Papers with generation step
Speeding Up Entmax (2022.findings-naacl)
Copied to clipboard
| Challenge: | Recent studies suggest that sparsity is a problem when the trained model is used for inference. |
| Approach: | They propose an alternative to softmax that produces a dense probability distribution but is slower than softmax. |
| Outcome: | The proposed method keeps its virtuous characteristics but is slower than softmax and achieves on par or better performance in machine translation task. |
Breaking the Ceiling of the LLM Community by Treating Token Generation as a Classification for Ensembling (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have demonstrated remarkable capabilities in a wide range of tasks and contexts. |
| Approach: | They propose to use a token-level ensembling method to exploit the probability information at each generation step and to avoid early incorrect tokens. |
| Outcome: | The proposed method breaks the existing community performance ceiling and improves on several benchmarks. |
Large Language Models Do Multi-Label Classification Differently (2025.emnlp-main)
Copied to clipboard
| Challenge: | Multi-label classification is prevalent in real-world settings, but the behavior of Large Language Models (LLMs) in this setting is understudied. |
| Approach: | They propose to use initial probability distributions to analyze output distributions of LLMs at each label generation step to find out how LLM models perform multi-label classification. |
| Outcome: | The proposed methods improve alignment and predictive performance over existing methods. |
Mem2Seq: Effectively Incorporating Knowledge Bases into End-to-End Task-Oriented Dialog Systems (P18-1)
Copied to clipboard
| Challenge: | End-to-end task-oriented dialog systems often suffer from the challenge of incorporating knowledge bases. |
| Approach: | They propose a novel yet simple end-to-end differentiable model called memory-tosequence to address this issue. |
| Outcome: | The proposed model can be trained faster and achieve state-of-the-art performance on three different task-oriented dialog datasets. |
Dynamic Topic Tracker for KB-to-Text Generation (2020.coling-main)
Copied to clipboard
| Challenge: | Existing KB-to-text generation models suffer from an off-topic problem . existing models generate unrelated clauses regardless of input data . |
| Approach: | They propose a dynamic topic tracker that learns a global hidden representation for topics and recognizes the corresponding topic during each generation step. |
| Outcome: | The proposed model improves the performance of sentence generation and mitigates off-topic problem. |
Addressing Overthinking in Large Vision-Language Models via Gated Perception-Reasoning Optimization (2026.findings-acl)
Copied to clipboard
Xingjian Diao, Zheyuan Liu, Chunhui Zhang, Weiyi Wu, Keyi Kong, Lin Shi, Kaize Ding, Soroush Vosoughi, Jiang Gui
| Challenge: | Prior work has attempted to mitigate this issue by using adaptive reasoning strategies, but these methods overlook a fundamental bottleneck: visual perception failures. |
| Approach: | They propose a meta-reasoning controller that dynamically routes computation among three decision paths at each generation step. |
| Outcome: | The proposed method outperforms slow-thinking methods while producing shorter responses. |
Constructing Your Model’s Value Distinction: Towards LLM Alignment with Anchor Words Tuning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a study of large language models (LLMs) shows that they can generate outputs that are honest, positive, harmless, etc. |
| Approach: | They propose a method that amplifies logits difference between positive and negative tokens . they propose to use the logits gap to generate positive and positive tokens after alignment . |
| Outcome: | The proposed method achieves effective alignment, but requires fewer computational resources compared to training-time alignment methods. |
Distilling Calibrated Knowledge for Stance Detection (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for stance detection ignore meaningful signals among categories offered by hard labels. |
| Approach: | They propose to use knowledge distillation to calibrate teacher predictions in each generation step. |
| Outcome: | The proposed method can calibrate teacher predictions in each generation step and improves stance detection accuracy. |
Bridging the Gap between Different Vocabularies for LLM Ensemble (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing methods to combine large language models with complete outputs have limited effectiveness . lexical gaps between different LLMs hinder dynamic correction and enhancement . |
| Approach: | They propose a method to ensemble large language models via Vocabulary Alignment (EVA) they learn mappings between vocabularies of different LLMs with overlapping tokens . |
| Outcome: | The proposed method bridges the lexical gap among various LLMs, enabling meticulous ensemble at each generation step. |
Think More, Hallucinate Less: Mitigating Hallucinations via Dual Process of Fast and Slow Thinking (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have focused on the issue of hallucination in large language models. |
| Approach: | They propose a framework that allows an explicit slow thinking generation process for mitigating hallucinations during inference. |
| Outcome: | The proposed framework outperforms baseline approaches on English and Chinese datasets. |
SuperWriter: Reflection-Driven Long-Form Generation with Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Long-form text generation remains a challenge for large language models . generating extended sequences often leads to degraded coherence and logical consistency . |
| Approach: | They propose a framework that integrates explicit structured thinking into long-form text generation. |
| Outcome: | The proposed framework surpasses even larger-scale models in evaluation and human evaluation. |
Laziness Is a Virtue When It Comes to Compositionality in Neural Semantic Parsing (2023.acl-long)
Copied to clipboard
Maxwell Crouse, Pavan Kapanipathi, Subhajit Chaudhury, Tahira Naseem, Ramon Fernandez Astudillo, Achille Fokoue, Tim Klinger
| Challenge: | Compositional generalization is a key feature of human intelligence and has been identified as a major point of weakness in neural methods for semantic parsing. |
| Approach: | They propose a neural parsing generation method that constructs logical forms from the bottom up, beginning from the logical form’s leaves. |
| Outcome: | The proposed method outperforms general-purpose parsers on a CFQ dataset and two other Text-to-SQL datasets while also being competitive with parser that have been tailored to each task. |
Hit the Sweet Spot! Span-Level Ensemble for Large Language Models (2025.coling-main)
Copied to clipboard
| Challenge: | a recent study focused on sample-level and token-level ensembles, which hinder dynamic correction and enhancement of outputs during the generation process. |
| Approach: | They propose a span-level ensemble method that balances real-time adjustments and accurate ensemble decisions. |
| Outcome: | The proposed method improves performance across language generation tasks significantly. |
Min-k Sampling: Decoupling Truncation from Temperature Scaling via Relative Logit Dynamics (2026.acl-long)
Copied to clipboard
Yuanhao Ding, Meimingwei Li, Esteban Garces Arias, Matthias Aßenmacher, Christian Heumann, Chongsheng Zhang
| Challenge: | Existing methods for decoding large language models have extreme sensitivity to temperature parameter T. |
| Approach: | They propose a dynamic truncation strategy that analyzes the local shape of the logit distribution to identify "semantic cliffs" they show that Min-k consistently improves text quality even under extreme temperature settings . |
| Outcome: | The proposed method achieves strict temperature invariance and low sensitivity to hyperparameter choices. |
Downstream Trade-offs of a Family of Text Watermarks (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) can generate humanlike responses to a variety of requests like writing emails, translating or summarizing content. |
| Approach: | They evaluate the performance of large language models (LLMs) watermarked using three different strategies over a diverse suite of tasks including those cast as k-class classification (CLS), multiple choice question answering (MCQ), short-form generation (e.g., open-ended question answering) and long-form generator (eg. translation) |
| Outcome: | The proposed models can cause significant drops in their effectiveness across a variety of tasks including CLS, MCQ, short-form generation and translation tasks. |
From Complex to Simple: Unraveling the Cognitive Tree for Reasoning with Small Language Models (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to solve complex logical reasoning problems are cumbersome for language models. |
| Approach: | They propose to use iterative methodology to construct a cognitive tree using language models . they propose to generate multiple responses by utilizing in-context examples . |
| Outcome: | The proposed model achieves a performance level comparable to that of GPT-3.5 . the proposed model contains fewer parameters than 5% of the model with 175B parameters . |
Adaptive Contrastive Search: Uncertainty-Guided Decoding for Open-Ended Text Generation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to decode text to the most probable sequence have been proposed to address these challenges by improving coherence, diversity, and resemblance to human-generated text. |
| Approach: | They propose a novel decoding strategy that extends contrastive search by incorporating an adaptive degeneration penalty informed by the model’s estimated uncertainty at each generation step. |
| Outcome: | The proposed approach improves creativity and coherence while maintaining coherency across model architectures, languages, and datasets. |
Explicit Bayesian Inference to Uncover the Latent Themes of Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have impressive generative capabilities, yet their inner mechanisms remain largely opaque. |
| Approach: | They propose a variational autoencoder-based neural topic model to interpret LLMs generation process through an explicit Bayesian framework by inferring latent topic variables via variational inference. |
| Outcome: | The proposed model outperforms state-of-the-art topic models on intrinsic measures of coherence and diversity on multiple datasets and shows significant gains on classification and summarization tasks. |
Token-level Inference-Time Alignment for Vision-Language Models (2026.findings-acl)
Copied to clipboard
Kejia Chen, Junjun Zheng, Jiawen Zhang, Manxi Lin, Xiao Pan, Jiacong Hu, Jian Lou, Zunlei Feng, Mingli Song
| Challenge: | Vision-Language Models (VLMs) often prioritize linguistic fluency over visual fidelity . despite widespread adoption, VLMs often exhibit a critical failure mode: hallucination . |
| Approach: | They propose a framework for Token-level Inference-Time Alignment that steers the decoding process without updating the base model parameters. |
| Outcome: | The proposed framework improves performance on 13 benchmarks across architectures . it boosts LLaVA-1.5-7B by 8.6% on MMVet and achieves a 74.0 MMStar score . |
ECO Decoding: Entropy-Based Control for Controllability and Fluency in Controllable Dialogue Generation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Controllable Dialogue Generation (CDG) enables chatbots to generate responses tailored to desired attributes like emotion and dialog-act. |
| Approach: | They propose a method which dynamically adjusts the control strength at each generation step according to the model’s entropy in both the language model and attribute classifier probability distributions. |
| Outcome: | The proposed method outperforms existing methods on DailyDialog and MultiWOZ datasets while maintaining fluency and grammar. |
PHOTON: Hierarchical Autoregressive Modeling for Lightspeed and Memory-Efficient Language Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Experimental results show that PHOTON is superior to competitive Transformer-based language models . despite their capabilities, the inference cost increases with context length under recent serving workloads. |
| Approach: | They propose a hierarchical autoregressive model that replaces horizontal scanning with vertical context scanning. |
| Outcome: | Experimental results show that PHOTON is superior to Transformer-based language models . it reduces decode-time KV-cache traffic yielding up to 103 higher throughput per unit memory . |
Unconditional Truthfulness: Learning Unconditional Uncertainty of Large Language Models (2025.emnlp-main)
Copied to clipboard
Artem Vazhentsev, Ekaterina Fadeeva, Rui Xing, Gleb Kuzmin, Ivan Lazichny, Alexander Panchenko, Preslav Nakov, Timothy Baldwin, Maxim Panov, Artem Shelmanov
| Challenge: | Uncertainty quantification (UQ) is a promising approach for detecting hallucinations and low-quality outputs of Large Language Models (LLMs). |
| Approach: | They propose to learn conditional dependency between autoregressive LLM generation steps from attention-based features and a two-staged training procedure to incorporate recurrent features. |
| Outcome: | The proposed method is highly effective for selective generation, achieving substantial improvements over rivaling unsupervised and supervised approaches. |
AdaptEvolve: Improving Efficiency of Evolutionary AI Agents through Adaptive Model Selection (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing routing strategies rely on static heuristics or external controllers to optimize performance. |
| Approach: | They propose a framework that leverages intrinsic generation confidence to estimate solvability. |
| Outcome: | Empirical results show that confidence-driven selection yields favorable Pareto frontier . computational cost of state-of-the-art large language models remains a key barrier to scalable deployment . |