Papers by Shaolin Zhu

MACS: Modality-Aware Capacity Scaling for Efficient Multimodal MoE Inference (2026.acl-long)

Copied to clipboard

Challenge:	Existing methods for expert parallelism inference suffer from a significant efficiency bottleneck . existing methods fail to address information heterogeneity and modality dynamics .
Approach:	They propose a training-free inference framework that scales experts without training . they propose an Entropy-Weighted Load mechanism to quantify the semantic value of visual tokens .
Outcome:	Experiments show that MACS outperforms existing methods on multimodal benchmarks.

FuxiTranyu: A Multilingual Large Language Model Trained with Balanced Data (2024.emnlp-industry)

Copied to clipboard

Haoran Sun, Renren Jin, Shaoyang Xu, Leiyu Pan, null Supryadi, Menglong Cui, Jiangcun Du, Yikun Lei, Lei Yang, Ling Shi, Juesi Xiao, Shaolin Zhu, Deyi Xiong

Language Models

Challenge:	Large language models exhibit significant performance discrepancies between high- and low-resource languages.
Approach:	They present an open-source multilingual LLM with 8 billion parameters and a multilingual instruction dataset.
Outcome:	The proposed model achieves consistent multilingual representations across languages.

CCSRD: Content-Centric Speech Representation Disentanglement Learning for End-to-End Speech Translation (2023.findings-emnlp)

Copied to clipboard

Xiaohu Zhao, Haoran Sun, Yikun Lei, Shaolin Zhu, Deyi Xiong

Semantic Text Processing

Challenge:	Existing speech-to-text translation models can extract features from speech inputs, but they may include non-linguistic speech factors such as pitch, timbre and speaker identity.
Approach:	They propose a content-centric speech representation disentanglement learning framework for speech translation that decomposes speech representations into content representations and non-linguistic representations via representation disentanglement learning.
Outcome:	The proposed framework outperforms state-of-the-art speech translation models and cascaded models on five translation directions.

Towards Robust In-Context Learning for Machine Translation with Large Language Models (2024.lrec-main)

Copied to clipboard

Shaolin Zhu, Menglong Cui, Deyi Xiong

Responsible & Trustworthy NLP

Challenge:	Experimental results demonstrate the effectiveness of our method, particularly in domain adaptation.
Approach:	They propose a method to retrieve translation pairs as demonstrations from an additional datastore to guide translation without updating the LLMs.
Outcome:	The proposed method reduces noise and improves translation performance in domain adaptation.

Towards a Deep Understanding of Multilingual End-to-End Speech Translation (2023.findings-emnlp)

Copied to clipboard

Haoran Sun, Xiaohu Zhao, Yikun Lei, Shaolin Zhu, Deyi Xiong

Multilinguality

Challenge:	Recent years have witnessed the rapid development of end-to-end speech-totext translation (ST) which has demonstrated remarkable performance and outperformed conventional cascaded systems.
Approach:	They employ Singular Value Canonical Correlation Analysis to analyze representations learnt in a multilingual end-to-end speech translation model trained over 22 languages.
Outcome:	The proposed approach outperforms existing cascaded systems in predicting phonetic features and improves translation quality.

MIT-10M: A Large Scale Parallel Corpus of Multilingual Image Translation (2025.coling-main)

Copied to clipboard

Bo Li, Shaolin Zhu, Lijie Wen

Multilinguality

Challenge:	Existing datasets suffer from limitations in scale, diversity, and quality, hindering the development and evaluation of IT models.
Approach:	They propose a large-scale parallel corpus of multilingual image translation with over 10M image-text pairs derived from real-world data.
Outcome:	The proposed model performs better in tackling challenging and complex image translation tasks in the real world.

Parallel sentences mining with transfer learning in an unsupervised setting (2021.naacl-srw)

Copied to clipboard

Yu Sun, Shaolin Zhu, Feng Yifan, Chenggang Mi

Language Models

Challenge:	Existing methods to mine parallel sentences in low-resource environments are not suitable for many low-level language pairs.
Approach:	They propose an approach based on transfer learning to mine parallel sentences in an unsupervised setting using bilingual corpora of low-resource language pairs.
Outcome:	The proposed model improves the performance of mined parallel sentences at two real-world low-resource language pairs compared with previous methods.

LANDeRMT: Dectecting and Routing Language-Aware Neurons for Selectively Finetuning LLMs to Machine Translation (2024.acl-long)

Copied to clipboard

Shaolin Zhu, Leiyu Pan, Bo Li, Deyi Xiong

Multilinguality

Challenge:	Existing studies have shown promising results in multilingual translation with limited bilingual supervision.
Approach:	They propose a Language-Aware Neuron Detecting and Routing framework that fine tunes LLMs to Machine Translation with diverse translation training data.
Outcome:	The proposed framework selectively finetunes LLMs to MT tasks with diverse translation training data.

TENP: Trapezoidal Expert Neuron Pruning For Mixture-of-Experts (2026.findings-acl)

Copied to clipboard

Jiangyang He, Shaolin Zhu, Deyi Xiong

Green & Sustainable NLP

Challenge:	Existing compression approaches remove entire experts, disrupting routing topology and harming performance, or rely on unstructured weight pruning with limited practical efficiency.
Approach:	They propose a structured Trapezoidal Expert Neuron Prunning framework that uses a trapezoidal pattern to identify and retain important experts while applying expert neuron pruning (ENP) to less important experts.
Outcome:	The proposed framework outperforms the full-parameter model by 10% on code generation tasks under a sparse activation of experts and a 40% routing sparsity.

DCIS: Efficient Length Extrapolation of LLMs via Divide-and-Conquer Scaling Factor Search (2025.emnlp-main)

Copied to clipboard

Lei Yang, Shaoyang Xu, Jianxiang Peng, Shaolin Zhu, Deyi Xiong

Semantic Text Processing

Challenge:	Existing frameworks for large language models with context length limitations are suboptimal for initialization and fine-tuning.
Approach:	They propose a RoPE-based fine-tuning framework that strategically determines the best scaling factors for LLMs by a Divide-and-Conquer Incremental Search algorithm.
Outcome:	The proposed framework mitigates performance decay at extended target lengths and can perform effectively without fine-tuning.

AdaDPI: Document-level Translation Adaptive Agent via Dynamic Parametric Internalization (2026.acl-long)

Copied to clipboard

Hong Ren, Liting Deng, Shaolin Zhu, Deyi Xiong

Machine Translation

Challenge:	Existing solutions, such as memory-based agents, rely on explicit context concatenation, which leads to context dilution, high inference latency, and superficial knowledge integration.
Approach:	They propose an adaptive agentic framework that shifts the DocMT paradigm from static retrieval to dynamic parametric internalization.
Outcome:	Extensive experiments on the discourse-rich GuoFeng and IWSLT2017 datasets show that AdaDPI outperforms the SoTA baselines by more than 5 points on the consistency metric.

MMNMT: Modularizing Multilingual Neural Machine Translation with Flexibly Assembled MoE and Dense Blocks (2023.emnlp-main)

Copied to clipboard

Shangjie Li, Xiangpeng Wei, Shaolin Zhu, Jun Xie, Baosong Yang, Deyi Xiong

Multilinguality

Challenge:	Mixture-of-Experts (MoE) based sparse architectures are prone to overfitting on low-resource language translation.
Approach:	They propose a modularized MNMT framework that flexibly assembles dense and MoE-based sparse modules to achieve the best of both worlds.
Outcome:	The proposed framework outperforms existing models on low-resource language translation and zero-shot translation on benchmark datasets.

Unveiling Multimodal Processing: Exploring Activation Patterns in Multimodal LLMs for Interpretability and Efficiency (2025.findings-emnlp)

Copied to clipboard

Chuan Wu, Meng Su, Youxuan Fang, Shaolin Zhu

Responsible & Trustworthy NLP

Challenge:	Recent advances in multimodal large language models have remained opaque.
Approach:	They propose a method to convert dense MLLMs into fine-grained Mixture-of-Experts architectures.
Outcome:	The proposed method outperforms random expert pruning and sparse activation and model pruning.

CKDST: Comprehensively and Effectively Distill Knowledge from Machine Translation to End-to-End Speech Translation (2023.findings-acl)

Copied to clipboard

Yikun Lei, Zhengshan Xue, Xiaohu Zhao, Haoran Sun, Shaolin Zhu, Xiaodong Lin, Deyi Xiong

Language Models

Challenge:	End-to-end speech-totext translation (ST) data are limited due to the limited resources.
Approach:	They propose a knowledge distillation framework for speech translation that integrates knowledge from machine translation and decouples knowledge from non-target class knowledge.
Outcome:	The proposed framework outperforms state-of-the-art models on a benchmark dataset.

PEIT: Bridging the Modality Gap with Pre-trained Models for End-to-End Image Translation (2023.acl-long)

Copied to clipboard

Shaolin Zhu, Shangjie Li, Yikun Lei, Deyi Xiong

Multimodality

Challenge:	Image translation is a task that translates an image containing text in the source language to the target language.
Approach:	They propose an end-to-end image translation framework that bridges the modality gap between visual inputs and textual inputs/outputs of machine translation (MT).
Outcome:	The proposed framework outperforms existing models on a large-scale image translation corpus . it significantly outperformed both cascaded and strong models on the e-commerce domain .

SARA: Unlocking Multilingual Knowledge in Mixture-of-Experts via Semantically Anchored Routing Alignment (2026.findings-acl)

Copied to clipboard

Tianyu Dong, Yangyang Liu, Jiang Zhou, Xinwei Wu, Xiaohu Zhao, Hao Wang, Heng Liu, Linlong Xu, Longyue Wang, Weihua Luo, Shaolin Zhu, Deyi Xiong

Multilinguality

Challenge:	Low-resource language tokens are often routed to different experts than those activated by high-resourced inputs, which hinders their efficacy in multilingual contexts.
Approach:	They propose a framework to transfer specialized capabilities from high-resource languages as anchors to low-resourced languages by using a symmetric Jensen-Shannon constraint.
Outcome:	The proposed framework outperforms standard instruction tuning on 5 low-resource languages and 3 benchmarks.

Efficiently Exploring Large Language Models for Document-Level Machine Translation with In-context Learning (2024.findings-acl)

Copied to clipboard

Menglong Cui, Jiangcun Du, Shaolin Zhu, Deyi Xiong

Responsible & Trustworthy NLP

Challenge:	Existing studies on sentence-level translation have focused on document level machine translation (DOCMT) document level translation is a complex task different from sentence- level translation.
Approach:	They propose a Context-Aware Prompting method which generates more accurate, coherent translations via in-context learning.
Outcome:	The proposed method is effective in literary translation tasks and zero pronoun translation tasks.

MLAS-LoRA: Language-Aware Parameters Detection and LoRA-Based Knowledge Transfer for Multilingual Machine Translation (2025.acl-long)

Copied to clipboard

Tianyu Dong, Bo Li, Jinsong Liu, Shaolin Zhu, Deyi Xiong

Multilinguality

Challenge:	Large language models (LLMs) have demonstrated strong performance even with limited parallel data.
Approach:	They propose a multiple language-aware LoRA knowledge transfer framework that selectively adapts LLMs to MT by transferring knowledge from a large teacher to a small student model.
Outcome:	The proposed framework outperforms baseline models on multilingual language pairs by +1.7 BLEU on average.

CYCLE-INSTRUCT: Fully Seed-Free Instruction Tuning via Dual Self-Training and Cycle Consistency (2025.emnlp-main)

Copied to clipboard

Zhanming Shen, Hao Chen, Yulei Tang, Shaolin Zhu, Wentao Ye, Xiaomeng Hu, Haobo Wang, Gang Chen, Junbo Zhao

Machine Translation

Challenge:	Existing methods for instruction tuning rely on expensive human-annotated seed data or powerful external teacher models.
Approach:	They propose a framework that achieves fully seed-free instruction tuning by employing a dual self-training loop where two models are bootstrapped solely from raw, unlabeled text.
Outcome:	The proposed framework outperforms seed-driven back-translation baselines and achieves comparable performance to strongly supervised methods.