Papers by Songming Zhang
Towards Understanding and Improving Knowledge Distillation for Neural Machine Translation (2023.acl-long)
Copied to clipboard
| Challenge: | Existing knowledge distillation techniques for neural machine translation lack special treatment on the top-1 information, which is limiting the potential of KD. |
| Approach: | They propose a method to distill knowledge from top-1 predictions of teachers and a technique to infuse more additional knowledge by distilling on the data without ground-truth targets. |
| Outcome: | The proposed method outperforms the vanilla word-level KD and outperfies the existing methods on three different students with different capacity gaps. |
Conditional Bilingual Mutual Information Based Adaptive Training for Neural Machine Translation (2022.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to improve neural machine translation use token-level adaptive training . however, standard models make predictions on condition of previous contexts . |
| Approach: | They propose a target-context-aware metric which can be supplemented by statistical metrics . they propose an adaptive training approach based on token- and sentence-level CBMI . |
| Outcome: | The proposed model outperforms the Transformer baseline and other similar approaches on English-German and Chinese-English tasks. |
CM-Align: Consistency-based Multilingual Alignment for Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Current large language models (LLMs) show a significant performance gap in alignment between English and other languages. |
| Approach: | They propose a consistency-based method to construct high-quality multilingual preference data for improving multilingual alignment. |
| Outcome: | The proposed method is based on three LLMs and three common tasks and shows that it performs better than current methods. |
Multilingual Knowledge Editing with Language-Agnostic Factual Neurons (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods to update factual knowledge overlook connections of same knowledge between different languages, resulting in knowledge conflicts and limited edit performance. |
| Approach: | They propose a method to edit multilingual knowledge simultaneously that avoids knowledge conflicts and improves edit performance. |
| Outcome: | The proposed method avoids knowledge conflicts and improves edit performance on bi-ZsRE and MzsRE benchmarks. |
A Quality-based Syntactic Template Retriever for Syntactically-Controlled Paraphrase Generation (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing syntactically-controlled paraphrase generation models perform well with human-annotated or well-chosen syntaktic templates. |
| Approach: | They propose a quality-based Syntactic Template Retriever to retrieve templates based on the quality of the to-be-generated paraphrases. |
| Outcome: | The proposed algorithm can generate high-quality paraphrases without sacrificing quality. |
TwinVoice: A Multi-dimensional Benchmark Towards Digital Twins via LLM Persona Simulation (2026.findings-acl)
Copied to clipboard
Bangde Du, Minghao Guo, Songming He, Ziyi Ye, Xi Zhu, Weihang Su, Shuqi Zhu, Yujia Zhou, Yongfeng Zhang, Qingyao Ai, Yiqun Liu
| Challenge: | Existing studies show that advanced LLMs produce text indistinguishable from human writing. |
| Approach: | They propose a benchmark to assess persona simulation across diverse contexts by decomposing the evaluation into six fundamental capabilities including opinion consistency, memory recall, logical reasoning, persona tone, and syntactic style. |
| Outcome: | The proposed model achieves moderate accuracy but falls short of the basic capabilities needed to simulate personas in real-world contexts. |
Think Natively: Unlocking Multilingual Reasoning with Consistency-Enhanced Reinforcement Learning (2026.acl-long)
Copied to clipboard
Xue Zhang, Yunlong Liang, Fandong Meng, Songming Zhang, Kaiyu Huang, Yufeng Chen, Xu Jinan, Jie Zhou
| Challenge: | Current Large Reasoning Models exhibit two critical limitations when processing non-English languages: (1) They struggle to maintain input-output language consistency; (2) They generally perform poorly with wrong reasoning paths and lower answer accuracy compared to English. |
| Approach: | They propose a language-consistency reward and a cross-lingual thinking alignment reward to improve the model's interpretability and accuracy. |
| Outcome: | The proposed model achieves nearly 100% language consistency and superior performance on two multilingual benchmarks (MMATH and PolyMath). |
Less, but Better: Efficient Multilingual Expansion for LLMs via Layer-wise Mixture-of-Experts (2025.acl-long)
Copied to clipboard
| Challenge: | Existing large language models (LLMs) have remarkable ability in high-resource languages, but their performance in multilingual scenarios is still limited. |
| Approach: | They propose a layer-wise expert allocation algorithm to determine the appropriate number of new experts for each layer. |
| Outcome: | The proposed method outperforms the previous state-of-the-art baseline with 60% fewer experts in the single-expansion setting and 33.3% fewer in the lifelong-expanding setting. |
AlignDistil: Token-Level Language Model Alignment as Adaptive Policy Distillation (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for LLM alignment optimize tokens using a sparse, response-level reward or preference annotation. |
| Approach: | They propose an RLHF-equivalent distillation method for token-level reward optimization that incorporates the reward learned by DPO into the RLHG objective and builds a token-based teacher distribution. |
| Outcome: | The proposed method bridges the accuracy gap between the reward from the DPO model and the pure reward model by building a contrastive DPO reward with a normal and a reverse DPO. |
Adversarially Improving NMT Robustness to ASR Errors with Confusion Sets (2022.aacl-short)
Copied to clipboard
| Challenge: | Existing methods for robustness against homophone errors are limited to homophones . substitution errors are the most common errors in NMT models . |
| Approach: | They propose an adversarial example generation method based on confusion sets that contain words easily confusable with a target word by ASR to conduct adversarially training for NMT models. |
| Outcome: | The proposed method improves on the clean test set and can be used in real-world scenarios. |
Dual-Space Knowledge Distillation for Large Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing large language models (LLMs) have strong generalization abilities due to their huge model capacities. |
| Approach: | They propose a dual-space knowledge distillation framework that unifies the output spaces of the two models for KD. |
| Outcome: | The proposed framework outperforms existing white-box KD frameworks on task-agnostic instruction-following benchmarks and can automatically align representations of two models with different vocabularies. |