Papers by Chenhui Chu
Text Classification with Negative Supervision (2020.acl-main)
Copied to clipboard
| Challenge: | Existing models for text representations have shown state-of-the-art performance on text classification tasks, however, the discrepancy between semantic similarity of texts and labelling standards affects classifiers. |
| Approach: | They propose a simple multitask learning model that uses negative supervision to generate distinct representations for texts with different labels. |
| Outcome: | The proposed model outperforms state-of-the-art models on classification tasks in three different languages. |
Video-guided Machine Translation with Spatial Hierarchical Attention Network (2021.acl-srw)
Copied to clipboard
| Challenge: | Existing studies use pretrained motion detection models as verb sense ambiguity representations to solve the verb sense problem. |
| Approach: | They propose to use video contents as auxiliary information to address the word sense ambiguity problem in machine translation. |
| Outcome: | Experiments on the VATEX dataset show that the proposed system achieves 35.86 BLEU-4 score, which is 0.51 score higher than the single model of the SOTA method. |
Multilingual Neural Machine Translation (2020.coling-tutorials)
Copied to clipboard
| Challenge: | In this tutorial, we will cover the latest advances in NMT to enhance low-resource translation. |
| Approach: | They will cover the latest advances in NMT approaches that leverage multilingualism . they will focus on topics such as language divergence, transfer learning and pivoting . |
| Outcome: | This tutorial will cover the latest advances in NMT to enhance low-resource translation models. |
Do LLMs Align Human Values Regarding Social Biases? Judging and Explaining Social Biases with LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models can lead to undesired consequences when misaligned with human values . previous studies have shown misalignment of LLMs with human value using expert-designed or agent-based emulated bias scenarios . |
| Approach: | They investigate whether large language models (LLMs) are misaligned with human values . they find no significant differences in understanding of HVSB between LLMs . |
| Outcome: | The results show that large language models do not have lower misalignment rates and attack success rates . the study also shows that smaller language models have the ability to explain HVSB . |
Attending Self-Attention: A Case Study of Visually Grounded Supervision in Vision-and-Language Transformers (2021.acl-srw)
Copied to clipboard
| Challenge: | a growing body of research has been focused on what attention heads learn during the pre-training of visual grounded language models. |
| Approach: | They propose to use visual grounding to supervise attention directly to learn visual ground. |
| Outcome: | The proposed method improves the performance of a state-of-the-art visual grounded language model on vision-and-language tasks. |
Identifying Source Language Expressions for Pre-editing in Machine Translation (2024.lrec-main)
Copied to clipboard
| Challenge: | MT-mediated communication can benefit from pre-editing source language texts to ensure accurate transmission of intended meaning in the target language. |
| Approach: | They hypothesize that such expressions tend to be distinctive features of texts originally written in the source language rather than translations generated from the target language into the source languages. |
| Outcome: | The proposed method identified characteristic expressions of the native language despite the noise and inherent nuances of the task. |
MPTc-Bench: Measuring Cross-market Generative Ability of Vision-Language Models via Movie Poster Transcreation (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent work adapts textual transcreation to image editing and formulates image transcreations to better match a target audience while preserving meaning. |
| Approach: | They propose a two-stage planner-editor pipeline in which an VLM planner specifies executable edits and an image editor renders them. |
| Outcome: | The proposed model can transcreate a visual asset for a different market while preserving its identity while matching market-specific design preferences and multilingual typography. |
Rapidly Developing High-quality Instruction Data and Evaluation Benchmark for Large Language Models with Minimal Human Effort: A Case Study on Japanese (2024.lrec-main)
Copied to clipboard
| Challenge: | Recent advances in large language models (LLMs) have aimed to refine their capacity to accurately follow human instructions and navigate intricate scenarios. |
| Approach: | They propose a method that uses a set of instructions to translate English into Japanese and then generates Japanese instruction data using GPT-4. |
| Outcome: | The proposed method outperforms Japanese-Alpaca models in the evaluation benchmarks without human references. |
Flexible Visual Grounding (2022.acl-srw)
Copied to clipboard
| Challenge: | Existing visual grounding datasets require queries to be answerable, but in multimedia data, many entities cannot be grounded to the image, resulting in unanswerable visual ground. |
| Approach: | They propose a method to ground to a pseudo image region for unanswerable queries . they add a query that cannot be grounded to the image and train it to ground . |
| Outcome: | The proposed model can handle answerable and unanswerable visual grounding with high accuracy on the proposed datasets. |
WRIME: A New Dataset for Emotional Intensity Estimation with Subjective and Objective Annotations (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing studies on emotion analysis use subjective emotional intensity labels by the writers and objective ones by the readers. |
| Approach: | They annotate 17,000 SNS posts with both the writer's subjective emotional intensity and the reader's objective emotional intensity to construct a Japanese emotion analysis dataset. |
| Outcome: | The results show that the reader cannot fully detect the emotions of the writer, especially anger and trust. |
What Language Do Non-English-Centric Large Language Models Think in? (2025.findings-acl)
Copied to clipboard
Chengzhi Zhong, Qianying Liu, Fei Cheng, Junfeng Jiang, Zhen Wan, Chenhui Chu, Yugo Murawaki, Sadao Kurohashi
| Challenge: | Despite their robust performance in English, these models often exhibit reduced proficiency in non-English languages, and their outputs may reflect an inherent bias toward English-centric perspectives. |
| Approach: | They categorize non-English-centric large language models into two groups: CPMs and BLMs, which are pre-trained on a balanced mix of multiple languages from scratch. |
| Outcome: | The proposed models exhibit a pronounced internal preference for English tokens when projected into the vocabulary space. |
iParaphrasing: Extracting Visually Grounded Paraphrases via an Image (C18-1)
Copied to clipboard
| Challenge: | iParaphrasing extracts visually grounded paraphrases, which are different phrasal expressions describing the same visual concept in an image. |
| Approach: | They propose a task to extract visually grounded paraphrases from images . they propose to model the similarity between the extracted VGPs using existing methods . |
| Outcome: | The proposed task extracts visually grounded paraphrases from images . the proposed method has the potential to improve multimodal language and image tasks . |
Vision-Enhanced Semantic Entity Recognition in Document Images via Visually-Asymmetric Consistency Learning (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing models train a visual encoder with weak cross-modal supervision signals, resulting in a limited capacity to capture non-textual features and suboptimal performance. |
| Approach: | They propose a Visually-Asymmetric coNsistenCy Learning approach that enhances the model’s ability to capture fine-grained visual and layout features through the incorporation of color priors. |
| Outcome: | The proposed approach outperforms the strong LayoutLM series baseline on benchmark datasets and provides insights for optimizing model performance. |
Language Lives in Sparse Dimensions: Toward Interpretable and Efficient Multilingual Control for Large Language Models (2026.eacl-long)
Copied to clipboard
| Challenge: | Prior studies show that large language models map multilingual content into English-aligned representations at intermediate layers before projecting them back into target-language token spaces in the later layers. |
| Approach: | They propose a method to identify and manipulate dimensions that are sparse and sparsity-based . they propose to use as few as 50 sentences of either parallel or monolingual data to manipulate these dimensions . |
| Outcome: | Experiments on a multilingual generation control task show the interpretability of these dimensions. |
A Survey of Domain Adaptation for Neural Machine Translation (C18-1)
Copied to clipboard
| Challenge: | Neural machine translation (NMT) is a deep learning based approach for machine translation. |
| Approach: | They propose to use a deep learning approach to train machine translation in scenarios where large-scale parallel corpora are available. |
| Outcome: | The proposed approach yields the state-of-the-art translation performance in resource rich scenarios. |
When Large Language Models Meet Speech: A Survey on Integration Approaches (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in large language models have spurred interest in expanding their application beyond text-based tasks. |
| Approach: | They propose to categorize the integration of speech with LLMs into three main approaches . they demonstrate how these methods are applied across various speech-related applications . |
| Outcome: | The proposed methods are applied across speech-related applications and highlight the challenges in this field to offer inspiration for future research. |
Speech-Hands: A Self-Reflection Voice Agentic Approach to Speech Recognition and Audio Reasoning with Omni Perception (2026.acl-long)
Copied to clipboard
Zhen Wan, Chao-Han Huck Yang, Jinchuan Tian, Hanrong Ye, Ankita Pasad, Szu-Wei Fu, Arushi Goel, Ryo Hachiuma, Shizhe Diao, Kunal Dhawan, Sreyan Ghosh, Yusuke Hirota, Zhehuai Chen, Rafael Valle, Chenhui Chu, Shinji Watanabe, Boris Ginsburg, Yu-Chiang Frank Wang
| Challenge: | naively fine-tuning an omni-model on speech recognition and external sound understanding tasks often degrades performance . Xie and Wu's framework, Speech-Hands, recasts the problem as an explicit self-reflection decision. |
| Approach: | They propose a voice-agentic framework that learns one critical omni-understanding skill: trusting itself versus external audio perception. |
| Outcome: | The proposed framework outperforms baseline models on the OpenASR leaderboard by 12.1% WER and high F1 on audio QA decisions. |
Abstractive Multi-Video Captioning: Benchmark Dataset Construction and Extensive Evaluation (2024.lrec-main)
Copied to clipboard
| Challenge: | Abstractive multi-video captioning focuses on abstracting multiple videos with natural language. |
| Approach: | They propose a task that generates an abstract caption of shared video content . they propose end-to-end and cascade approaches to abstractive multi-video captioning . |
| Outcome: | The proposed task generates an abstract caption of shared content in a video group containing multiple videos. |
SpeechIQ: Speech-Agentic Intelligence Quotient Across Cognitive Levels in Voice Understanding by Large Language Models (2025.acl-long)
Copied to clipboard
Zhen Wan, Chao-Han Huck Yang, Yahan Yu, Jinchuan Tian, Sheng Li, Ke Hu, Zhehuai Chen, Shinji Watanabe, Fei Cheng, Chenhui Chu, Sadao Kurohashi
| Challenge: | SIQ quantifies voice understanding abilities and provides unified comparisons between cascaded methods and end-to-end models. |
| Approach: | They propose a human cognition-inspired evaluation pipeline for voice understanding large language models (LLM_Voice) that quantifies voice understanding abilities and provides unified comparisons between cascaded methods and end-to-end models. |
| Outcome: | The proposed framework quantifies voice understanding abilities and provides unified comparisons between cascaded methods and end-to-end models, identifies annotation errors in existing benchmarks, and detects hallucinations in LLM_Voice. |
Recursive Neural Network Based Preordering for English-to-Japanese Machine Translation (P18-3)
Copied to clipboard
| Challenge: | Existing methods for preordering require a manual feature design, making language dependent design difficult. |
| Approach: | They propose a preordering method with recursive neural networks that learn features from raw inputs. |
| Outcome: | The proposed method is comparable to the state-of-the-art method but without a manual feature design. |
VISA: An Ambiguous Subtitles Dataset for Visual Scene-aware Machine Translation (2022.lrec-1)
Copied to clipboard
| Challenge: | Existing multimodal machine translation datasets contain images and video captions or general subtitles which rarely contain linguistic ambiguity. |
| Approach: | They propose a dataset that consists of Japanese-English parallel sentence pairs and corresponding video clips. |
| Outcome: | The proposed dataset is challenging for the latest MMT system and can facilitate MMT research. |
Constructing a Public Meeting Corpus (2020.lrec-1)
Copied to clipboard
Koji Tanaka, Chenhui Chu, Haolin Ren, Benjamin Renoust, Yuta Nakashima, Noriko Takemura, Hajime Nagahara, Takao Fujikawa
| Challenge: | Existing corpora are created from text that has already been digitized. |
| Approach: | They propose a full pipeline of analysis of a large corpus about a century of public meeting in historical Australian news papers, from construction to visual exploration. |
| Outcome: | The proposed method achieves a high recall rate and an F-score of 87.8% on a historical Australian newspaper database. |
CoVoGER: A Multilingual Multitask Benchmark for Speech-to-text Generative Error Correction with Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models can fix recognition or translation errors that traditional rescoring cannot fix. |
| Approach: | They propose a benchmark for GER that covers both ASR and speech-to-text translation across 15 languages and 28 language pairs. |
| Outcome: | The proposed benchmark is built on common voice 20.0 and CoVoST-2 with Whisper and SeamlessM4T. |
Progressive LoRA for Multimodal Continual Instruction Tuning (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to MCIT address Catastrophic Forgetting and Knowledge Transfer (KT) but using a fixed number of shared LoRA blocks across tasks can lead to knowledge interference. |
| Approach: | They propose a framework that uses a fixed number of shared LoRA blocks to reduce knowledge interference. |
| Outcome: | The proposed framework outperforms existing approaches on the latest MCIT benchmark. |
Lightweight Cross-Lingual Sentence Representation Learning (2021.acl-long)
Copied to clipboard
| Challenge: | Existing models for learning fixed-dimensional cross-lingual sentence representations are impractical due to memory limitations. |
| Approach: | They propose a lightweight dual-transformer architecture with just 2 layers for generating memory-efficient cross-lingual sentence representations. |
| Outcome: | The proposed model improves performance on training tasks and improves memory efficiency. |
Annotation of Adverse Drug Reactions in Patients’ Weblogs (2020.lrec-1)
Copied to clipboard
| Challenge: | Adverse drug reactions are a severe problem that significantly degrade quality of life and make the therapeutic approach unacceptable. |
| Approach: | They crawled patient’s weblog articles shared on an online patient-networking platform and annotated the effects of drugs therein reported. |
| Outcome: | The proposed dataset is unique for the richness of annotated information, including detailed descriptions of drug reactions with full context. |
ARKitSceneRefer: Text-based Localization of Small Objects in Diverse Real-World 3D Indoor Scenes (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing datasets for 3D referring expression comprehension cover large objects and small objects, such as cooking tools and office supplies. |
| Approach: | They propose a 3D referring expression comprehension dataset that uses 3D scenes to ground text representations onto objects in 3D environments. |
| Outcome: | The proposed dataset covers 15k objects of 1,605 indoor scenes and is significantly larger than existing datasets. |
Generative Error Correction for Emotion-aware Speech-to-text Translation (2025.findings-acl)
Copied to clipboard
| Challenge: | Despite recent advances in speech-to-text translation, the impact of the emotion content has been overlooked. |
| Approach: | They propose to use generative error correction (GER) to generate the translation based on the decoded N-best hypotheses and combine emotion and sentiment labels into the LLM finetuning process to enable the model to consider the emotion content. |
| Outcome: | The proposed model can translate speech in English-Chinese using GER and emotion and sentiment labels. |
Towards Human-Like Machine Comprehension: Few-Shot Relational Learning in Visually-Rich Documents (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing document AI approaches fail to consider key-value relations in visually-rich documents . a few-shot approach is proposed to extract key- value relation triplets in VRDs . |
| Approach: | They propose a few-shot relational learning approach targeting the extraction of key-value relation triplets in Visually-Rich Documents. |
| Outcome: | The proposed method outperforms existing methods in visually-rich documents. |
MELD-ST: An Emotion-aware Speech Translation Dataset (2024.findings-acl)
Copied to clipboard
Sirou Chen, Sakiko Yahata, Shuichiro Shimizu, Zhengdong Yang, Yihang Li, Chenhui Chu, Sadao Kurohashi
| Challenge: | Emotion plays a crucial role in human conversation. |
| Approach: | They present a MELD-ST dataset for the emotion-aware speech translation task . they show that fine-tuning with emotion labels can enhance translation performance . |
| Outcome: | The proposed dataset shows that fine tuning with emotion labels can improve translation performance in some settings. |
When do Contrastive Word Alignments Improve Many-to-many Neural Machine Translation? (2022.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods to improve pre-training for many-to-many neural machine translation use manual cleaning of bilingual dictionaries, which are unavailable for most language pairs. |
| Approach: | They propose a word-level contrastive objective to leverage word alignments for many-to-many neural machine translation (NMT) Empirical results show that this leads to 0.8 BLEU gains for several language pairs. |
| Outcome: | Empirical results show that the proposed objective leads to 0.8 BLEU gains for several language pairs. |
DocTrack: A Visually-Rich Document Dataset Really Aligned with Human Eye Movement for Machine Reading (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Document AI models that can read visually rich documents have a long way to go before they can read them as accurately, continuously, and flexibly as humans do. |
| Approach: | They propose a visually-rich document dataset that aligns with human eye-movement information using eye-tracking technology. |
| Outcome: | The proposed dataset can help in designing better document AI models and human reading robots in the future. |
Video-Helpful Multimodal Machine Translation (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing multimodal machine translation datasets contain images and video captions or instructional video subtitles, which rarely contain linguistic ambiguity. |
| Approach: | They propose an MMT dataset that contains ambiguous subtitles and a video-helpful evaluation set. |
| Outcome: | The proposed model performs significantly better than existing models on ambiguous subtitles dataset . it is based on a training set and video-helpful evaluation set . |
Understanding the Prompt Sensitivity (2026.acl-long)
Copied to clipboard
| Challenge: | Prompt sensitivity is a measure of how strongly the output of a large language model (LLM) depends on the exact wording of its input prompt. |
| Approach: | They consider LLMs as multivariate functions and perform a first-order Taylor expansion to analyze the relationship between meaning-preserving prompts, their gradients, and log probabilities of the model’s next token. |
| Outcome: | The proposed model disperses meaning-preserving inputs, making it difficult to reduce to 0. The proposed models also dispersing prompt variants are more likely to introduce prompt sensitivity risks in LLMs. |
Towards Speech Dialogue Translation Mediating Speakers of Different Languages (2023.findings-acl)
Copied to clipboard
| Challenge: | a new task is proposed to mediate speakers of different languages using speech dialogue translation . we consider context as an important aspect that needs to be addressed in this task . speech translation (ST) has also recently shown success in monologue translation - but no study has focused on ST of dialogues . |
| Approach: | They propose a task to mediate speakers of different languages using speech dialogue translation . they construct a speechBSD dataset and conduct baseline experiments . |
| Outcome: | The proposed task mediates speakers of different languages using speech dialogue translation dataset . it shows that bilingual context performs better in our settings . |
Rethinking Meeting Effectiveness: A Benchmark and Framework for Temporal Fine-grained Automatic Meeting Effectiveness Evaluation (2026.acl-long)
Copied to clipboard
| Challenge: | Current methods for evaluating meeting effectiveness rely on post-hoc surveys that yield a single coarse-grained score for an entire meeting. |
| Approach: | They propose a new paradigm for evaluating meeting effectiveness centered on novel criteria and temporal fine-grained approach . they introduce a meta-evaluation dataset containing 2,459 human-annotated segments from 130 AMI Corpus meetings. |
| Outcome: | The proposed framework can be used to evaluate meeting effectiveness across different meeting types and from business scenarios to unstructured discussions. |
MM-LLMs: Recent Advances in MultiModal Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | MultiModal Large Language Models (MM-LLMs) have undergone significant advances in the past year . traditional MM models incur substantial computational costs, especially when trained from scratch . |
| Approach: | They propose a taxonomy encompassing 126 MM-LLMs and summarize key training recipes to enhance their potency. |
| Outcome: | The proposed models preserve the reasoning and decision-making capabilities of LLMs and empower diverse range of MM tasks. |
Exploring the Impact of Layer Normalization for Zero-shot Neural Machine Translation (2023.acl-short)
Copied to clipboard
| Challenge: | Recent studies have shown that layer normalization (LayerNorm) overfits training data and therefore has low generalizability for ZST. |
| Approach: | They propose to use the Transformer architecture to set the default layer normalization setting for zero-shot translation (ZST) they also propose to set LayerNorm after residual connections to outperform PreNorm by 12.3 BLEU points. |
| Outcome: | The proposed model outperforms the current model by 12.3 BLEU points on 54 directions on OPUS, IWSLT, and Europarl datasets. |
BERTSeg: BERT Based Unsupervised Subword Segmentation for Neural Machine Translation (2022.aacl-short)
Copied to clipboard
| Challenge: | Existing subword segmenters are frequency-based without semantics information or neural-based but trained on parallel corpora. |
| Approach: | They propose an unsupervised neural subword segmenter for neural machine translation that utilizes contextualized semantic embeddings of words from characterBERT and maximizes the generation probability of subword segments. |
| Outcome: | The proposed method improves translation performance on ALT, IWSLT15 Vi->En, WMT16 Ro->En and WMT15 Fi->En datasets. |
Flexible Weight Tuning and Weight Fusion Strategies for Continual Named Entity Recognition (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for Named Entity Recognition (CNER) use knowledge distillation to retain old knowledge, but they are too expensive and fail to integrate with existing state-of-the-art models. |
| Approach: | They propose a weight tuning and weightfusion strategy to learn new entity types while mitigating catastrophic forgetting of old models. |
| Outcome: | The proposed strategies improve the performance of existing models and are model-agnostic. |
Exploiting Multilingualism through Multistage Fine-Tuning for Low-Resource Neural Machine Translation (D19-1)
Copied to clipboard
| Challenge: | Using multi-parallel corpora for transfer learning is a useful technique for low-resource NMT. |
| Approach: | They compare multi-parallel corpora for transfer learning in a low-resource setting . their results show that multi-paralleled corpors are extremely useful . |
| Outcome: | The proposed model can give 3–9 BLEU score gains over a one-to-one model. |
How Does Cognitive Bias Affect Large Language Models? A Case Study on the Anchoring Effect in Price Negotiation Simulations (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Cognitive biases can be observed in LLMs, affecting their reliability in real-world applications. |
| Approach: | They investigate the anchoring effect in LLM-driven price negotiations . reasoning models are less prone to the anchor effect, they find . |
| Outcome: | The proposed study shows that LLMs are influenced by the anchoring effect like humans . reasoning models are less prone to the anchor effect, but personality traits are not affected . |