Papers with fine-grained control
Synthetic Data for Evaluation: Supporting LLM-as-a-Judge Workflows with EvalAssist (2025.emnlp-demos)
Copied to clipboard
Martín Santillán Cooper, Zahra Ashktorab, Hyo Jin Do, Erik Miehling, Werner Geyer, Jasmina Gajcin, Elizabeth M. Daly, Qian Pan, Michael Desmond
| Challenge: | EvalAssist is a web-based application designed to assist human-centered evaluation of language model outputs. |
| Approach: | They propose a synthetic data generation tool integrated into EvalAssist to assist human-centered evaluation of language model outputs. |
| Outcome: | The proposed tool supports flexible prompting, RAG-based grounding, persona diversity, and iterative generation workflows. |
Counterfactuals to Control Latent Disentangled Text Representations for Style Transfer (2021.acl-short)
Copied to clipboard
| Challenge: | Existing methods for unsupervised text style transfer focus on transferring a specific attribute, but this technique has never been explored in natural language generation tasks. |
| Approach: | They propose a counterfactual-based method to modify latent representations by posing a ‘what-if’ scenario. |
| Outcome: | The proposed method is tested on multiple attribute transfer tasks like Sentiment, Formality and Excitement to support the hypothesis. |
♫ MuSiQue: Multihop Questions via Single-hop Question Composition (2022.tacl-1)
Copied to clipboard
| Challenge: | Existing multihop reasoning benchmarks are largely solvable via shortcuts . a bottom–up approach allows us to create a multihop QA dataset that requires proper multihop thinking. |
| Approach: | They propose a bottom–up approach that selects composable pairs of single-hop questions that are connected and adds stringent filters to the construction process. |
| Outcome: | The proposed approach creates a multihop question answering dataset with 25K 2–4 hop questions. |
FORG3D: Flexible Object Rendering for Generating Vision-Language Spatial Reasoning Data from 3D Scenes (2025.acl-demo)
Copied to clipboard
| Challenge: | FORG3D synthesizes vision-language data for use in cognitive experiments . it provides precise control over object placement, orientation, and camera configurations . |
| Approach: | They propose a 3D rendering toolkit that synthesizes vision-language data with Blender and Python. |
| Outcome: | The toolkit synthesizes vision-language data for two primary purposes: supporting cognitive experiments and improving visual reasoning capabilities of large vision-linguistic models. |
Improving LLM Reasoning through Interpretable Role-Playing Steering (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for role-playing rely on prompt engineering, which lacks stability and interpretability. |
| Approach: | They propose a framework that extracts latent representations from role-play prompts and constructs a steering vector that can be injected into the model's residual stream with controllable intensity. |
| Outcome: | The proposed framework extracts latent representations from role-play prompts, selects the most relevant features based on activation patterns, and constructs a steering vector that can be injected into the model’s residual stream with controllable intensity. |
A Close Look into the Calibration of Pre-trained Language Models (2023.acl-long)
Copied to clipboard
| Challenge: | Pre-trained language models (PLMs) may fail in giving reliable estimates of their predictive uncertainty. |
| Approach: | They conduct fine-grained control experiments to study the dynamic change in PLMs’ calibration performance in training. |
| Outcome: | The proposed methods significantly reduce PLMs’ confidence in wrong predictions. |
CARES: Context-Aware Resolution Selector for VLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Large vision–language models process images at native or high resolution to remain effective across tasks. |
| Approach: | They propose a lightweight preprocessing module that predicts the minimum sufficient input resolution for large vision–language models. |
| Outcome: | CARES predicts when a pre-trained VLM's response converges to its peak ability to answer correctly, reducing compute by up to 80%. |
The Case for a Single Model that can Both Generate Continuations and Fill-in-the-Blank (2022.findings-naacl)
Copied to clipboard
| Challenge: | a natural language generation system can be used to create text at the end of a passage . fill in the blank (FITB) is a task of inserting text into a specified position in a text . |
| Approach: | They evaluate the feasibility of using a single model to perform both tasks . they show that models pre-trained with a FitB-style objective are capable of both tasks. |
| Outcome: | The proposed model can perform both fill in the blank and continuation tasks. |
Can We Steer Reasoning Direction by Thinking Intervention? (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large Reason Models suffer from overthinking and erroneous reasoning problems due to the lack of fine-grained control over their reasoning behaviors. |
| Approach: | They propose a paradigm to enable fine-grained control over LRMs’ reasoning behaviors by aligning reasoning trajectories with specific cognitive patterns. |
| Outcome: | The proposed paradigm achieves integration intervention throughout model reasoning processes. |
Posterior Control of Blackbox Generation (2020.acl-main)
Copied to clipboard
| Challenge: | Existing methods for conditional natural language generation are limited in their ability to produce controlled output. |
| Approach: | They propose to augment neural generation models with discrete control states learned through a structured latent-variable approach. |
| Outcome: | The proposed approach improves over benchmarks while providing fine-grained control. |
Representation Interventions Enable Lifelong Knowledge Memory Control in LLMs (2026.acl-long)
Copied to clipboard
Xuyuan Liu, Shengyu Chen, Xinshuai Dong, Yanchi Liu, Xujiang Zhao, Haoyu Wang, Yujun Yan, Haifeng Chen, Zhengzhang Chen
| Challenge: | Large language models (LLMs) produce outdated or inaccurate content. Updating their knowledge efficiently and accurately without costly retraining is a major challenge. |
| Approach: | They propose a robust and scalable method that treats knowledge control as interventions within the model’s representation space. |
| Outcome: | The proposed method achieves fine-grained control over complex, unstructured knowledge while maintaining general utility with frozen base weights. |
From Token to Action: State Machine Reasoning to Mitigate Overthinking in Information Retrieval (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Chain-of-Thought (CoT) prompting often leads to overthinking in large language models . redundant trajectories that revisit similar states and misguided reasoning that diverges from user intent are two key challenges in information retrieval. |
| Approach: | They propose a transition-based reasoning framework that supports early stopping and fine-grained control. |
| Outcome: | The proposed framework improves retrieval performance by 3.4% while reducing token usage by 74.4%. |
DocReRank: Single-Page Hard Negative Query Generation for Training Multi-Modal RAG Rerankers (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing models focus on identifying relevant documents, but embedding similarity often limits accuracy. |
| Approach: | They propose a method to generate hard negative queries per page instead of negative pages per query . they propose to refine ranking of an initial set of retrieved documents using hard negative mining . |
| Outcome: | The proposed approach outperforms existing models and significantly improves retrieval performance. |
Fine-Grained Controllable Text Generation Using Non-Residual Prompting (2022.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to control the text generation process are not expressive enough. |
| Approach: | They propose an encoder-decoder architecture that enables intermediate text prompts at arbitrary time steps. |
| Outcome: | The proposed architecture is expressive and versatile on multiple experimental settings. |
Using Structured Content Plans for Fine-grained Syntactic Control in Pretrained Language Model Generation (2022.coling-1)
Copied to clipboard
| Challenge: | Large pretrained language models can generate powerful text but cannot be controlled at a sub-sentential level. |
| Approach: | They propose to make such fine-grained control possible in pretrained LMs by generating text directly from a semantic representation, Abstract Meaning Representation (BART), which is augmented at the node level with syntactic control tags. |
| Outcome: | The proposed method can generate text from a semantic representation, which is augmented at the node level with syntactic control tags. |
Controlling Machine Translation for Multiple Attributes with Additive Interventions (2021.emnlp-main)
Copied to clipboard
| Challenge: | A standard approach for exerting control in MT is to prepend the input with a special tag to signal the desired output attribute. |
| Approach: | They propose a vector-valued approach which allows for fine-grained control over multiple attributes simultaneously via a weighted linear combination of the corresponding vectors. |
| Outcome: | The proposed approach achieves better control over a wider range of tasks than tagging and even fine-tuning a model trained without annotations. |
CHAE: Fine-Grained Controllable Story Generation with Characters, Actions and Emotions (2022.coling-1)
Copied to clipboard
| Challenge: | Existing studies on story generation focus on coarse-grained control of the story, neglecting the details of the narrative. |
| Approach: | They propose a model for fine-grained control on the story that allows the generation of customized stories with characters, corresponding actions and emotions arbitrarily assigned. |
| Outcome: | The proposed method has strong controllability to generate customized stories according to the fine-grained personalized guidance. |
Unifying Continuous and Discrete Text Diffusion with Non-simultaneous Diffusion Processes (2025.acl-long)
Copied to clipboard
| Challenge: | Experimental results demonstrate NeoDiff’s superior performance compared to baselines of non-autoregressive continuous and discrete diffusion models, iterative-based methods and autoregressive diffusion-based approaches. |
| Approach: | They propose a discrete and continuous diffusion model that integrates the strengths of discrete, continuous and continuous approaches. |
| Outcome: | The proposed model unifies the theories of discrete and continuous diffusion models, offering a more principled and effective framework for text generation. |
SGDPO: Self-Guided Direct Preference Optimization for Language Model Alignment (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for aligning Large Language Models with human values are limited and results of DPO are not resilient. |
| Approach: | They propose a self-guided direct preference optimization algorithm that incorporates a pilot term to steer the gradient flow during the optimization process. |
| Outcome: | The proposed method can generate human-preferred response up to 9.19% higher than previous methods. |
FineSteer: A Unified Framework for Fine-Grained Inference-Time Steering in Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for inference-time steering fail to be effective, utility-preserving and training-efficient due to rigid, one-size-fits-all designs and limited adaptability. |
| Approach: | They propose a steering framework that decomposes inference-time steering into two stages . they propose 'conditional steering' mechanism that preserves model utility by avoiding unnecessary steering . a 'mixture-of-Steering-Experts' mechanism captures multimodal nature of desired steering behaviors . |
| Outcome: | The proposed framework outperforms the state-of-the-art methods on safety and truthfulness benchmarks. |
Universal Acoustic Adversarial Attacks for Flexible Control of Speech-LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | acoustic adversarial attacks on speech LLMs may make them more vulnerable to adversarials . flexible speech encoders and large language models have enabled speech Llms to handle a wide range of processing tasks. |
| Approach: | They investigate universal adversarial attacks on speech LLMs by pre-trained speech encoders and large language models. |
| Outcome: | The proposed model can handle a wide range of spoken language processing tasks. |
Nested Browser-Use Learning for Agentic Information Seeking (2026.acl-long)
Copied to clipboard
Baixuan Li, Jialong Wu, Wenbiao Yin, Kuan Li, Zhongwang Zhang, Huifeng Yin, Zhengwei Tao, Liwen Zhang, Pengjun Xie, Jingren Zhou, Yong Jiang, Wentao Zhang, Zhiqiang Gao
| Challenge: | Existing information-seeking (IS) agents rely on the web for their information acquisition. |
| Approach: | They propose a browser-action framework that decouples interaction control from page exploration through a nested structure. |
| Outcome: | Empirical results show that NestBrowse offers clear benefits in practice. |
Linguistically-Controlled Paraphrase Generation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Controlled paraphrase generation produces paraphrases that preserve meaning while allowing precise control over linguistic attributes of output. |
| Approach: | They introduce an encoder-decoder framework that enables fine-grained control over 40 linguistic attributes in English. |
| Outcome: | The proposed framework reduces attribute error by up to 34% over existing models . |
Incomplete Prompt Jailbreaks in Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are increasingly released as open-weight models with safeguards against harmful requests. |
| Approach: | They formalize incomplete prompt jailbreaks as incomplete prompts elicit harmful continuations . they identify two functional neurons that delay refusal until sentence termination . |
| Outcome: | The proposed model fails to generalize across content domains and attractor types . the proposed model can be used to perform more precise and robust IPJ defenses . |
MicroEdit: Neuron-level Knowledge Disentanglement and Localization in Lifelong Model Editing (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for enhancing large language models are designed for single or limited edits, lacking the capacity to support long-term, multi-round knowledge updates. |
| Approach: | They propose a neuron-level editing method that performs minimal interventions within large language models (LLMs) by leveraging a sparse autoencoder, MicroEdit disentangles knowledge representations and activates only a minimal set of necessary neurons for precise parameter updates. |
| Outcome: | Extensive experiments show that MicroEdit outperforms prior methods and robustly handles lifelong knowledge editing across QA and Hallucination settings on LLaM and Mistral. |
A Simple Yet Effective Method for Non-Refusing Context Relevant Fine-grained Safety Steering in LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning large language models to meet safety policies are costly and impractical. |
| Approach: | They propose a method to fine-tune large language models to meet evolving safety policies by applying a gradient-free, unsupervised approach. |
| Outcome: | The proposed method provides precise control, avoids blanket refusals, and directs models to generate safe, relevant content. |
Where Paths Split: Localized, Calibrated Control of Moral Reasoning in Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models display heterogeneous moral preferences across settings. |
| Approach: | They propose a method for steering toward a desired ethical framework while preserving general competence. |
| Outcome: | The proposed method outperforms baselines while providing interpretable mechanism. |