understanding, and reasoning, comprising 4,264 question-answering (QA) pairs derived from 1,084 expert-curated images.
Approach: They propose to use 4,264 question-answering (QA) pairs derived from 1,084 expert-curated images to evaluate the visual perception of multimodal large language models (MLLMs) . they also propose to utilize cross-panel relation understanding to evaluate MLLM’s ability to decipher intricate cross-panel relations.
Outcome: The proposed model is based on 4,264 question-answering pairs derived from 1,084 expert-curated images.
Field Embedding: A Unified Grain-Based Framework for Word Representation (2021.naacl-main)

Copied to clipboard

Challenge: Current methods focus on learning word embeddings while linguistic information is discarded after the learning.
Approach: They propose a framework field embedding to jointly learn word and grain embedds by incorporating morphological, phonetic, and syntactical linguistic fields.
Outcome: The proposed framework integrates morphological, phonetic, and syntactical linguistic fields to learn word embeddings and grain embedds.
Attention-Based Capsule Networks with Dynamic Routing for Relation Extraction (D18-1)

Copied to clipboard

Challenge: Existing neural networks focus on instance representation, and subsampling fails to retain precise spatial relationships between higher-level parts.
Approach: They propose a neural approach based on capsule networks with attention mechanisms to extract relational information from a capsule.
Outcome: The proposed method improves the precision of the predicted relations with different benchmarks.
CritiQ: Mining Data Quality Criteria from Human Preferences (2025.acl-long)

Copied to clipboard

Challenge: Existing methods to train language models rely on manual design, perplexity, or careful prompt engineering.
Approach: They propose a method that automatically mines criteria from human preferences for data quality with only 30 human-annotated pairs and performs efficient data selection.
Outcome: The proposed method improves on human-annotated test sets and shows high accuracy on code, math, and logic domains.
From Scores to Preferences: Redefining Evaluation Paradigm for Speech Quality Reward Modeling (2026.findings-acl)

Copied to clipboard

Challenge: Experimental results show that the MOS-aware GRM significantly improves fine-grained speech quality discrimination.
Approach: They propose a MOS-aware reward model that incorporates MOS gap into reward function during reinforcement learning.
Outcome: The proposed model significantly improves fine-grained speech quality discrimination.
Mitigating Object Hallucinations in MLLMs via Multi-Frequency Perturbations (2025.findings-emnlp)

Copied to clipboard

Challenge: Multimodal large language models have demonstrated remarkable performance in visual-language tasks, but their authenticity is often compromised by object hallucinations.
Approach: They propose a multi-frequency perturbation method that leverages both low-frequency and high-frequency features of images to perturb visual feature representations and explicitly suppress redundant frequency-domain features during inference.
Outcome: The proposed method significantly mitigates object hallucinations across various model architectures.
Beyond Scaling: Measuring and Predicting the Upper Bound of Knowledge Retention in Language Model Pre-Training (2026.acl-long)

Copied to clipboard

Challenge: Existing methods to predict performance of large language models are lacking . authors propose a size-dependent mutual information predictor for closed-book question answering accuracy .
Approach: They propose a size-dependent mutual information predictor that integrates knowledge frequency, knowledge specificity, and model size to forecast closed-book question answering accuracy.
Outcome: The proposed method outperforms baseline models and achieves R2 > 0.7 in predicting QA accuracy without additional training.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations