Papers by Gary Lee
Can LLMs Estimate Cognitive Complexity of Reading Comprehension Items? (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for estimating the cognitive complexity of reading comprehension items are expensive, time-consuming, and subject to rater variability. |
| Approach: | They propose to use two dimensions to estimate cognitive complexity of RC items to focus on evidence Scope and transformation level to estimate the cognitive complexity. |
| Outcome: | The proposed models can estimate the cognitive complexity of items by focusing on two dimensions—Evidence Scope and Transformation Level—that indicate the degree of cognitive burden involved in reasoning about the answer. |
Mixture-of-Experts with Intermediate CTC Supervision for Accented Speech Recognition (2026.acl-long)
Copied to clipboard
| Challenge: | Accented speech remains a persistent challenge for automatic speech recognition (ASR) Accent-agnostic approaches improve robustness but struggle with heavily accented or unseen varieties . |
| Approach: | They propose a Mixture-of-Experts architecture with intermediate CTC supervision that promotes expert specialization and generalization. |
| Outcome: | Experiments show that the proposed architecture improves on accented speech . the proposed framework is based on a mixture-of-experts architecture with intermediate supervision . |
Autoregressive Score Generation for Multi-trait Essay Scoring (2024.findings-eacl)
Copied to clipboard
| Challenge: | Existing holistic approaches to score essays using pre-trained BERT-based models are inefficient, leading to inferior qualities in data-scarce traits. |
| Approach: | They propose an autoregressive prediction of multi-trait scores using pre-trained T5 models. |
| Outcome: | The proposed model shows over 5% improvement in prompts and traits compared to previous models . |
Revisiting Early Detection of Sexual Predators via Turn-level Optimization (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing methods to detect online grooming rely on chat-level risk labels and fail to identify optimal intervention points. |
| Approach: | They propose a speed control reinforcement learning strategy based on luring communication theory to capture the predator’s turn-level entrapment and a new reward function that balances the trade-off between speed and accuracy based upon the LCT. |
| Outcome: | The proposed method preempts online grooming while identifying optimal early intervention points. |
Leveraging What’s Overfixed: Post-Correction via LLM Grammatical Error Overcorrection (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to reduce overcorrection often result in significantly decreased recall, limiting the usability of correction systems. |
| Approach: | They propose a novel approach that leverages the strengths of large language models to balance recall and precision by triggering overcorrection via LLMs and fine-tuning smaller models to identify and refine erroneous outputs. |
| Outcome: | The proposed approach maximizes recall and precision by leveraging the generative power of LLMs while preserving the reliability of smaller supervised models. |
Why Do Multilingual Reasoning Gaps Emerge in Reasoning Language Models? (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent reasoning language models (RLMs) achieve strong performance on complex reasoning tasks, yet they still exhibit a multilingual reasoning gap. |
| Approach: | They propose a strategy that incorporates an English translation into the initial reasoning trace when an understanding failure is detected. |
| Outcome: | The proposed strategy incorporates an English translation into the initial reasoning trace when an understanding failure is detected. |
Merging Triggers, Breaking Backdoors: Defensive Poisoning for Instruction-Tuned Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are vulnerable to backdoor attacks, where adversaries poison a small subset of data to implant hidden behaviors. |
| Approach: | They propose a training pipeline that immunizes instruction-tuned LLMs against backdoor attacks. |
| Outcome: | The proposed defenses lower attack success rates while preserving instruction-following ability. |
PanicToCalm: A Proactive Counseling Agent for Panic Attacks (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing models for training such models are limited due to ethical and logistical issues. |
| Approach: | They propose a dataset that includes high-distress episodes constructed from first-person narratives and structured around the principles of Psychological First Aid. |
| Outcome: | The proposed model outperforms baseline models in counselor-side metrics and client affect improvement. |
Retrieval-Augmented Fine-Tuning With Preference Optimization For Visual Program Generation (2025.acl-long)
Copied to clipboard
| Challenge: | Visual programming languages (VPLs) allow users to create programs through graphical interfaces, which results in easier accessibility and widespread usage in various domains. |
| Approach: | They propose to train VPLs from user instructions using large language models . they propose to use retrieval-augmented fine-tuning to leverage repetitive use of subroutines . |
| Outcome: | The proposed method outperforms prompting-based methods for LD generation accuracy even with smaller backbone models. |
Progressive Facial Granularity Aggregation with Bilateral Attribute-based Enhancement for Face-to-Speech Synthesis (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for generating speech from facial images rely on pre-trained visual encoders and fine-tune them to align with speech embeddings. |
| Approach: | They propose to derive corresponding voices from facial images using face-to-voice synthesis, which derives corresponding voice from facial image. |
| Outcome: | The proposed approach significantly improves face-voice congruence and synthesis stability. |
Self-Correcting Code Generation Using Small Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a recent study has demonstrated that self-correction is a powerful tool for code generation, but whether it is effective for smaller models remains unexplored. |
| Approach: | They propose a method that trains small language models to maintain correct outputs while progressively correcting incorrect outputs as turns proceed. |
| Outcome: | The proposed approach improves the ability of small language models for multi-turn code correction. |
Safeguarding RAG Pipelines with GMTP: A Gradient-based Masked Token Probability Method for Poisoned Document Detection (2025.findings-acl)
Copied to clipboard
| Challenge: | Retrieval-Augmented Generation (RAG) provides external knowledge for accurate and up-to-date responses, but external knowledge is vulnerable to poisoning and unauthorized injections. |
| Approach: | They propose a Gradient-based Masked Token Probability defense method to detect and filter out adversarially crafted documents by examining gradients of the retriever’s similarity function. |
| Outcome: | Experiments show that the proposed method eliminates over 90% of poisoned content while retaining relevant documents. |
Adaptive Planning for Multi-Attribute Controllable Summarization with Monte Carlo Tree Search (2026.acl-long)
Copied to clipboard
| Challenge: | Controllable summarization is a form of outputs that tailors summaries to user-specified attributes. |
| Approach: | They propose an adaptive planning framework that reframes the task as planning the order of sequential attribute control with a customized Monte Carlo Tree Search. |
| Outcome: | The proposed framework surpasses LLM-based self-planning models and fine-tuned baselines in multi-attribute controllable summarization. |
Cross-lingual Back-Parsing: Utterance Synthesis from Meaning Representation for Zero-Resource Semantic Parsing (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to extend semantic parsing (SP) beyond English are challenging due to the complex slot alignment step after translation. |
| Approach: | They propose a method to enhance cross-lingual transfer for SP by utilizing mPLMs. |
| Outcome: | The proposed method synthesizes target language utterances from source meaning representations while maintaining high slot value alignment rates. |
KoBLEX: Open Legal Question Answering with Multi-hop Reasoning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have demonstrated remarkable performances in general domains and are now extending into the expert domain of law. |
| Approach: | They propose a Korean Benchmark for Legal EXplainable QA (KoBLEX) that evaluates provision-grounded, multi-hop legal reasoning. |
| Outcome: | The proposed method outperforms baselines and shows a high correlation with human judgments. |
K-COMP: Retrieval-Augmented Medical Domain Question Answering With Knowledge-Injected Compressor (2025.naacl-long)
Copied to clipboard
| Challenge: | Documents retrieved for closed domains require high expertise, so reader model may have difficulty comprehending the text. |
| Approach: | They propose a system which augments the prior knowledge required to answer correctly by adding thousands of tokens to the retrieved documents. |
| Outcome: | The proposed system provides the knowledge required to answer correctly and generates prior knowledge to facilitate the answer process prior to compression of the retrieved passages. |
Multi-Facet Blending for Faceted Query-by-Example Retrieval (2025.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to faceted query-by-example (QBE) rely on document-level comparisons using basic indicators like citations . however, this limited their use to citation-based domains and fails to capture the intricacies of facet constraints. |
| Approach: | They propose a multi-facet blending augmentation method that exploits modularity by decomposing and recomposing to synthesize facet-specific training sets. |
| Outcome: | The proposed method decomposes documents into facet units and generates (ir)relevant pairs, thereby synthesizing facet-specific training sets. |
Multimodal Cognitive Reframing Therapy via Multi-hop Psychotherapeutic Reasoning (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing studies have focused on text-based cognitive reframing, but neglected the importance of non-verbal evidence in real-life therapy. |
| Approach: | They propose a dataset that pairs each GPT-4-generated dialogue with an image that reflects the virtual client’s facial expressions to better mirror real psychotherapy, where facial expression leads to interpreting implicit emotional evidence. |
| Outcome: | The proposed approach outperforms existing methods with LLMs and vision-language models and provides more thoughtful and empathetic suggestions. |
Autoregressive Multi-trait Essay Scoring via Reinforcement Learning with Scoring-aware Multiple Rewards (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing reinforcement learning (RL) applications in AES are limited to classification models despite associated performance degradation. |
| Approach: | They propose to integrate actual evaluation schemes into the training process by designing QWK-based rewards with a mean-squared error penalty for multi-trait AES. |
| Outcome: | The proposed scoring-aware multi-reward reinforcement learning integrates actual evaluation schemes into the training process. |
DyPCL: Dynamic Phoneme-level Contrastive Learning for Dysarthric Speech Recognition (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing studies have focused on data augmentation and feature extraction methods to improve dysarthric speech recognition. |
| Approach: | They propose a Dynamic Phoneme-level Contrastive Learning method which decomposes the speech utterance into phoneme segments for phoneme- level contrastive learning. |
| Outcome: | The proposed method outperforms baseline models and achieves an average 22.10% reduction in word error rate (WER) across the overall dysarthria group. |
Multi-Dimensional Optimization for Text Summarization via Reinforcement Learning (2024.acl-long)
Copied to clipboard
| Challenge: | Existing summarization methods target a specific dimension, resulting in poor quality summaries. |
| Approach: | They propose multi-objective reinforcement learning tailored to generate balanced summaries across all dimensions. |
| Outcome: | The proposed model achieves significant performance gains compared to baseline models on representative summarization datasets on four dimensions. |
Multi-Level Attention Aggregation for Language-Agnostic Speaker Replication (2024.eacl-short)
Copied to clipboard
| Challenge: | Recent advances in speech synthesis research have enabled the generation of natural-sounding speech, which has prompted a notable shift in TTS research towards the synthesis of speech in the voices of both seen and unseen speakers. |
| Approach: | They propose a multi-level attention aggregation approach that probes and amplifies various speaker-specific attributes in a hierarchical manner. |
| Outcome: | The proposed model achieves substantial speaker similarity and generalizes to out-of-domain (OOD) cases. |
Adversarial DPO: Harnessing Harmful Data for Reducing Toxicity with Minimal Impact on Coherence and Evasiveness in Dialogue Agents (2024.findings-naacl)
Copied to clipboard
| Challenge: | Existing toxicity within large language models can negatively impact the user experience, causing performance degradation. |
| Approach: | They propose an adversarial DPO algorithm that improves direct preference optimization (DPO) by incorporating harmful data into the generative model. |
| Outcome: | The proposed training algorithm improves the model’s resilience against harmful conversations while minimizing performance degradation. |
MiLQ: Benchmarking IR Models for Bilingual Web Search with Mixed Language Queries (2025.emnlp-main)
Copied to clipboard
| Challenge: | Information Retrieval (IR) research on mixed-language queries remains sparse and outdated. |
| Approach: | They propose a test set for mixed-language queries that is realistic and preferred by bilingual speakers. |
| Outcome: | The proposed benchmarks show that multilingual IR models perform moderately on MiLQ and inconsistently across native, English, and mixed-language queries. |
PicPersona-TOD : A Dataset for Personalizing Utterance Style in Task-Oriented Dialogue with Image Persona (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing systems produce generic, monotonic responses that lack individuality and fail to adapt to users’ personal attributes. |
| Approach: | They propose a dataset that incorporates user images as part of the persona, enabling personalized responses tailored to user-specific factors such as age or emotional context. |
| Outcome: | The proposed dataset enhances user experience, with personalized responses contributing to a more engaging interaction. |
Exploring Iterative Controllable Summarization with Large Language Models (2026.findings-eacl)
Copied to clipboard
| Challenge: | Large language models (LLMs) excel at abstractive summarization tasks, but their ability to precisely control summary attributes remains underexplored. |
| Approach: | They propose a guide-to-explain framework for controllable summarization that enables the model to identify misaligned attributes in the initial draft and guides it to self-explan errors in the previous output. |
| Outcome: | The proposed framework generates well-adjusted summaries that satisfy the desired attributes with robust effectiveness while requiring surprisingly fewer iterations than other iterative approaches. |
Audio-Based Linguistic Feature Extraction for Enhancing Multi-lingual and Low-Resource Text-to-Speech (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to synthesize speech for low-resource languages require a substantial amount of source language corpora to generate the linguistic knowledge that can be reused for speech synthesis. |
| Approach: | They propose a method that extracts linguistic features from audio input while effectively filtering out miscellaneous acoustic information including speaker-specific attributes like timbre. |
| Outcome: | The proposed method extracts linguistic features from audio input while effectively filtering out miscellaneous acoustic information including speaker-specific attributes like timbre. |
Cross-lingual Transfer for Automatic Question Generation by Learning Interrogative Structures in Target Languages (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing automatic question generation datasets focus on English, resulting in data gaps for other languages. |
| Approach: | They propose a cross-lingual transfer method that allows models to generate questions in low-resource languages. |
| Outcome: | The proposed method outperforms other models and achieves comparable performance across languages. |
EnSToM: Enhancing Dialogue Systems with Entropy-Scaled Steering Vectors for Topic Maintenance (2025.findings-acl)
Copied to clipboard
| Challenge: | Small large language models (sLLMs) are lightweight and efficient, but struggle to maintain topic consistency in task-oriented dialogue systems. |
| Approach: | They propose an approach to ensure topic consistency in task-oriented dialogue systems by manipulating internal activations during inference. |
| Outcome: | The proposed approach achieves significant performance gain with a relatively small data size compared to fine-tuning approaches. |
A Multi-Agent Framework for Feature-Constrained Difficulty Control in Reading Comprehension Item Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for difficulty-controlled reading comprehension item generation rely on a single agent prompting approach. |
| Approach: | They propose a multi-agent framework for Feature-constrained Item Generation where multiple LLM agents collaborate to generate and iteratively revise items based on intended constraints. |
| Outcome: | The proposed method generates items with monotonically increasing difficulty at higher rates than baselines. |
MIRROR: Multimodal Cognitive Reframing Therapy for Rolling with Resistance (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have explored the use of large language models (LLMs) in psychotherapy, however text-based cognitive behavioral therapy models struggle with client resistance, which weakens therapeutic alliance. |
| Approach: | They propose a multimodal approach that incorporates nonverbal cues and a synthetic dataset that pairs each client’s statements with corresponding facial images to train vision language models. |
| Outcome: | The proposed approach outperforms existing text-based cognitive behavioral therapy models in managing client resistance and fostering therapeutic alliance. |
Prompt-Guided Selective Masking Loss for Context-Aware Emotive Text-to-Speech (2025.findings-naacl)
Copied to clipboard
| Challenge: | Emotional dialogue speech synthesis (EDSS) aims to generate expressive speech by leveraging the dialogue context between interlocutors. |
| Approach: | They propose a large language model to generate holistic emotion tags based on prior dialogue context and pinpoint key words in the target utterance that align with the predicted emotional state. |
| Outcome: | The proposed method improves emotional expressiveness and facilitates automatic emotion speech generation during inference. |
Towards Prompt Generalization: Grammar-aware Cross-Prompt Automated Essay Scoring (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing approaches to score essays on unseen prompts are challenging to use in educational situations. |
| Approach: | They propose a grammar-aware cross-prompt trait scoring model which internally captures prompt-independent syntactic aspects to learn generic essay representation. |
| Outcome: | Empirical results show that the proposed model improves prompt-independent and grammar-related traits and achieves notable QWK gains in the most challenging cross-prompt scenario. |
DeRAGEC: Denoising Named Entity Candidates with Synthetic Rationale for ASR Error Correction (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent studies have demonstrated that postprocessing speech recognition transcriptions with large language models can significantly enhance the accuracy of Automatic Speech Recognition (ASR). |
| Approach: | They propose a method to improve Named Entity (NE) correction in Automatic Speech Recognition systems by leveraging phonetic similarity and augmented definitions. |
| Outcome: | The proposed method outperforms baseline methods on common voice and STOP datasets and achieves a 28% reduction in WER and NE hit ratio. |
Difficulty-Controllable Cloze Question Distractor Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for generating high-quality distractors lack adaptability and control over difficulty levels. |
| Approach: | They propose a two-way distractor generation process to generate plausible distractors using an ensemble QA system and a multitask learning strategy to train a difficulty-controllable generation model. |
| Outcome: | The proposed method significantly outperforms GPT-4o in aligning distractor difficulty with human perception. |
Behavior-Aware Item Modeling via Dynamic Procedural Solution Representations for Knowledge Tracing (2026.findings-acl)
Copied to clipboard
| Challenge: | Knowledge Tracing (KT) aims to predict learners’ future performance from past interactions, but they overlook the procedural dynamics of problem solving. |
| Approach: | They propose a framework that enriches item representations by integrating dynamic procedural solution information. |
| Outcome: | Experiments on XES3G5M and NIPS34 show that BAIM outperforms strong pretraining-based baselines, achieving particularly large gains under repeated learner interactions. |