Papers by Su Lei

15 papers
QUITO-X: A New Perspective on Context Compression from the Information Bottleneck Theory (2025.findings-emnlp)

Copied to clipboard

Challenge: Existing methods for compressing context by removing redundant tokens are inconsistent with the objective of retaining the most important tokens when conditioning on a given query.
Approach: They propose a method that uses information bottleneck theory to compress context . they propose to remove redundant tokens using metrics such as self-information or perplexity .
Outcome: The proposed method achieves a 25% increase in compression rate compared to the state-of-the-art .
Lost in Decomposition: Analyzing and Mitigating the Limitations of Long Context Methods via Context Dependency (2026.findings-acl)

Copied to clipboard

Challenge: Existing workflow-based long context methods do not perform well on specific datasets . performance degradation is associated with the indiscriminate application of long context models .
Approach: They propose a training-free adaptive routing strategy to improve long context large language models' robustness.
Outcome: The proposed method can be generalized to all types of datasets, but performance degradation is a concern.
Sparse-RL: Breaking the Memory Wall in LLM Reinforcement Learning via Stable Sparse Rollouts (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for storing key-value caches during long-horizon rollouts cause performance collapses.
Approach: They propose a new training paradigm that empowers stable RL training under sparse rollouts.
Outcome: The proposed model reduces rollout overhead while maintaining the performance.
MenatQA: A New Dataset for Testing the Temporal Comprehension and Reasoning Abilities of Large Language Models (2023.findings-emnlp)

Copied to clipboard

Challenge: Large language models (LLMs) have shown nearly saturated performance on many NLP tasks.
Approach: They construct multiple sensitive factors time QA which encompasses three temporal factors . they test current mainstream LLMs with different parameter sizes .
Outcome: The proposed model incorporates three temporal factors with 2,853 samples . the results show that LLMs fall behind smaller models on these factors .
GEM: A General Evaluation Benchmark for Multimodal Tasks (2021.findings-acl)

Copied to clipboard

Challenge: Existing datasets that focus on natural language tasks are not considered as a general evaluation benchmark for multimodal tasks.
Approach: They present a general evaluation benchmark for multimodal tasks, GEM 1 . they compare it with existing multimodal vision-language datasets .
Outcome: The proposed model is compared with existing vision-language datasets focusing on natural language tasks . it is the largest vision-linguistic dataset covering image-language tasks and video-language task at the same time .
Distilling Large Embeddings via Hyperspherical Householder Quantization (2026.acl-long)

Copied to clipboard

Challenge: Existing methods for quantizing large embeddings rely on Euclidean quantization, which is poorly aligned with the angular geometry induced by contrastive embeddment training.
Approach: They propose a geometry-aware distillation method that compresses large embeddings into short discrete representations via iterative Householder transformations on the unit hypersphere.
Outcome: The proposed method reduces decoding cost and maintains strong semantic retrieval accuracy.
RACQC: Advanced Retrieval-Augmented Generation for Chinese Query Correction (2025.findings-emnlp)

Copied to clipboard

Challenge: Large language models (LLMs) exhibit remarkable capabilities across many tasks, but face critical challenges in the CSC scenario: (1) poor generalization to rare entities in open-domain searches; and (2) failure to adapt to temporal entity variations due to static parameters, resulting in serious over-correction issues.
Approach: They propose a Chinese Spelling Check system with RAG and multi-task learning that integrates dynamic knowledge retrieval and entity-centric RAG to address rare entities.
Outcome: The proposed system outperforms existing baselines in the CSC task and achieves a maximum improvement of +9.92% on the search scenario benchmark and +3.2% on the general-domain dataset.
Intent-Driven Semantic ID Generation for Grounded Conversational News Recommendation (2026.acl-industry)

Copied to clipboard

Challenge: a new approach to news recommendation grounds each suggestion in a rapidly evolving article corpus while addressing implicit user intents that lack explicit retrievable keywords.
Approach: They propose an intent-driven Semantic ID generation paradigm to address these challenges . they map diverse intents to hierarchical SID prefixes and then fuzzy-match them to current news pool .
Outcome: The proposed model achieves 0% hallucination and 12.4% L1 match on a mainstream Chinese news platform.
Towards Linear Time Neural Machine Translation with Capsule Networks (D19-1)

Copied to clipboard

Challenge: Neural Machine Translation (NMT) is an endto-end learning approach to machine translation.
Approach: They propose a capsule network with dynamic routing for linear time Neural Machine Translation . they map the source sentence into a matrix with pre-determined size and apply a deep LSTM network to decode the target sequence from the source representation.
Outcome: The proposed network achieves comparable results with the Transformer system on English-German and English-French tasks.
Teaching Small Language Models to Reason for Knowledge-Intensive Multi-Hop Question Answering (2024.findings-acl)

Copied to clipboard

Challenge: Large language models can teach small language models to solve complex reasoning tasks by Chain-of-thought Distillation (CoTD) e.g., mathematical question answering.
Approach: They propose a method that distills two student models to solve a multi-hop question . they use chain-of-thought distillation to generate step-by-step reasoning paths .
Outcome: The proposed method surpasses existing methods on knowledge-intensive multi-hop questions.
On Transferability of Prompt Tuning for Natural Language Processing (2022.naacl-main)

Copied to clipboard

Challenge: Pre-trained language models (PLMs) can achieve comparable performance to full-parameter fine-tuning by tuning a few soft prompts, but require much more training time than fine-timing.
Approach: They empirically investigate the transferability of soft prompts across different downstream tasks and PLMs to determine what decides prompt transferability.
Outcome: The proposed method can achieve comparable performance to full-parameter fine-tuning by tuning a few soft prompts, but requires much more training time than fine-timing.
Dialogue Structure Annotation for Multi-Floor Interaction (L18-1)

Copied to clipboard

Challenge: Existing annotation schemes do not address dialogue structure.
Approach: They propose an annotation scheme for meso-level dialogue structure that clusters utterances from multiple participants and floors into units according to realization of an initiator's intent.
Outcome: The proposed annotation scheme is used to annotate a corpus of human-robot interaction dialogues.
Multi-Task Pre-Training for Plug-and-Play Task-Oriented Dialogue System (2022.acl-long)

Copied to clipboard

Challenge: Existing pre-trained language models often form a cascaded generation problem . this can lead to error accumulation across different sub-tasks and greater data annotation overhead.
Approach: They propose a plug-and-play model for task-oriented dialogue that learns primary TOD task completion skills from heterogeneous dialog corpora.
Outcome: The proposed model learns primary TOD task completion skills from heterogeneous dialog corpora.
TaCL: Improving BERT Pre-training with Token-aware Contrastive Learning (2022.findings-naacl)

Copied to clipboard

Challenge: Existing pre-trained MLMs produce an anisotropic distribution of token representations . this is not ideal for tasks that require discriminative semantic meanings of distinct tokens - a problem that exists in pre-training models .
Approach: They propose a continual pre-training approach that encourages BERT to learn an isotropic distribution of token representations.
Outcome: The proposed approach improves on a wide range of English and Chinese benchmarks.
A Systematic Investigation of KB-Text Embedding Alignment at Scale (2021.acl-long)

Copied to clipboard

Challenge: Knowledge bases (KBs) and text often contain complementary knowledge.
Approach: They propose a framework for aligning KB and text embeddings for joint reasoning . they also evaluate alignment methods to infuse textual information into KB embeddables .
Outcome: The proposed framework can be used to predict link prediction on emerging entities and events using textual information.

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