Papers by Shaolei Zhang
Mitigating the Inconsistency Between Word Saliency and Model Confidence with Pathological Contrastive Training (2022.findings-acl)
Copied to clipboard
| Challenge: | Neural networks are used for various NLP tasks, but their complexity makes them difficult to interpret. |
| Approach: | They propose a framework to mitigate the model pathology and obtain more interpretable models by using contrastive learning and saliency-based samples augmentation to calibrate the sentences representation. |
| Outcome: | The proposed framework can mitigate the model pathology and generate more interpretable models while keeping the model performance. |
LLaMA-Omni 2: LLM-based Real-time Spoken Chatbot with Autoregressive Streaming Speech Synthesis (2025.acl-long)
Copied to clipboard
| Challenge: | LLaMA-Omni 2 is a series of speech language models (SpeechLMs) based on large language models. |
| Approach: | They introduce a series of speech language models capable of real-time speech interaction . LLaMA-Omni 2 trains on 200K multi-turn speech dialogue samples . |
| Outcome: | The proposed speech language models surpass state-of-the-art models on spoken question answering and speech instruction. |
Turning Fixed to Adaptive: Integrating Post-Evaluation into Simultaneous Machine Translation (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to perform adaptive and fixed translations lack evaluation before taking actions. |
| Approach: | They propose a method to perform adaptive translation policy via post-evaluation into fixed policy . their method evaluates rationality of next action by measuring change in source content . |
| Outcome: | The proposed method exceeds strong baselines under all latency. |
Modeling Concentrated Cross-Attention for Neural Machine Translation with Gaussian Mixture Model (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Dot-product attention only considers the pair-wise correlation between words, resulting in dispersion when dealing with long sentences and neglecting source neighboring relationships. |
| Approach: | They propose to model concentrated attention in cross-attention using a Gaussian Mixture Model to model cross- attention in a language model. |
| Outcome: | Experiments on three datasets show that the proposed method outperforms the baseline and has significant improvement on alignment quality, N-gram accuracy, and long sentence translation. |
Gaussian Multi-head Attention for Simultaneous Machine Translation (2022.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for siMT do not explicitly model the alignment to perform the control. |
| Approach: | They propose to model alignment and translation in a unified manner by Gaussian Multi-head Attention (GMA) they propose to integrate alignment-related priors into the translation model to determine final attention. |
| Outcome: | The proposed method outperforms strong baselines on trade-off between translation and latency. |
Learning Optimal Policy for Simultaneous Machine Translation via Binary Search (2023.acl-long)
Copied to clipboard
| Challenge: | Simultaneous machine translation model needs a precise translation policy to achieve good latency-quality trade-offs. |
| Approach: | They propose a method for building the optimal translation policy online via binary search by employing explicit supervision. |
| Outcome: | Experiments on four translation tasks show that the proposed method exceeds strong baselines across all latency scenarios. |
Modeling Dual Read/Write Paths for Simultaneous Machine Translation (2022.acl-long)
Copied to clipboard
| Challenge: | Simultaneous machine translation (SiMT) outputs translation while reading source sentence . existing methods do not direct the read/write path, resulting in poor performance . |
| Approach: | They propose a method which introduces duality constraints to direct the read/write path . they propose to map the read path in two SiMT models to satisfy duality constraint . |
| Outcome: | Experiments on En-Vi and De-En tasks show that the proposed method outperforms baselines under all latency. |
Wait-info Policy: Balancing Source and Target at Information Level for Simultaneous Machine Translation (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to balance source and target information at the token level are limited by the number of received source tokens. |
| Approach: | They propose a Wait-info Policy to balance source and target at the information level . they quantify the amount of info contained in each token and compare it with previous outputs . |
| Outcome: | The proposed method outperforms baselines under and achieves better balance . it is based on comparisons between the total info of previous target outputs and received source inputs . |
StreamSpeech: Simultaneous Speech-to-Speech Translation with Multi-task Learning (2024.acl-long)
Copied to clipboard
| Challenge: | Existing simultaneous translation methods focus on text-to-text and speech-totext translation. |
| Approach: | They propose a Simul-S2ST model that jointly learns translation and simultaneous policy in a unified framework of multi-task learning. |
| Outcome: | The proposed model can perform offline and simultaneous speech recognition, speech translation and speech synthesis via an "All-in-One" seamless model. |
TruthX: Alleviating Hallucinations by Editing Large Language Models in Truthful Space (2024.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) sometimes produce untruthful responses despite knowing the correct knowledge. |
| Approach: | They propose an inference-time intervention method to activate the truthfulness of Large Language Models (LLMs) by editing the features within LLM’s internal representations that govern the truthful. |
| Outcome: | The proposed method improves the truthfulness of 13 advanced LLMs by an average of 20% on TruthfulQA benchmark. |
Truth-Aware Context Selection: Mitigating Hallucinations of Large Language Models Being Misled by Untruthful Contexts (2024.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are easily misled by untruthful contexts provided by users or knowledge augmentation tools, leading to hallucinations. |
| Approach: | They propose a lightweight method to adaptively recognize and mask untruthful context from the inputs and a new evaluation metric to further study the LLMs’ ability to accept truthful information and resist untrusted information. |
| Outcome: | The proposed method can detect and mask untruthful context from the inputs and significantly improve the quality of LLMs’ responses when presented with misleading information. |
A Non-autoregressive Generation Framework for End-to-End Simultaneous Speech-to-Any Translation (2024.acl-long)
Copied to clipboard
| Challenge: | Existing translation pipelines require additional cascade components to achieve speech-to-speech translation. |
| Approach: | They propose a non-autoregressive generation framework for simultaneous speech translation . it integrates both text-to-text and speech-tospeech tasks into a unified framework . |
| Outcome: | The proposed framework outperforms state-of-the-art models in speech-to-text and speech- to-speech tasks. |
Non-autoregressive Streaming Transformer for Simultaneous Translation (2023.emnlp-main)
Copied to clipboard
| Challenge: | Simultaneous machine translation models are trained to strike a balance between latency and translation quality. |
| Approach: | They propose a non-autoregressive streaming Transformer which generates blank tokens and decodes repetitive tokens to adjust its READ/WRITE strategy flexibly. |
| Outcome: | The proposed model outperforms previous strong autoregressive models on various benchmarks on siMT. |
AlignX: Advancing Multilingual Large Language Models with Multilingual Representation Alignment (2025.emnlp-main)
Copied to clipboard
| Challenge: | Multilingual large language models (LLMs) possess impressive multilingual understanding and generation capabilities, but performance and cross-lingual alignment often lag for non-dominant languages. |
| Approach: | They propose a representation-level framework to enhance multilingual performance of pre-trained LLMs by integrating multilingual semantic alignment and language feature integration. |
| Outcome: | The proposed framework improves multilingual capability of pre-trained LLMs by bringing representations closer and improving cross-lingual alignment. |
Decoder-only Streaming Transformer for Simultaneous Translation (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods for siMT focus on the Encoder-Decoder architecture, but there are limitations in training and inference. |
| Approach: | They propose a model that generates translation while reading source tokens . they propose Streaming Self-Attention mechanism tailored for the Decoder-only architecture . |
| Outcome: | The proposed model achieves state-of-the-art performance on three translation tasks. |
Can We Achieve High-quality Direct Speech-to-Speech Translation without Parallel Speech Data? (2024.acl-long)
Copied to clipboard
| Challenge: | Existing two-pass direct speech-to-speech translation models require parallel speech data to train, which is challenging to collect. |
| Approach: | They propose a two-pass direct speech-to-speech translation (S2ST) model that decomposes the task into speech- to-text translation (s2TT) and text-tospech (TTS) they propose 'composer' S2ST model that integrates pretrained S2TT and TTS models into a direct S2 ST model. |
| Outcome: | The proposed model integrates pretrained S2TT and TTS models into a direct S2ST model without parallel speech data. |
RealMem: Benchmarking LLMs in Real-World Memory-Driven Interaction (2026.findings-acl)
Copied to clipboard
Haonan Bian, Zhiyuan Yao, Sen Hu, Zishan Xu, Shaolei Zhang, Yifu Guo, Ziliang Yang, Xueran Han, Huacan Wang, Ronghao Chen
| Challenge: | Existing benchmarks focus on casual conversation or task-oriented dialogue, failing to capture “long-term project-oriented” interactions where agents must track evolving goals. |
| Approach: | They propose a benchmark that simulates the dynamic evolution of memory in real-world projects. |
| Outcome: | The proposed benchmarks simulate the dynamic evolution of memory in real-world projects. |
Reducing Position Bias in Simultaneous Machine Translation with Length-Aware Framework (2022.acl-long)
Copied to clipboard
| Challenge: | Existing methods for simultaneous machine translation (SiMT) are more challenging since the source sentence is always incomplete during translating. |
| Approach: | They propose a framework to reduce the position bias by bridging the structural gap between SiMT and full-sentence MT. |
| Outcome: | The proposed framework reduces the position bias by bridging the structural gap between SiMT and full-sentence MT. |
Universal Simultaneous Machine Translation with Mixture-of-Experts Wait-k Policy (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for simultaneous machine translation require multiple models for different latency levels, resulting in large computational costs. |
| Approach: | They propose a universal SiMT model with Mixture-of-Experts Wait-k Policy to achieve the best translation quality under arbitrary latency with only one model. |
| Outcome: | The proposed model outperforms all the strong baselines under different latency levels including the state-of-the-art adaptive policy. |
Simultaneous Machine Translation with Tailored Reference (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing SiMT models are trained using the same reference disregarding the varying amounts of available source information at different latency. |
| Approach: | They propose a method that provides tailored reference for the SiMT models trained at different latency by rephrasing ground-truth to the tailored reference. |
| Outcome: | The proposed method achieves state-of-the-art translation performance on three translation tasks. |
Information-Transport-based Policy for Simultaneous Translation (2022.emnlp-main)
Copied to clipboard
| Challenge: | Simultaneous translation (ST) outputs translation while receiving source inputs . low latency restriction restricts ST to translating target tokens based on current received source tokens. |
| Approach: | They propose a system that outputs translation while receiving source inputs . it uses a read/write policy to decide whether to translate a target token or wait for the next source token . |
| Outcome: | The proposed model outperforms baselines and achieves state-of-the-art on text-to-text and speech-to text tasks. |
IG-Pruning: Input-Guided Block Pruning for Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for efficient inference rely on fixed block masks, which can lead to suboptimal performance. |
| Approach: | They propose an input-aware block-wise pruning method that dynamically selects layer masks at inference time. |
| Outcome: | The proposed method outperforms state-of-the-art static depth pruning methods . it is particularly suitable for resource-constrained deployment scenarios . |
End-to-End Simultaneous Speech Translation with Differentiable Segmentation (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to perform simultaneous speech translation always separate segmentation from the underlying model. |
| Approach: | They propose to use Differentiable Segmentation (DiSeg) to learn segmentation from the translation model. |
| Outcome: | Experimental results show that the proposed model can learn segmentation from the translation model. |