Papers with autoregressive
Infusing Sequential Information into Conditional Masked Translation Model with Self-Review Mechanism (2020.coling-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive models generate target words in parallel, but with a large latency due to the left-to-right dependency. |
| Approach: | They propose to train a conditional masked translation model and refine results within several iterations to remedy a flawed translation by non-autoregressive models. |
| Outcome: | The proposed model outperforms state-of-the-art models by over 1 BLEU while using less training computations. |
Non-Autoregressive Sequence Generation (2022.acl-tutorials)
Copied to clipboard
| Challenge: | Non-autoregressive sequence generation (NAR) models generate output sequences in parallel to speed up generation process. |
| Approach: | This tutorial provides a thorough introduction and review of non-autoregressive sequence generation . it aims to generate the entire or partial output sequences in parallel to speed up the generation process . |
| Outcome: | This tutorial provides a thorough introduction and review of non-autoregressive sequence generation . it aims to reduce the performance gap between state-of-the-art models due to lack of modeling power . |
Binary and Ternary Natural Language Generation (2023.acl-long)
Copied to clipboard
| Challenge: | ternary and binary neural networks have proven difficult to optimize since both parameter and output space are discretized . authors demonstrate ternaries and binary models on downstream tasks of summarization and machine translation . |
| Approach: | They propose to use ternary and binary neural networks to optimize for multiplication-free computation . they propose to apply statistics-based quantization for the weights and elastic quantization of the activations to the transformer text generation model. |
| Outcome: | The proposed model outperforms the best existing models on machine translation tasks. |
Leveraging Key Information Modeling to Improve Less-Data Constrained News Headline Generation via Duality Fine-Tuning (2022.aacl-main)
Copied to clipboard
| Challenge: | Recent language generative models are mostly trained on large-scale datasets, while in some real scenarios, the training datasets are often expensive and would be small-scale. |
| Approach: | They propose a novel duality fine-tuning method to capture more information from limited data and build connections between tasks. |
| Outcome: | The proposed method can capture more information from limited data, build connections between separate tasks, and is suitable for less-data constrained generation tasks. |
GRITHopper: Decomposition-Free Multi-Hop Dense Retrieval (2026.eacl-long)
Copied to clipboard
| Challenge: | Decomposition-based multi-hop retrieval methods rely on autoregressive steps to break down complex queries, which breaks end-to-end differentiability and is computationally expensive. |
| Approach: | They propose a multi-hop dense retrieval model that integrates causal language modeling with dense retrievals. |
| Outcome: | The proposed model outperforms existing methods on in-distribution and out-of-difference benchmarks. |
DeepRapper: Neural Rap Generation with Rhyme and Rhythm Modeling (2021.acl-long)
Copied to clipboard
| Challenge: | Existing systems for rap generation focus on rhyming lyrics but ignore rhythmic beats . rap lyrics need to be semantically meaningful and fashionable to convey interesting stories . |
| Approach: | They develop a Transformer-based rap generation system that can model both rhymes and rhythms. |
| Outcome: | The proposed system generates high-quality raps with rhymes and rhythms . it is based on a Transformer-based language model . |
Recursive Non-Autoregressive Graph-to-Graph Transformer for Dependency Parsing with Iterative Refinement (2021.tacl-1)
Copied to clipboard
| Challenge: | RNGTr is a non-recursive Graph-to-Graph Transformer for iterative refinement of graphs . it can improve the accuracy of initial parsers on 13 languages . |
| Approach: | They propose a recursive non-autoregressive Graph-to-Graph Transformer architecture for iterative refinement of arbitrary graphs and apply it to syntactic dependency parsing. |
| Outcome: | The proposed architecture can improve state-of-the-art on 13 languages and the German CoNLL2009 corpus. |
OPT-Tree: Speculative Decoding with Adaptive Draft Tree Structure (2025.tacl-1)
Copied to clipboard
| Challenge: | Autoregressive language models generate one token in one step, limiting inference efficiency . Existing methods do not adapt to different situations to maximize acceptance length . speculative decoding has shown great potential for lossless acceleration . |
| Approach: | They propose an algorithm to construct adaptive and scalable draft trees for autoregressive language models. |
| Outcome: | Experimental results show that OPT-Tree outperforms existing draft trees and achieves speed-up ratio of up to 3.2 compared with autoregressive decoding. |
LAD: LoRA-Adapted Diffusion (2025.emnlp-demos)
Copied to clipboard
| Challenge: | Autoregressive models dominate text generation but suffer from left-to-right decoding constraints that limit efficiency and bidirectional reasoning. |
| Approach: | They propose a framework for non-autoregressive generation that adapts LLaMA models for iterative, bidirectional sequence refinement using LoRA adapters. |
| Outcome: | The proposed framework adapts LLaMA models for iterative, bidirectional sequence refinement using LoRA adapters. |
KNU-HYUNDAI’s NMT system for Scientific Paper and Patent Tasks onWAT 2019 (D19-52)
Copied to clipboard
Cheoneum Park, Young-Jun Jung, Kihoon Kim, Geonyeong Kim, Jae-Won Jeon, Seongmin Lee, Junseok Kim, Changki Lee
| Challenge: | We submitted our transformer-based neural machine translation system to the translation tasks of the 6th workshop on Asian Translation (WAT 2019). |
| Approach: | They propose a transformer-based neural machine translation system for Chinese-Japanese, English-Japanese, and Korean->Japanoise translation tasks. |
| Outcome: | The proposed system performed well on the two translation tasks and was ranked first in terms of the BLEU scores in all the JPC2 subtasks. |
OpenPrompt: An Open-source Framework for Prompt-learning (2022.acl-demo)
Copied to clipboard
| Challenge: | Prompt-learning is a new paradigm in natural language processing, adapting pre-trained language models to cloze-style prediction, autoregressive modeling, or sequence to sequence generation. |
| Approach: | They propose a framework for prompt-learning that integrates pre-trained language models with a unified framework. |
| Outcome: | The proposed framework is easy to use and flexible enough to integrate with other frameworks. |
Lexical Substitution as Causal Language Modeling (2024.starsem-1)
Copied to clipboard
| Challenge: | Existing methods for lexical substitution task lacks autoregressive decoding capabilities. |
| Approach: | They propose a framework that uses causal language modeling (CLM) for lexical substitution task. |
| Outcome: | The proposed system outperforms GeneSis, the best previously published supervised LST method. |
A Study of Non-autoregressive Model for Sequence Generation (2020.acl-main)
Copied to clipboard
| Challenge: | Non-autoregressive (NAR) models generate all tokens in parallel, resulting in faster generation speed compared to autoregressive models. |
| Approach: | They propose to use knowledge distillation and source-target alignment to bridge the gap between NAR and autoregressive models in various tasks. |
| Outcome: | The proposed techniques can speed up NAR models in some tasks but not all . the proposed techniques reduce target token dependency while allowing for faster inference . |
GL-GIN: Fast and Accurate Non-Autoregressive Model for Joint Multiple Intent Detection and Slot Filling (2021.acl-long)
Copied to clipboard
| Challenge: | Existing joint models for multi-intent SLU only consider intent detection while ignoring slot filling task. |
| Approach: | They propose a non-autoregressive model for joint multiple intent detection and slot filling . their framework is 11.5 times faster than existing joint models . |
| Outcome: | The proposed model is 11.5 times faster than existing models and is faster than current models. |
fairseq Sˆ2: A Scalable and Integrable Speech Synthesis Toolkit (2021.emnlp-demo)
Copied to clipboard
| Challenge: | Speech synthesis is the task of generating speech waveforms with desired characteristics, including but not limited to textual content, speaker identity, and speaking styles. |
| Approach: | They propose a fairseq extension for speech synthesis that implements autoregressive and non-AR text-to-speech models and their multi-speaker variants. |
| Outcome: | The proposed extension can train autoregressive and non-AR models and their multi-speaker variants with less curated data and has automatic metrics to facilitate faster iteration and analysis. |
Non-Autoregressive Text Generation with Pre-trained Language Models (2021.eacl-main)
Copied to clipboard
| Challenge: | Autoregressive generation models generate tokens in a left-to-right, token-by-token fashion, resulting in lag in inference. |
| Approach: | They propose to use BERT as the backbone of a non-autoregressive generation model for greatly improved performance. |
| Outcome: | The proposed model outperforms existing non-autoregressive models and achieves competitive performance with many strong autoregressive model. |
BOLT: Fast Energy-based Controlled Text Generation with Tunable Biases (2023.acl-short)
Copied to clipboard
| Challenge: | Energy-based models (EBMs) have gained popularity for controlled text generation due to their high applicability to a wide range of constraints. |
| Approach: | They propose a language model with tunable biases to adjust the language model’s output logits. |
| Outcome: | The proposed model maintains the generator’s autoregressive nature to assert a strong control on token-wise conditional dependencies and overall fluency, and converges faster. |
EDITOR: An Edit-Based Transformer with Repositioning for Neural Machine Translation with Soft Lexical Constraints (2021.tacl-1)
Copied to clipboard
| Challenge: | Empirically, EDITOR uses soft lexical constraints more effectively than the Levenshtein Transformer while speeding up decoding dramatically compared to constrained beam search. |
| Approach: | They propose an Edit-Based TransfOrmer with Repositioning that integrates lexical preferences into output sequences by iterative editing hypotheses. |
| Outcome: | The proposed model uses soft lexical constraints more effectively than the Levenshtein Transformer while speeding up decoding dramatically compared to constrained beam search. |
Reinforcement Learning for Edit-Based Non-Autoregressive Neural Machine Translation (2024.naacl-srw)
Copied to clipboard
| Challenge: | Non-autoregressive (NAR) language models have a performance gap due to the large decoding space and difficulty in capturing dependency between target words accurately. |
| Approach: | They propose to use reinforcement learning to enhance the performance of edit-based NAR models by using stepwise reward maximization and episodic reward maximisation. |
| Outcome: | The proposed model outperforms autoregressive models in the evaluation of an edit-based model. |
PADA: Example-based Prompt Learning for on-the-fly Adaptation to Unseen Domains (2022.tacl-1)
Copied to clipboard
| Challenge: | Domain Adaptation (DA) algorithms suffer degradation when applied to out-of-distribution examples. |
| Approach: | They propose an example-based autoregressive Prompt learning algorithm for on-the-fly Any-Domain Adaptation . the algorithm is trained to generate a unique prompt that maps the test example to a semantic space . |
| Outcome: | The proposed model outperforms baselines in 14 multi-source adaptation scenarios. |
Probabilistically Masked Language Model Capable of Autoregressive Generation in Arbitrary Word Order (2020.acl-main)
Copied to clipboard
| Challenge: | Large-scale pretrained language models such as masked language model (MLM) have brought significant improvements to many NLU and NLG tasks. |
| Approach: | They propose a probabilistic masking scheme for the masked language model and a model with a uniform prior distribution on the masking ratio. |
| Outcome: | The proposed model outperforms BERT on a bunch of downstream NLG tasks. |
Train Once, and Decode As You Like (2020.coling-main)
Copied to clipboard
| Challenge: | Existing approaches to machine translation support autoregressive, semi-autoregressive and refinement-based non-auto-regressives. |
| Approach: | They propose a unified approach for supporting different generation manners of machine translation including autoregressive, semi-autoregressive and refinement-based non-auto-regressives. |
| Outcome: | The proposed approach achieves better or competitive translation performance compared with strong baseline models in all the settings. |
Benchmarking Diffusion Models for Machine Translation (2024.eacl-srw)
Copied to clipboard
| Challenge: | Diffusion models have shown great potential on many generative tasks, but their application to natural language processing (NLP) is still a less explored direction. |
| Approach: | They adapt two diffusion-based text generation models, Diffusion-LM and DiffuSeq, to perform machine translation. |
| Outcome: | The proposed models struggle more on long-range dependencies than other models. |
GLM: General Language Model Pretraining with Autoregressive Blank Infilling (2022.acl-long)
Copied to clipboard
| Challenge: | Existing pretraining frameworks do not perform well for all tasks of three main categories, such as natural language understanding (NLU), unconditional generation, and conditional generation. |
| Approach: | They propose a general language model based on autoregressive blank infilling to address this challenge. |
| Outcome: | The proposed model outperforms BERT, T5, and GPT on a wide range of tasks across NLU, conditional and unconditional generation tasks. |
Entity Linking in the Job Market Domain (2024.findings-eacl)
Copied to clipboard
| Challenge: | In Natural Language Processing, entity linking (EL) has centered around Wikipedia, but yet remains underexplored for the job market domain. |
| Approach: | They propose to use a bi-encoder and an autoregressive model to link fine-grained span-level skill mentions to a specific taxonomy entry to quantify labor market demands. |
| Outcome: | The proposed model outperforms GENRE in strict evaluation, but performs better in loose evaluation. |
Why Are Positional Encodings Nonessential for Deep Autoregressive Transformers? A Petroglyph Revisited (2025.findings-acl)
Copied to clipboard
| Challenge: | Autoregressive Transformer language models do not require explicit positional encodings (PEs) this is because a cascade of (permutation invariant) set processors collectively exhibit sequence-sensitive behavior in the autoregressively setting. |
| Approach: | They propose to explain why autoregressive Transformers require explicit positional encodings (PEs) this property has been known since early efforts adopting the Transformer for language modeling . |
| Outcome: | The proposed model can distinguish sequences with permuted tokens without the need for explicit PEs. |
xLM: A Python Package for Non-Autoregressive Language Models (2026.eacl-demo)
Copied to clipboard
| Challenge: | Autoregressive language models generate text sequentially from left to right by adding one token at a time. |
| Approach: | They propose a python package that provides a suite of small non-autoregressive language models that can be used by researchers. |
| Outcome: | The proposed package makes implementing small non-autoregressive language models faster and provides a suite of pre-trained models that can be used by the research community. |
Mix and Match: Learning-free Controllable Text Generationusing Energy Language Models (2022.acl-long)
Copied to clipboard
| Challenge: | Recent work on controlled text generation has required attribute-based fine-tuning of the base language model or restricted the parameterization of the attribute discriminator. |
| Approach: | They propose a global score-based alternative for controllable text generation that combines arbitrary pre-trained black-box models for achieving desired attributes in the generated text. |
| Outcome: | The proposed method outperforms methods that require extra training or fine-tuning . the proposed method is based on a model with energy values of a linear combination of scores from black-box models . |
Fast Bilingual Grapheme-To-Phoneme Conversion (2022.naacl-industry)
Copied to clipboard
| Challenge: | Autoregressive transformers (ART)-based grapheme-to-phoneme (G2P) models have been proposed for bi/multilingual text-to speech systems. |
| Approach: | They propose a bilingual grapheme-to-phoneme (G2P) model with autoregressive transformers for fast and exact decoding and data augmentation for predicting output length. |
| Outcome: | The proposed model achieves better performance than the previous model and 2700% faster inference speed. |
Jointly Masked Sequence-to-Sequence Model for Non-Autoregressive Neural Machine Translation (2020.acl-main)
Copied to clipboard
| Challenge: | masked language models have been used for natural language processing tasks but few studies have adopted it in the sequence-to-sequence models. |
| Approach: | They propose to combine encoder and decoder to train a masked sequence-to-sequence model . they propose to train the encoder more rigorously by masking the encoded input . |
| Outcome: | The proposed model achieves 27.69/32.24 BLEU scores on English-German/German-English tasks with 5+ times speed up compared with an autoregressive model. |
Transparency Helps Reveal When Language Models Learn Meaning (2023.tacl-1)
Copied to clipboard
| Challenge: | Existing language models are trained to optimize unsupervised objectives on text . despite their centrality, current models do not represent natural language semantics well . |
| Approach: | They show that autoregressive and masked language models learn to emulate semantic relations between expressions when context-dependent . they argue that a learner that has access to all Java code can never learn execution . |
| Outcome: | a new study shows that language models fail to represent natural language semantics well . the authors show that the model learning fails when denotations are changed to be context-dependent . |
Revisiting the Markov Property for Machine Translation (2024.findings-eacl)
Copied to clipboard
| Challenge: | Statistical machine translation (SMT) has employed Markov models, but autoregressive models are less effective. |
| Approach: | They propose to use a Markov Autoregressive Transformer to model neural machine translation using four WMT benchmarks. |
| Outcome: | The proposed model performs better than autoregressive models on four WMT benchmarks. |
Scaling Neural ITN for Numbers and Temporal Expressions in Tamil: Findings for an Agglutinative Low-resource Language (2023.emnlp-industry)
Copied to clipboard
| Challenge: | Inverse Text Normalisation (ITN) is a textrewriting task that converts verbalized text to written form. |
| Approach: | They propose to use a seq2seq model, a non-autoregressive text editor and a sequence tagger + rules combination to fine-tune three pre-trained neural models. |
| Outcome: | The proposed model improves with bootstrapping and data augmentation, and bootstrap alone shows a percentage improvement of 14.12 %. |
Control-DAG: Constrained Decoding for Non-Autoregressive Directed Acyclic T5 using Weighted Finite State Automata (2024.naacl-short)
Copied to clipboard
| Challenge: | Existing non-autoregressive (NAR) models fail to generate specified entity names in up to 40% of responses and produce OOV errors. |
| Approach: | They propose a constrained decoding algorithm for Directed Acyclic T5 model which offers lexical, vocabulary and length control. |
| Outcome: | The proposed model significantly improves on Schema Guided Dialogue and DART datasets, establishing strong results for Task-Oriented Dialog and Data-to-Text NLG. |
Is Encoder-Decoder Redundant for Neural Machine Translation? (2022.aacl-main)
Copied to clipboard
| Challenge: | Encoder-decoder architecture is widely adopted for sequence-to-sequence modeling tasks. |
| Approach: | They propose to combine bilingual and multilingual translations to train a language model to do translation. |
| Outcome: | The proposed approach performs on par with the baseline encoder-decoder Transformer . the proposed approach is compared with the translation model in the target language . |
Semi-Autoregressive Neural Machine Translation (D18-1)
Copied to clipboard
| Challenge: | Existing approaches to neural machine translation are typically autoregressive but suffer from low parallelizability and thus slow at decoding long sequences. |
| Approach: | They propose a semi-autoregressive Transformer model for fast sequence generation that keeps the autoregressive property in global but relieves in local . |
| Outcome: | The proposed model achieves 5.58 speedup while maintaining 88% translation quality, significantly better than previous non-autoregressive methods. |
BeLLM: Backward Dependency Enhanced Large Language Model for Sentence Embeddings (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing LLMs adopt autoregressive architectures without explicit backward dependency modeling. |
| Approach: | They propose a backward dependency enhanced large language model that transforms attention layers from uni-to-bi-directional to learn sentence embeddings. |
| Outcome: | The proposed model achieves state-of-the-art performance in varying scenarios. |
Non-Autoregressive Neural Machine Translation with Consistency Regularization Optimized Variational Framework (2022.naacl-main)
Copied to clipboard
| Challenge: | Variational Autoencoder (VAE) is an effective framework to model the interdependency for non-autoregressive neural machine translation (NAT). |
| Approach: | They propose to use Variational Autoencoder to model interdependency for non-autoregressive neural machine translation (NAT) a posterior consistency regularization approach is proposed to improve translation quality . |
| Outcome: | The proposed model is 1.5/0.7 and 0.8/0.3 BLEU points faster than the baseline model. |
CheckersGPT: Learning World Models through Language Modeling (2024.acl-srw)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have shown impressive performance on various tasks, but the underlying process behind predicting the desired next token remains a black box. |
| Approach: | They train a GPT-style autoregressive language model using only the next character prediction objective and then train corresponding model with different layer sizes. |
| Outcome: | The proposed model shows a hint of learning a world model representation of the board positions on a simulated game of checkers and human gameplay dataset. |
Pan More Gold from the Sand: Refining Open-domain Dialogue Training with Noisy Self-Retrieval Generation (2022.coling-1)
Copied to clipboard
| Challenge: | Existing methods for generating open-domain dialogue systems underutilize training data. |
| Approach: | They propose a retrieval-generation training framework that takes advantage of heterogeneous training data by considering them as "evidence" they use BERTScore retrieval framework which gives better qualities of the training data, they show . |
| Outcome: | The proposed method performs well on zero-shot experiments and is more robust to real-world data. |
DiffusPoll: Conditional Text Diffusion Model for Poll Generation (2024.findings-acl)
Copied to clipboard
| Challenge: | Social media platforms manipulate public opinion through sheer numbers and cause biases, authors say . they say new paradigm for poll generation can generate high-quality samples while preserving diversity . |
| Approach: | They propose a non-autoregressive diffusion model that uses masks to generate polls . they use attribute tags to enhance the quality of polls and to diversify poll options . |
| Outcome: | The proposed model matches the Transformer model while offering greater diversity and quality. |
Efficient Sample-Specific Encoder Perturbations (2024.naacl-short)
Copied to clipboard
| Challenge: | Existing approaches to training encoder-decoder systems often depend on teacher-forcing with the likelihood criteria, e.g. next token prediction of the reference sequence. |
| Approach: | They propose an inference-efficient way to modify the behaviour of an encoder-decoder system according to a specific attribute of interest by using a small proxy network. |
| Outcome: | The proposed framework improves the COMET performance of Flan-T5 on Machine Translation and the WER of Whisper foundation models on Speech Recognition. |
Aleph-Alpha-GermanWeb: Improving German-language LLM pre-training with model-based data curation and synthetic data generation (2026.eacl-long)
Copied to clipboard
Thomas F Burns, Letitia Parcalabescu, Stephan Waeldchen, Michael Barlow, Gregor Ziegltrum, Volker Stampa, Bastian Harren, Björn Deiseroth
| Challenge: | Recent studies show that data quality can significantly boost performance and training efficiency for large language models. |
| Approach: | They propose a German-language dataset curation pipeline that combines heuristic and model-based filtering techniques with synthetic data generation. |
| Outcome: | The proposed pipeline can be used to create a large-scale German pre-training dataset using common Crawl web data, fineweb2 and synthetically generated data conditioned on real, organic web data. |
An Empirical Study on Adversarial Attack on NMT: Languages and Positions Matter (2021.acl-short)
Copied to clipboard
| Challenge: | Existing approaches to generating NMT adversarial examples inject perturbations into source sentences or target translations to improve the robustness of NMT models. |
| Approach: | They investigate adversarial attack on NMT from two aspects: languages (the source vs. the target language) and positions (front v. rear). |
| Outcome: | The proposed approach is more effective than adversarial attacks by sampling positions randomly or according to gradients. |
LEAF: Large Language Diffusion Model for Time Series Forecasting (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Recent work has applied large language models (LLMs) into time series forecasting, but they lack an understanding of holistic temporal patterns with potential error accumulation. |
| Approach: | They propose a framework that marries Larg e Langu age Diffusion Model with time series forecasting (LEAF) they propose converting time series into tokens and adopting language diffusion models to capture temporal dependencies. |
| Outcome: | The proposed framework generates future predictions with a diffusion model from a holistic view. |
RAMQA: A Unified Framework for Retrieval-Augmented Multi-Modal Question Answering (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing ranking methods rely on small encoder-based ranking models, which are incompatible with modern decoder--based generative large language models (LLMs) Existing methods based on small LLaVA rankers are incompatible with advanced LLMs. |
| Approach: | They propose a framework that combines learning-to-rank methods with generative permutation-enhanced ranking techniques. |
| Outcome: | The proposed framework improves on two benchmarks, WebQA and MultiModalQA, showing significant improvements over baselines. |
TRACE: Training and Inference-Time Interpretability Analysis for Language Models (2025.emnlp-demos)
Copied to clipboard
| Challenge: | Existing tools for interpretability analysis of transformer models are post hoc, rely on scalar metrics or require nontrivial integration effort. |
| Approach: | They propose a modular toolkit for training and inference-time interpretability analysis of transformer models. |
| Outcome: | Experiments with autoregressive transformers show that TRACE reveals developmental phenomena overlooked by traditional scalar metrics such as loss or accuracy. |
How Large Language Models are Transforming Machine-Paraphrase Plagiarism (2022.emnlp-main)
Copied to clipboard
| Challenge: | Autoregressive paraphrasing tools can be used to generate convincing plagiarized texts with minimal effort. |
| Approach: | They evaluate the detection performance of large autoregressive models for machine-paraphrase generation on scientific articles from arXiv, student theses, and Wikipedia. |
| Outcome: | The proposed models generate paraphrases indistinguishable from original work and human experts rate the quality of generated examples as high as originals. |
PrefixNLI: Detecting Factual Inconsistencies as Soon as They Arise (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are prone to factually inconsistent statements, known as hallucinations. |
| Approach: | They propose to train a specialized model that detects inconsistencies over text prefixes to improve generation faithfulness by 5-14 F1 points. |
| Outcome: | The proposed model outperforms baseline models by 5-14 F1 points in prefix-level entailment. |
Autoregressive Structured Prediction with Language Models (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Recent years have seen a paradigm shift in NLP towards using pretrained language models for a wide range of tasks. |
| Approach: | They propose to model structures as sequences of actions in autoregressive manner with PLMs . their approach allows in-structure dependencies to be learned without any loss . |
| Outcome: | The proposed approach achieves state-of-the-art on all structured prediction tasks. |
Iterative Refinement in the Continuous Space for Non-Autoregressive Neural Machine Translation (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive inference procedures that refine in token space often require computational overhead. |
| Approach: | They propose an efficient inference procedure that iteratively refines translation purely in the continuous space using a latent variable instead of the latent variables. |
| Outcome: | The proposed procedure is twice as efficient and more effective than the existing EM-like inference procedure. |
Diving into the Decoding Space of Non-Autoregressive Models via Lexically Constrained Search (2026.acl-short)
Copied to clipboard
| Challenge: | Non-autoregressive (NAR) models have been mainly developed to improve decoding efficiency. |
| Approach: | They propose a search-based decoding algorithm which is comparable to the autoregressive Grid Beam Search (GBS) method. |
| Outcome: | The proposed method does not suffer from the MAP degradation issue as the autoregressive method does. |
Don’t Let Discourse Confine Your Model: Sequence Perturbations for Improved Event Language Models (2021.acl-short)
Copied to clipboard
| Challenge: | Existing approaches to train event language models on text constrain them to follow discourse order of events. |
| Approach: | They propose a method to perturb event sequences so that they can relax model dependence on text order. |
| Outcome: | The proposed technique improves performance on applications and out-of-domain events data. |
A Better Way to Do Masked Language Model Scoring (2023.acl-short)
Copied to clipboard
| Challenge: | Existing methods to estimate log-likelihood of sentences are not available for masked language models . et al., 2020: e.on.org/sci/sql/src/spll-word-l2r estimates sentence pseudo-log-likellihoods for a given sentence under autoregressive models are difficult to perform on controlled benchmarks. |
| Approach: | They propose to estimate sentence pseudo-log-likelihood scores by successively masking each sentence token, retrieving its score using the rest of the sentence as context, and summing the resulting values. |
| Outcome: | The proposed method outperforms both the original and the adapted method on masked language models. |
CEV-LM: Controlled Edit Vector Language Model for Shaping Natural Language Generations (2024.eacl-long)
Copied to clipboard
| Challenge: | Existing control approaches primarily adjust the semantic (e.g., emotion, topics), structural (e-speech, parts-of-seech), and lexical (el-s-sp-s) properties of text, but are insufficient to accomplish complex objectives such as pacing which control the complexity and readability of the text. |
| Approach: | They propose a lightweight semi-autoregressive language model that uses edit vectors to control three complementary metrics that quantify the shape of text. |
| Outcome: | The proposed model provides significantly more targeted and precise control of speed, volume, and circuitousness while using less training data, and containing fewer parameters. |
Non-Autoregressive Machine Translation with Latent Alignments (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive machine translation methods are lacking in the field of latent alignments. |
| Approach: | They propose two strong methods for non-autoregressive machine translation that model latent alignments with dynamic programming. |
| Outcome: | The proposed models achieve state-of-the-art on the WMT’14 EnDe task, compared with the autoregressive Transformer baseline. |
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. |
GEC-DePenD: Non-Autoregressive Grammatical Error Correction with Decoupled Permutation and Decoding (2023.acl-long)
Copied to clipboard
| Challenge: | grammatical error correction is an important NLP task that is usually solved with autoregressive sequence-to-sequence models. |
| Approach: | They propose a non-autoregressive approach to grammatical error correction that decouples a permutation network and a decoder network that fills in specific tokens. |
| Outcome: | The proposed approach improves over previously known non-autoregressive methods and reaches the level of autoregressive approaches that do not use language-specific synthetic data generation methods. |
Diffusion-NAT: Self-Prompting Discrete Diffusion for Non-Autoregressive Text Generation (2024.eacl-long)
Copied to clipboard
| Challenge: | Existing non-autoregressive (NAR) text-to-text generation methods are unable to generate coherent and fluent texts due to discrete nature of text. |
| Approach: | They propose to integrate discrete diffusion models (DDM) into NAR text-to-text generation and integrate BART to improve the performance. |
| Outcome: | The proposed method outperforms competing methods and surpasses autoregressive methods on 7 datasets. |
How much do contextualized representations encode long-range context? (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing studies of contextualized representations focus on short sequences of tens to hundreds of tokens, whereas modern language models handle hundreds of thousands of token in a single context window. |
| Approach: | They use a perturbation setup and a metric to capture contextualization of long-range patterns from the perspective of representation geometry. |
| Outcome: | The proposed model can encode long-range contexts, but it's not fully recurrent, the authors say . their results suggest improvements in existing language models . |
Detection-Correction Structure via General Language Model for Grammatical Error Correction (2024.acl-long)
Copied to clipboard
| Challenge: | Grammatical error correction (GEC) is a task dedicated to rectifying texts with minimal edits. |
| Approach: | They propose a detection-correction structure based on the general language model which integrates detection and correction into a single model. |
| Outcome: | The proposed model outperforms the state-of-the-art models on English and Chinese datasets. |
Integrating Translation Memories into Non-Autoregressive Machine Translation (2023.eacl-main)
Copied to clipboard
| Challenge: | Non-autoregressive machine translation (NAT) has made great progress, but most studies focus on standard translation tasks. |
| Approach: | They propose to train an edit-based NAT model with a Translation Memory (TM) they propose to modify the data presentation and introduce an extra deletion operation to reduce decoding load. |
| Outcome: | The proposed model performs on par with an autoregressive approach while reducing the decoding load. |
Enriching Non-Autoregressive Transformer with Syntactic and Semantic Structures for Neural Machine Translation (2021.eacl-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive models have boosted the efficiency of neural machine translation, but their performance is significantly worse than that of autoregressive counterparts. |
| Approach: | They propose to incorporate syntactic and semantic structures among natural languages into a non-autoregressive Transformer for the task of neural machine translation. |
| Outcome: | The proposed model achieves faster speed and keeps translation quality compared with other models. |
Drop Dropout on Single Epoch Language Model Pretraining (2025.findings-acl)
Copied to clipboard
| Challenge: | Initial dropout was seen as a breakthrough regularization technique that reduced overfitting, yet single-epoch pretraining tasks common to modern LLMs yield minimal overfit. |
| Approach: | They propose to use dropout during single-epoch pretraining to reduce overfitting in language modeling, morpho-syntax, question answering, and MNLI to improve performance. |
| Outcome: | The results show that dropout is not used in large LLMs and improves performance in language modeling, morpho-syntax, question answering, and MNLI. |
Syntactically Supervised Transformers for Faster Neural Machine Translation (P19-1)
Copied to clipboard
| Challenge: | Standard decoders for neural machine translation generate a single token per timestep, which slows inference . a series of controlled experiments demonstrates that SynST decodes sentences 5x faster than the baseline autoregressive Transformer. |
| Approach: | They propose a syntactically supervised Transformer that generates all target tokens in one shot . synST is a variant of the Transformer architecture that autoregressively predicts a chunked parse tree . |
| Outcome: | The proposed method decodes sentences 5x faster than the baseline method on En-De and En-Fr datasets while achieving higher BLEU scores. |
mT6: Multilingual Pretrained Text-to-Text Transformer with Translation Pairs (2021.emnlp-main)
Copied to clipboard
Zewen Chi, Li Dong, Shuming Ma, Shaohan Huang, Saksham Singhal, Xian-Ling Mao, Heyan Huang, Xia Song, Furu Wei
| Challenge: | Multilingual T5 pretrains a sequence-to-sequence model on monolingual texts, but it has shown promising results on many cross-lingual tasks. |
| Approach: | They propose a partially non-autoregressive objective for text-to-text pre-training and propose mT6 to improve cross-lingual transferability over multilingual T5. |
| Outcome: | The proposed model improves cross-lingual transferability over existing models. |
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. |
DiffuVST: Narrating Fictional Scenes with Global-History-Guided Denoising Models (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for visual storytelling suffer from low inference speed and are not well-suited for synthetic scenes. |
| Approach: | They propose a diffusion-based system that generates visual descriptions as a single conditional denoising process. |
| Outcome: | The proposed system improves inter-sentence coherence and image-to-text fidelity. |
Non-Autoregressive Machine Translation: It’s Not as Fast as it Seems (2022.naacl-main)
Copied to clipboard
| Challenge: | Efficient machine translation models are commercially important as they can increase inference speeds, reduce costs and carbon emissions. |
| Approach: | They compare NAR models with autoregressive models to evaluate their performance . they point out flaws in evaluation methodology and argue for consistent evaluation . |
| Outcome: | The proposed model is faster on GPUs, but slower under more realistic usage conditions. |
Tug-of-war between idioms’ figurative and literal interpretations in LLMs (2026.eacl-long)
Copied to clipboard
| Challenge: | We find that idioms have non-compositional figurative interpretations that diverge from the idiomatic literal interpretation. |
| Approach: | They employ causal tracing to analyze how pretrained causal transformers deal with idiom ambiguity. |
| Outcome: | The proposed model leverages the idiom's context and refines it if it conflicts with the retrieved interpretation. |
NAST: A Non-Autoregressive Generator with Word Alignment for Unsupervised Text Style Transfer (2021.findings-acl)
Copied to clipboard
| Challenge: | Autoregressive text style transfer models often ignore part of the source sentence and generate some irrelevant words with strong styles. |
| Approach: | They propose a non-autoregressive generator for unsupervised text style transfer which explicitly models word alignments to suppress irrelevant words. |
| Outcome: | The proposed generator significantly improves performance and provides explainable word alignments. |
Guided by the Plan: Enhancing Faithful Autoregressive Text-to-Audio Generation with Guided Decoding (2026.eacl-long)
Copied to clipboard
| Challenge: | Autoregressive (AR) models excel at generating temporally coherent audio by producing tokens sequentially, yet they often falter in faithfully following complex textual prompts. |
| Approach: | They propose a lightweight auxiliary model trained with a GAE-inspired objective to predict final instruction-following quality from partial generations. |
| Outcome: | The proposed model achieves 10 points improvement in CLAP score over baseline AR models while maintaining computational parity with best-of-N decoding. |
TESS: Text-to-Text Self-Conditioned Simplex Diffusion (2024.eacl-long)
Copied to clipboard
Rabeeh Karimi Mahabadi, Hamish Ivison, Jaesung Tae, James Henderson, Iz Beltagy, Matthew Peters, Arman Cohan
| Challenge: | Existing models for diffusion generation are expensive and discrete, resulting in a large number of diffusion steps to generate text. |
| Approach: | They propose a text diffusion model that is fully non-autoregressive and employs a new form of self-conditioning and applies the diffusion process on the logit simplex space rather than the learned embedding space. |
| Outcome: | The proposed model outperforms state-of-the-art non-autoregressive models, requires fewer diffusion steps with minimal drop in performance, and is competitive with pretrained autoregressive sequence-to-sequence models. |
Source and Target Bidirectional Knowledge Distillation for End-to-end Speech Translation (2021.naacl-main)
Copied to clipboard
| Challenge: | End-to-end speech translation models can be trained to leverage source text . however, since the input modalities are different, it is difficult to leverage the source text successfully. |
| Approach: | They propose to leverage source transcriptions via pre-training and joint training with ASR and NMT tasks. |
| Outcome: | The proposed model predicts paraphrased transcriptions as an auxiliary task with a single decoder. |
Don’t Take It Literally: An Edit-Invariant Sequence Loss for Text Generation (2022.naacl-main)
Copied to clipboard
Guangyi Liu, Zichao Yang, Tianhua Tao, Xiaodan Liang, Junwei Bao, Zhen Li, Xiaodong He, Shuguang Cui, Zhiting Hu
| Challenge: | Neural text generation models are typically trained by maximizing log-likelihood with the sequence cross entropy (CE) loss. |
| Approach: | They propose an Edit-Invariant Sequence Loss method which computes the matching loss of a target sequence with all n-grams in the generated sequence. |
| Outcome: | The proposed method outperforms the common CE loss and strong baselines on a wide range of tasks. |
SlotRefine: A Fast Non-Autoregressive Model for Joint Intent Detection and Slot Filling (2020.emnlp-main)
Copied to clipboard
| Challenge: | Slot filling and intent detection are two main tasks in spoken language understanding systems. |
| Approach: | They propose a non-autoregressive slot filling model with two-pass iteration mechanism to handle uncoordinated slots problem. |
| Outcome: | The proposed model significantly outperforms previous models in slot filling task while speeding up decoding. |
Align-Refine: Non-Autoregressive Speech Recognition via Iterative Realignment (2021.naacl-main)
Copied to clipboard
| Challenge: | Non-autoregressive encoder-decoder models improve decoding speed, but generation quality suffers . editing at the level of output sequences limits model flexibility. |
| Approach: | They propose *iterative realignment* which iteratively realigns connectionist temporal alignments. |
| Outcome: | The proposed model matches an autoregressive baseline with a 14x speedup on the WSJ dataset; on LibriSpeech, it achieves an LM-free test-other WER of 9.0% (19% relative improvement on comparable work). |
Glancing Transformer for Non-Autoregressive Neural Machine Translation (2021.acl-long)
Copied to clipboard
| Challenge: | Existing non-autoregressive neural machine translation methods are either inferior to Transformer or require multiple decoding passes, leading to reduced speedup. |
| Approach: | They propose a Glancing Language Model (GLM) for single-pass parallel generation models and Glancing Transformer (GLAT) with only single- pass decoding, GLAT is able to generate high-quality translation with 8-15 speedup. |
| Outcome: | The proposed model outperforms all previous non-autoregressive methods on multiple language directions and is nearly comparable to Transformer. |
EdiT5: Semi-Autoregressive Text Editing with T5 Warm-Start (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Pre-trained seq2seq models have established strong baselines for text-to-text transduction tasks. |
| Approach: | They propose a semi-autoregressive text-editing approach that combines the strengths of non-auto-regressively text- editing and autoregressive decoding. |
| Outcome: | The proposed model is faster at inference times than conventional models while being capable of modeling flexible input-output transformations. |
Mapping the Course for Prompt-based Structured Prediction (2026.eacl-long)
Copied to clipboard
| Challenge: | Large language models have demonstrated strong performance in a wide-range of language tasks without task-specific fine-tuning. |
| Approach: | They combine large language models with combinatorial inference to marry predictive power of LLMs with structural consistency provided by inference methods. |
| Outcome: | The proposed model incorporates symbolic inference to provide consistent and accurate predictions on challenging tasks. |
Span Pointer Networks for Non-Autoregressive Task-Oriented Semantic Parsing (2021.findings-emnlp)
Copied to clipboard
Akshat Shrivastava, Pierce Chuang, Arun Babu, Shrey Desai, Abhinav Arora, Alexander Zotov, Ahmed Aly
| Challenge: | a novel approach to map utterances to semantic frames is based on non-autoregressive parsers that shift the decoding task from text generation to span prediction. |
| Approach: | They propose a non-autoregressive, task-oriented parser which shifts the decoding task from text generation to span prediction and produces endpoints as opposed to text. |
| Outcome: | The proposed model bridges the quality gap between non-autoregressive and autoregressive parsers, achieving 87 EM on TOPv2 and shows a 70% reduction in latency and 83% reduction in memory at beam size 5 compared to prior non-regressives. |
Dynamic and Efficient Inference for Text Generation via BERT Family (2023.acl-long)
Copied to clipboard
| Challenge: | Existing methods to improve performance of pre-trained language models are limited due to large-scale parameters and the universal autoregressive decoding paradigm. |
| Approach: | They propose a novel fine-tuning method which can make a single pre-trained model support Dynamic and Efficient infERence and achieve an adaptive trade-off between model performance and latency. |
| Outcome: | The proposed method achieves higher BLEU scores than the strong autoregressive Transformer model on translation tasks with 3 12 times speedup and faster inference speed compared with the BART model on four GLGE benchmark tasks. |
PLANET: Dynamic Content Planning in Autoregressive Transformers for Long-form Text Generation (2022.acl-long)
Copied to clipboard
| Challenge: | Existing methods for text generation still suffer from incoherence problems . Neural sequence-to-sequence (seq2sequ) models generate fluent results . |
| Approach: | They propose a novel generation framework that leverages autoregressive self-attention mechanism to conduct content planning and surface realization dynamically. |
| Outcome: | The proposed framework outperforms baseline models and generates more coherent texts with richer contents. |
Saber: Efficient Sampling with Adaptive Acceleration and Backtracking Enhanced Remasking for Diffusion Language Model in Code Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Diffusion language models (DLMs) offer advantages in parallel generation and bidirectional context modeling, but they face a critical trade-off between inference speed and output quality for tasks with strict structural constraints such as code generation. |
| Approach: | They propose an efficient sampling algorithm that reduces the number of tokens unmasked per step based on the model’s evolving confidence. |
| Outcome: | The proposed method improves Pass@1 accuracy by 1.9% while achieving 251.4% inference speedup. |
In-Context Example Ordering Guided by Label Distributions (2024.findings-naacl)
Copied to clipboard
| Challenge: | Prior work has shown that ICL is sensitive to different natural language instructions and different orderings of in-context examples. |
| Approach: | They propose two principles for in-context example ordering guided by model’s probability predictions. |
| Outcome: | The proposed model outperforms baseline models on 13 text classification datasets and nine autoregressive LLMs with 700M to 13B parameters. |
Thinking Clearly, Talking Fast: Concept-Guided Non-Autoregressive Generation for Open-Domain Dialogue Systems (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing models with seq2seq framework lack ability to effectively manage concept transitions . lack of concept management strategies might lead to incoherent dialogue due to loosely connected concepts . |
| Approach: | They propose a concept-guided non-autoregressive model for open-domain dialogue generation that learns to identify multiple associated concepts from a conceptual graph and a customized Insertion Transformer to perform concept-directed generation to complete a response. |
| Outcome: | The proposed model outperforms state-of-the-art models in automatic and human evaluations with substantially faster inference speed. |
Mask Tokens as Prophet: Fine-Grained Cache Eviction for Efficient dLLM Inference (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing cache eviction strategies for autoregressive language models fail to account for the role of mask tokens and specific characteristics in dLLMs. |
| Approach: | They propose a training-free cache eviction framework tailored to dLLMs that denies a fully masked sequence and allows parallel decoding at the expense of memory and computation. |
| Outcome: | The proposed framework reduces the cost of memory and cache eviction and improves efficiency by reducing allocation in intermediate layers and concentrating resources on prompt-preferring heads. |
Retrieval, Analogy, and Composition: A framework for Compositional Generalization in Image Captioning (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches fail to generalize well to concepts that are not observed during training. |
| Approach: | They propose a framework that revolves around probing several similar image caption training instances and performing analogical reasoning over relevant entities in retrieved prototypes. |
| Outcome: | The proposed framework improves on the widely used image captioning benchmarks and on composition-related evaluation metrics. |
A Label-Aware Autoregressive Framework for Cross-Domain NER (2022.findings-naacl)
Copied to clipboard
| Challenge: | Existing approaches to named entity recognition (NER) focus on reducing discrepancy between tokens and tokens, but transfer of valuable label information is often not considered or ignored. |
| Approach: | They propose a framework that borrows entity information from the source domain to enhance NER in the target domain. |
| Outcome: | The proposed model improves over the state-of-the-art model on several datasets. |
Improving Non-autoregressive Neural Machine Translation with Monolingual Data (2020.acl-main)
Copied to clipboard
| Challenge: | Neural machine translation is usually done via knowledge distillation from an autoregressive (AR) model. |
| Approach: | They leverage large monolingual corpora to improve the NAR model's performance by transferring the autoregressive model' s generalization ability while preventing overfitting. |
| Outcome: | The proposed methods on the WMT14 En-De and WMT16 En-Ro news translation tasks show that monolingual data augmentation improves the NAR model to approach the teacher AR model’s performance. |
Non-Autoregressive Neural Machine Translation: A Call for Clarity (2022.emnlp-main)
Copied to clipboard
| Challenge: | Non-autoregressive translation models require a single forward pass to generate the output sequence instead of iteratively producing each predicted token. |
| Approach: | They propose to use a single forward pass to generate the output sequence instead of iteratively producing each predicted token. |
| Outcome: | The proposed models improve translation quality and speed under third-party testing environments. |
MIDLM: Multi-Intent Detection with Bidirectional Large Language Models (2025.coling-main)
Copied to clipboard
| Challenge: | Existing models that use autoregressive architectures restrict the sharing of token information within a sentence. |
| Approach: | They propose a framework that integrates intent number detection and multi-intent selection to enable autoregressive LLMs to leverage bidirectional information awareness through post-training. |
| Outcome: | The proposed framework outperforms existing models and pretrained baselines in the multi-intent detection task. |
Autoregressive Pre-Training on Pixels and Texts (2024.emnlp-main)
Copied to clipboard
| Challenge: | pixel-based language modeling integrates visual and textual data to improve performance of language models. |
| Approach: | They propose a method that integrates visual and textual data into an autoregressive framework. |
| Outcome: | The proposed method improves performance of pixel-based language models by incorporating visual and textual data. |
DynaMo: Accelerating Language Model Inference with Dynamic Multi-Token Sampling (2024.naacl-long)
Copied to clipboard
| Challenge: | Rapid explosion in model sizes has resulted in high inference times . open-source LLMs are democratizing research in natural language processing . |
| Approach: | They propose a suite of multi-token prediction language models that reduce net inference times by leveraging traditional autoregressive weights. |
| Outcome: | The proposed model achieves same-quality generated text as baseline (Pythia-6.9B) with only 5.87% and 2.67% parameter and training time overheads. |
Chaining Simultaneous Thoughts for Numerical Reasoning (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Numerical reasoning over text is an essential skill for AI systems . structure modeling is effective, but structures restrict how a model should grasp the reasoning process . |
| Approach: | They propose a numerical reasoner that models reasoning steps using a directed acyclic graph without pre-defined decoding dependencies. |
| Outcome: | The proposed model produces diverse reasoning steps without pre-defined dependencies and compares relevant ones to reach a solution. |
Retrieving Examples from Memory for Retrieval Augmented Neural Machine Translation: A Systematic Comparison (2024.findings-naacl)
Copied to clipboard
| Challenge: | Existing approaches to extract examples from memory are limited, but the upstream retrieval step is still unexplored. |
| Approach: | They propose to use a standard autoregressive model, edit-based model and a large language model with in-context learning to investigate the effect of retrieval methods on translation scores. |
| Outcome: | The proposed architectures improve translation scores and increase diversity of examples. |
Retrieval as Generation: A Unified Framework with Self-Triggered Information Planning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing models that ground retrieval on external evidence are limited in their ability to implement retrieval-augmented generation. |
| Approach: | They propose a retrieval-augmented generation model that embeds retrieval control directly into generation. |
| Outcome: | The proposed model surpasses strong RAG baselines and uses substantially fewer parameters. |
Sequence-to-Sequence Knowledge Graph Completion and Question Answering (2022.acl-long)
Copied to clipboard
| Challenge: | Knowledge graph embedding (KGE) models represent each entity and relation of a knowledge graph (KG) with low-dimensional embeddable vectors. |
| Approach: | They propose to use an off-the-shelf encoder-decoder Transformer model to generate a knowledge graph embedding model that can be used for KG link prediction and incomplete KG question answering. |
| Outcome: | The proposed model outperforms baselines on multiple large-scale datasets without extensive hyperparameter tuning. |
Don’t Generate, Classify! Low-Latency Prompt Optimization with Structured Complementary Prompt (2026.eacl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated strong performance across diverse tasks, but their performance varies significantly across different prompts. |
| Approach: | They propose a framework that reframes prompt engineering as a classification problem. |
| Outcome: | The proposed framework improves answer quality by up to 26.5% in win rate compared to prior methods while reducing latency by upto 1,956 times. |
REBEL: Relation Extraction By End-to-end Language generation (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to extract relation triplets from text often involve multiple-step pipelines that propagate errors or are limited to a small number of relation types. |
| Approach: | They propose to use autoregressive seq2seq models to simplify Relation Extraction by expressing triplets as a sequence of text and a model that performs end-to-end relation extraction for more than 200 different relation types. |
| Outcome: | The proposed model achieves state-of-the-art on an array of Relation Extraction and Relation Classification benchmarks and achieves top performance in most of them. |
TABED: Test-Time Adaptive Ensemble Drafting for Robust Speculative Decoding in LVLMs (2026.findings-eacl)
Copied to clipboard
Minjae Lee, Wonjun Kang, Byeongkeun Ahn, Christian Classen, Kevin Galim, Seunghyuk Oh, Minghao Yan, Hyung Il Koo, Kangwook Lee
| Challenge: | Large Vision Language Models (LVLMs) are advanced models that process multiple modalities, such as images, audio, and video, alongside text. |
| Approach: | They propose to use a method to generate and verify draft tokens in parallel . they compare existing methods with small draft models and observe performance fluctuations . |
| Outcome: | The proposed method achieves an average walltime speedup of 1.74 over autoregressive decoding and a 5% improvement over single drafting methods. |
Knowledgeable In-Context Tuning: Exploring and Exploiting Factual Knowledge for In-Context Learning (2024.findings-naacl)
Copied to clipboard
| Challenge: | Existing studies have explored multiple aspects that affect the performance of large language models (LLMs) such as input-output mapping, extensive data resources, and the ability to train on labeled examples. |
| Approach: | They propose a framework that injects knowledge into LLMs during continual self-supervised pre-training and judiciously selects examples with high knowledge relevance. |
| Outcome: | The proposed framework outperforms baseline models and improves by more than 13% and 7% on text classification and question-answering tasks. |
Segment-Level Diffusion: A Framework for Controllable Long-Form Generation with Diffusion Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Diffusion models have shown promise in text generation, but often struggle with generating long, coherent, and contextually accurate text. |
| Approach: | They propose a framework that enhances diffusion-based text generation through text segmentation, robust representation training with adversarial and contrastive learning, and improved latent-space guidance. |
| Outcome: | The proposed framework improves diffusion-based text generation and improves scalability and fluency. |
Diffusion with Truncated Blocks: Fast and High-Quality Text Generation using Truncated Block Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Diffusion-based Large Language Models (dLLMs) generate text by iteratively denoising masked sequences. |
| Approach: | They propose a method that iteratively denoises masked sequences to reduce the model's attention dilution by token-level noise while models employing sequence-level noising exhibit a reduced effect. |
| Outcome: | The proposed method improves the performance and efficiency of Diffusion-based large language models by iterating on masked sequences. |
Diffusion vs. Autoregressive Language Models: A Text Embedding Perspective (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language model (LLM)-based embedding models surpass BERT and T5 on general-purpose text embeddable tasks. |
| Approach: | They propose to adopt diffusion language models for text embeddings to overcome limitations in unidirectional attention used during autoregressive pre-training. |
| Outcome: | The proposed model outperforms the existing LLM-based embedding model on reasoning tasks by 20% and 2% on traditional embeddable benchmarks. |
Threshold Filtering Packing for Supervised Fine-Tuning: Training Related Samples within Packs (2025.naacl-long)
Copied to clipboard
| Challenge: | Randomly concatenating data points can lead to cross-contamination due to the significant difference in their subject matter. |
| Approach: | They propose a method that randomly concatenates data of varying lengths until reaching the designed maximum length to optimize context length and reduce padding. |
| Outcome: | The proposed method significantly improves performance on GSM8K and HumanEval, and also improves fairness and accuracy by 15%. |
A Batch Normalized Inference Network Keeps the KL Vanishing Away (2020.acl-main)
Copied to clipboard
| Challenge: | Variational Autoencoder (VAE) is widely used to approximate a model’s posterior on latent variables. |
| Approach: | They propose to let the Kullback–Leibler divergence individual follow a distribution across the whole dataset and analyze that it is sufficient to prevent posterior collapse by keeping the expectation of the KL’s distribution positive. |
| Outcome: | The proposed approach can avoid posterior collapse effectively and efficiently without introducing any new model component or modifying the objective. |
Non-Autoregressive Semantic Parsing for Compositional Task-Oriented Dialog (2021.naacl-main)
Copied to clipboard
| Challenge: | Semantic parsing using sequence-to-sequence models is stymied by higher compute requirements and higher latency. |
| Approach: | They propose a non-autoregressive approach to predict semantic parse trees with an efficient seq2seq model architecture. |
| Outcome: | The proposed architecture achieves an 81% reduction in latency on TOP dataset and retains competitive performance over non-pretrained models on three different semantic parsing datasets. |
Using Pre-Trained Language Models for Producing Counter Narratives Against Hate Speech: a Comparative Study (2022.findings-acl)
Copied to clipboard
| Challenge: | Autoregressive models combined with stochastic decodings are the most promising for generating CNs with regard to an unseen target of hate. |
| Approach: | They propose to use pre-trained language models to generate counter-narratives in English by adding an automatic post-editing step to refine generated CNs. |
| Outcome: | The proposed pipeline could be used to generate counter-narratives in English using pre-trained language models and stochastic decoding mechanisms. |
ENGINE: Energy-Based Inference Networks for Non-Autoregressive Machine Translation (2020.acl-main)
Copied to clipboard
| Challenge: | a non-autoregressive machine translation model can minimize the autoregressive teacher's energy . engINE is an inference network trained to minimize the teacher' energy based on distilled corpora . |
| Approach: | They propose to train a non-autoregressive machine translation model to minimize autoregressive teacher energy by using an inference network instead of distilled corpora. |
| Outcome: | The proposed model achieves state-of-the-art non-autoregressive results on two datasets . the proposed model is trained to minimize the autoregressive teacher energy . |
MIO: A Foundation Model on Multimodal Tokens (2025.emnlp-main)
Copied to clipboard
Zekun Moore Wang, King Zhu, Chunpu Xu, Wangchunshu Zhou, Jiaheng Liu, Yibo Zhang, Jessie Wang, Ning Shi, Siyu Li, Yizhi Li, Haoran Que, Zhaoxiang Zhang, Yuanxing Zhang, Ge Zhang, Ke Xu, Jie Fu, Wenhao Huang
| Challenge: | Existing models lack multimodal understanding capabilities, resulting in closed-source model that does not support multimodal interleaved sequences. |
| Approach: | They propose a foundation model built on multimodal tokens capable of understanding and generating speech, text, images, and videos in an end-to-end, autoregressive manner. |
| Outcome: | The proposed model is able to understand speech, text, images, and videos in an end-to-end, autoregressive manner. |
Search-Oriented Conversational Query Editing (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing CQR models are not learned toward improving the downstream search performance . existing models generate the rewrite token-by-token from scratch . |
| Approach: | They propose a text editing-based CQR model tailored for conversational search . they propose rewrite tokens are selected from the dialogue in a non-autoregressive fashion . |
| Outcome: | The proposed model outperforms state-of-the-art models on three conversational search benchmarks while having low rewriting latency. |
MEGA: Multilingual Evaluation of Generative AI (2023.emnlp-main)
Copied to clipboard
Kabir Ahuja, Harshita Diddee, Rishav Hada, Millicent Ochieng, Krithika Ramesh, Prachi Jain, Akshay Nambi, Tanuja Ganu, Sameer Segal, Mohamed Ahmed, Kalika Bali, Sunayana Sitaram
| Challenge: | Large Large Models (LLMs) have shown impressive performance on many natural language processing tasks such as language understanding, reasoning, and language generation. |
| Approach: | They present a framework for evaluating generative LLMs in the multilingual setting and provide directions for future progress in the field. |
| Outcome: | The proposed framework evaluates generative models on 16 NLP datasets across 70 typologically diverse languages and compares them to state-of-the-art non-autoregressive models. |
Enabling Autoregressive Models to Fill In Masked Tokens (2026.findings-eacl)
Copied to clipboard
| Challenge: | Autoregressive (AR) and masked language modeling (MLM) models are incapable of mucked infilling, which is the ability to predict mangled tokens between past and future context. |
| Approach: | They propose a method that leverages the strengths of autoregressive and masked language modeling to achieve state-of-the-art mucked infilling performance. |
| Outcome: | The proposed approach outperforms existing methods on masked infilling tasks. |
When Only Time Will Tell: Interpreting How Transformers Process Local Ambiguities Through the Lens of Restart-Incrementality (2024.acl-long)
Copied to clipboard
| Challenge: | In incremental models, one interpretation is possible, but models that can revise can do so if the ambiguity is resolved. |
| Approach: | They propose an interpretable way to analyse incremental states in a bidirectional way . they propose to use a model that can update internal states to reflect the garden path effect . |
| Outcome: | The proposed model shows that it can perform revisions and recover if the label is incorrect. |
Autoregressive Affective Language Forecasting: A Self-Supervised Task (2020.coling-main)
Copied to clipboard
| Challenge: | Using natural language, we can model emotional language in time based on past changes of language. |
| Approach: | They propose a task of affective language forecasting to model emotional language in time based on past changes of language. |
| Outcome: | The proposed model outperforms existing models on a Twitter dataset of 1,900 users and scores for 6 emotions and 2 additional linguistic attributes. |
Learning to Rewrite for Non-Autoregressive Neural Machine Translation (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive neural machine translations have poor inference speed but weak recognition of erroneous translation pieces. |
| Approach: | They propose an architecture to explicitly learn to rewrite the erroneous translation pieces. |
| Outcome: | The proposed architecture can achieve better performance while significantly reducing decoding time. |
Meta-Learning Online Adaptation of Language Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Large language models encode impressively broad world knowledge in their parameters, but the knowledge in static models becomes stale. |
| Approach: | They propose to use a small, autoregressive model to reweight the language modeling loss for each token during online fine-tuning to maximize the out-of-date model’s ability to answer questions about a document. |
| Outcome: | The proposed approach improves the model's ability to answer questions on streams of thousands of documents compared with fine-tuning and baseline heuristics. |
Diffusion Glancing Transformer for Parallel Sequence-to-Sequence Learning (2024.naacl-long)
Copied to clipboard
| Challenge: | Experimental results show that non-autoregressive generation models are superior in generation efficiency but inferior in generation quality. |
| Approach: | They propose a diffusion glancing transformer which employs a modality diffusion process and residual glancy sampling to improve multi-modality modeling. |
| Outcome: | The proposed model outperforms autoregressive and non-autoregressive models on machine translation and text generation benchmarks. |
Non-Autoregressive Sentence Ordering (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing sentence ordering approaches only leverage unilateral dependencies during decoding and cannot fully explore the semantic dependency between sentences. |
| Approach: | They propose a non-autoregressive ordering network that explores bilateral dependencies between sentences and predicts sentences for each position in parallel. |
| Outcome: | The proposed model outperforms existing autoregressive sentence ordering approaches and yields competitive performance compared with the state-of-the-arts. |
Learning to Recover from Multi-Modality Errors for Non-Autoregressive Neural Machine Translation (2020.acl-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive neural machine translation models suffer from multi-modality problem . despite their autoregressivity, most NMT models suffer with slow decoding speed . |
| Approach: | They propose a semi-autoregressive model which generates a translation as a sequence of segments while each segment is predicted token-by-token. |
| Outcome: | The proposed model can achieve 4 times speedup while maintaining comparable performance. |
One Reference Is Not Enough: Diverse Distillation with Reference Selection for Non-Autoregressive Translation (2022.naacl-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive neural machine translation models suffer from multimodality problem . multi-modality is not solved by a teacher forcing algorithm, limiting model capability . |
| Approach: | They propose a method that generates multiple reference translations for each source sentence . they compare the NAT output with all references and select the one that best fits the simulated model . |
| Outcome: | The proposed method achieves 29.82 BLEU with only one decoding pass on WMT14 En-De . |
Leveraging Order-Free Tag Relations for Context-Aware Recommendation (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to tag recommendation neglect orderlessness and inter-dependency . Empirical results on Instagram and Stack Overflow show that our method is significantly superior to the previous approaches. |
| Approach: | They propose a sequence-oblivious generation method for tag recommendation . the next tag to be generated is independent of the order of the generated tags . they also propose regressive generation methods that take orderlessness into account . |
| Outcome: | Empirical results show that the proposed method is superior to previous approaches . the proposed system is based on two domains, Instagram and Stack Overflow . |
Insertion-based Tree Decoding (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing non-autoregressive decoders that are sub-linear can speed up inference for longer sequences. |
| Approach: | They propose a sub-linear nonautoregressive tree decoder that uses tree-based insertion operations to generate trees in sub-lines . they evaluate their approach on semantic parsing and compare it against strong baselines . |
| Outcome: | The proposed approach achieves competitive accuracies while reducing the number of decoding steps. |
Unlocking the Potential of Diffusion Language Models through Template Infilling (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods rely on prefix-based prompting, resulting in a lack of stability and a large computational time. |
| Approach: | They propose a conditioning methodology tailored for Diffusion Language Models that distributes structural anchors across the target response, establishing a global template before infilling masked segments. |
| Outcome: | The proposed method improves on mathematical reasoning, code generation, and trip planning benchmarks while maintaining speed and robustness. |
Retrieving Sequential Information for Non-Autoregressive Neural Machine Translation (P19-1)
Copied to clipboard
| Challenge: | Experimental results show that the Reinforce-NAT system surpasses the baseline NAT system by a significant margin on BLEU without decelerating the decoding speed. |
| Approach: | They propose a sequence-level training method and a Transformer decoder to fuse the target sequential information into the top layer of the decoded Transformer. |
| Outcome: | The proposed model surpasses the baseline NAT system on BLEU without decelerating the decoding speed and achieves comparable translation performance to the autoregressive Transformer model with considerable speedup. |
A Set Prediction Network For Extractive Summarization (2023.findings-acl)
Copied to clipboard
| Challenge: | Recent approaches to extracting salient sentences from source document are naive and lack dependencies between sentences. |
| Approach: | They propose a set prediction network to detect redundancy relationship between sentences . they use a non-autoregressive decoder to predict sentences in parallel . |
| Outcome: | The proposed method outperforms previous state-of-the-art models on extracted summary datasets. |
On the Role of Bidirectionality in Language Model Pre-Training (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Prior work on language model pre-training explored different architectures and learning objectives, but differences in data, hyperparameters and evaluation make a principled comparison difficult. |
| Approach: | They propose a framework that generalizes prior approaches to pre-training language models by focusing on bidirectionality and controlling each of them separately. |
| Outcome: | The proposed framework generalizes prior approaches including fully unidirectional models like GPT, fully bidirectional models and hybrid models like CM3 and prefix LM. |
Empirical Analysis of Decoding Biases in Masked Diffusion Models (2026.acl-long)
Copied to clipboard
Pengcheng Huang, Tianming Liu, Zhenghao Liu, Yukun Yan, Shuo Wang, Tong Xiao, Zulong Chen, Maosong Sun
| Challenge: | Existing MDMs employ uncertainty-based decoding strategies that limit their reasoning ability and ultimately degrade generation quality. |
| Approach: | They propose a framework that regularizes uncertainty-based decoding by incorporating two complementary priors to shape global decoding trajectories and promote content informativeness. |
| Outcome: | The proposed framework outperforms existing decoding strategies by more than 7% while achieving comparable performance to autoregressive models of similar parameter scales. |
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. |
Sequence-to-Lattice Models for Fast Translation (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing non-autoregressive machine translation models have decoders that are difficult to port to NAT models. |
| Approach: | They propose a sequence-to-lattice model that replaces the decoder with a search lattice. |
| Outcome: | The proposed model is faster than past non-autoregressive generation approaches and more accurate than reducing the number of decoder layers. |
BasqBBQ: A QA Benchmark for Assessing Social Biases in LLMs for Basque, a Low-Resource Language (2025.coling-main)
Copied to clipboard
| Challenge: | Existing pre-trained language models can propagate social biases in under-resourced languages like Basque. |
| Approach: | They propose a benchmark to assess biases in Basque using a multiple-choice question-answering task. |
| Outcome: | The proposed dataset is the first to assess biases in Basque across eight domains . larger models achieve better accuracy, but ambiguous cases remain challenging . |
Viterbi Decoding of Directed Acyclic Transformer for Non-Autoregressive Machine Translation (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Non-autoregressive models lack the ability to capture sequential dependency . Existing approaches to model sequential dependency have to apply a sequential decision process at inference time . |
| Approach: | They propose a Viterbi decoding framework to capture sequential dependency . they propose to find the optimal translation path under any length constraint . |
| Outcome: | The proposed framework improves the performance of DA-Transformer while maintaining similar speedup. |
One Token Is Enough: Improving Diffusion Language Models with a Sink Token (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing Diffusion Language Models lack a structural constraint to stabilize attention sinks. |
| Approach: | They propose a simple but effective extra sink token that is constrained to attend to itself while remaining globally visible to all other tokens. |
| Outcome: | The proposed token is able to stabilize attention sinks and improve model performance. |
Tree-Structured Non-Autoregressive Decoding for Sequence-to-Sequence Text Generation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Autoregressive Transformers suffer from high inference latency due to sequential token generation. |
| Approach: | They propose a tree-structured non-autoregressive decoding paradigm that bridges autoregressive and non-automatic decoding. |
| Outcome: | The proposed paradigm outperforms autoregressive and non-autoregressive decoding in machine translation and paraphrase generation. |
A Non-Autoregressive Edit-Based Approach to Controllable Text Simplification (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing models that generate generic simplified outputs for a given source text have been used to specify output properties. |
| Approach: | They propose a non-autoregressive model that iteratively edits an input sequence and incorporates lexical complexity information into the refinement process to generate simplifications that better match the desired output complexity. |
| Outcome: | The proposed model incorporates lexical complexity information into the refinement process to achieve more complex simplification operations such as content deletion and paraphrasing, as well as sentence splitting. |
FloorPlan-LLaMa: Aligning Architects’ Feedback and Domain Knowledge in Architectural Floor Plan Generation (2025.acl-long)
Copied to clipboard
| Challenge: | Existing evaluation methods for floor plan generation rely on statistical metrics like FID, GED, and PSNR, which fail to evaluate using domain knowledge. |
| Approach: | They propose to use a first floor plan dataset to train a floor plan generation model based on a multi-dimensional preference score and a textual analysis to integrate architects’ professional expertise and preferences. |
| Outcome: | The proposed model outperforms baseline models in text-conditional and class-condition tasks and is more rational and aligns better with human preferences. |
TrigReason: Trigger-Based Collaboration between Small and Large Reasoning Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Reasoning Models suffer from high inference latency due to autoregressive reasoning . SpecReason adopts a polling-based design that repeatedly invokes the LRM for verification at every step . |
| Approach: | They propose a trigger-based collaborative reasoning framework that delegates most reasoning to the SRM and activates LRM intervention only when necessary. |
| Outcome: | The proposed framework reduces latency and API cost by 73.3% under edge–cloud conditions. |
End-to-End Non-Autoregressive Neural Machine Translation with Connectionist Temporal Classification (D18-1)
Copied to clipboard
| Challenge: | Autoregressive decoding is the only part of sequence-to-sequence models that prevents massive parallelization at inference time. |
| Approach: | They propose a non-autoregressive architecture based on connectionist temporal classification . they conduct experiments on the WMT English-Romanian and English-German datasets . |
| Outcome: | The proposed model achieves a significant speedup over autoregressive models . the model can be trained end-to-end and maintains translation quality comparable to other models compared to autoregression models based on connectionist temporal classification . |
Context-Aware Non-Autoregressive Document-Level Translation with Sentence-Aligned Connectionist Temporal Classification (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing studies employ autoregressive translation (AT) methods to encode sentences . however, the AT methods struggle with error accumulation when the length of sentences increases. |
| Approach: | They propose a context-aware non-autoregressive framework with the sentence-aligned connectionist temporal classification loss for document-level neural machine translation. |
| Outcome: | The proposed framework achieves 46X speedup on three benchmarks compared to strong baselines. |
Autoregressive Text Generation Beyond Feedback Loops (D19-1)
Copied to clipboard
| Challenge: | Autoregressive feedback exposes the evolution of the hidden state trajectory to potential biases from well-known train-test discrepancies. |
| Approach: | They combine a latent state space model with a CRF observation model to investigate the state evolution of a hidden state trajectory. |
| Outcome: | The proposed model performs better on unconditional sentence generation compared to baselines while avoiding some prototypical failure modes. |
Multi-Granularity Optimization for Non-Autoregressive Translation (2022.emnlp-main)
Copied to clipboard
| Challenge: | Non-autoregressive machine translation suffers severe performance deterioration due to the naive independence assumption. |
| Approach: | They propose a method which collects model behaviours on translation segments of various granularities and integrates feedback for backpropagation to reduce latency. |
| Outcome: | Experiments on four benchmark datasets show that the proposed method outperforms baseline models trained with cross-entropy loss and achieves the best performance on WMT’16 EnRo and highly competitive results on WTM’14 EnDe. |
Alignment-Augmented Speculative Decoding with Alignment Sampling and Conditional Verification (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to accelerate autoregressive generation of large language models require training costs. |
| Approach: | They propose a training-free alignment-augmented speculative decoding algorithm . it leverages the output distribution obtained in the prefilling phase to provide more aligned draft candidates . |
| Outcome: | The proposed method increases the average generation score by 3.3 points for the LLaMA3 model. |
Finding the Optimal Vocabulary Size for Neural Machine Translation (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Class imbalance is said to exist when one or more classes are not of approximately equal frequency in data. |
| Approach: | They cast neural machine translation (NMT) as a classification task in an autoregressive setting and examine its limitations. |
| Outcome: | The proposed model performs better on multiple languages with large data sizes with different vocabulary sizes. |
Discrete Diffusion Language Model for Efficient Text Summarization (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing discrete diffusion models fail on conditional long-text generation due to incompatibility between the backbone architectures and the random noising process. |
| Approach: | They propose a semantic-aware noising process that enables Transformer backbones to handle long sequences effectively. |
| Outcome: | The proposed model outperforms existing models on three benchmark summarization datasets while achieving much faster inference speed compared to autoregressive models. |
Parallelism and Generation Order in Masked Diffusion Language Models: Limits Today, Potential Tomorrow (2026.findings-acl)
Copied to clipboard
Yangyang Zhong, Yanmei Gu, Zhengqing Zang, Xiaomeng Li, Yuqi Ding, Xibei Jia, Yuting Shen, Zhenzhong Lan, Liwang Zhu, Weiping Liu, Junlin Zhou, Haisheng Liu, Zhong Xin Yu, Pengxin Luo, Donglian Qi, Yunfeng Yan, Junbo Zhao
| Challenge: | Autoregressive (AR) language models dominate modern natural language processing due to strong likelihood-based training objectives and reliable left-to-right decoding. |
| Approach: | They characterize MDLM behavior along two dimensions: parallelism strength and generation order . authors propose a Generate-then-Edit paradigm that mitigates dependency loss . |
| Outcome: | The proposed model improves on tasks that require "backward information" the Generate-then-Edit paradigm improves parallel decoding efficiency while reducing dependency loss. |
An Exploration of Encoder-Decoder Approaches to Multi-Label Classification for Legal and Biomedical Text (2023.findings-acl)
Copied to clipboard
| Challenge: | Standard methods for multi-label text classification rely on encoder-only pre-trained models . encoder decoder models have proven more effective in other classification tasks . |
| Approach: | They compare four methods for multi-label classification based on encoder-only models . they use a pre-trained model for multilabel text classification . |
| Outcome: | The proposed methods outperform encoder-only methods on complex datasets and labeling schemes. |
FastCorrect 2: Fast Error Correction on Multiple Candidates for Automatic Speech Recognition (2021.findings-emnlp)
Copied to clipboard
Yichong Leng, Xu Tan, Rui Wang, Linchen Zhu, Jin Xu, Wenjie Liu, Linquan Liu, Xiang-Yang Li, Tao Qin, Edward Lin, Tie-Yan Liu
| Challenge: | Error correction is widely used in automatic speech recognition (ASR) to post-process the generated sentence. |
| Approach: | They propose a fast correction model that takes multiple ASR candidates as input for better correction accuracy. |
| Outcome: | The proposed model can reduce the word error rate (WER) with multiple candidates by 3.2% and 2.6%. |
AGRec: Adapting Autoregressive Decoders with Graph Reasoning for LLM-based Sequential Recommendation (2025.findings-acl)
Copied to clipboard
| Challenge: | Autoregressive decoders in large language models excel at capturing sequential behaviors for generative recommendations, but they lack graph-structured user-item interactions, which are widely recognized as beneficial. |
| Approach: | They propose a novel algorithm that adapts LLMs’ decoders with graph reasoning for recommendation by augmenting the decoding logits with an auxiliary GNN model to optimize token generation. |
| Outcome: | The proposed model outperforms state-of-the-art models in sequential recommendations. |
Helping the Weak Makes You Strong: Simple Multi-Task Learning Improves Non-Autoregressive Translators (2022.emnlp-main)
Copied to clipboard
| Challenge: | Non-autoregressive (NAR) neural machine translation models require a conditional independence assumption on target sequences, resulting in less informative learning signals. |
| Approach: | They propose a model-agnostic multi-task learning framework to provide more informative learning signals for NAR models under conventional MLE training. |
| Outcome: | The proposed framework improves accuracy of multiple NAR baselines without additional decoding overhead. |
Lessons Learned from GPT-SW3: Building the First Large-Scale Generative Language Model for Swedish (2022.lrec-1)
Copied to clipboard
Ariel Ekgren, Amaru Cuba Gyllensten, Evangelia Gogoulou, Alice Heiman, Severine Verlinden, Joey Öhman, Fredrik Carlsson, Magnus Sahlgren
| Challenge: | a prerequisite for building large-scale generative models for other languages is access to large amounts of high-quality text data and powerful computational resources. |
| Approach: | They present a 3.5 billion parameter autoregressive language model, trained on a 100 GB Swedish corpus. |
| Outcome: | The proposed model performs well on a 100 GB Swedish corpus and is competent in comparison with existing models of similar size. |
A Bounding Box is Worth One Token - Interleaving Layout and Text in a Large Language Model for Document Understanding (2025.findings-acl)
Copied to clipboard
Jinghui Lu, Haiyang Yu, Yanjie Wang, Yongjie Ye, Jingqun Tang, Ziwei Yang, Binghong Wu, Qi Liu, Hao Feng, Han Wang, Hao Liu, Can Huang
| Challenge: | Existing methods for integrating spatial layouts with text have limitations . existing methods produce overly long text sequences or lack autoregressive traits of LLMs . |
| Approach: | They introduce Interleaving Layout and Text in a Large Language Model (LayTextLLM) they use OCR-derived text and spatial layouts to integrate with LLMs for document understanding . |
| Outcome: | The proposed model shows an increase in performance in KIE and VQA tasks. |
Too Big to Fail: Larger Language Models are Disproportionately Resilient to Induction of Dementia-Related Linguistic Anomalies (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing studies show that the attention mechanism in transformer-based NLMs may present an analogue to the notions of cognitive and brain reserve. |
| Approach: | They propose a bidirectional ablation method that masks attention heads to display degradation of similar magnitude to masking in smaller models. |
| Outcome: | The proposed method exhibits properties attributed to the concepts of cognitive and brain reserve in human brain studies. |
Prompt-Based Editing for Text Style Transfer (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Text style transfer is a type of textual prompt that generates style-transferred texts word by word . early prediction errors may affect future word predictions. |
| Approach: | They propose a prompt-based editing approach to text style transfer using a pretrained language model. |
| Outcome: | The proposed approach outperforms existing systems with 20 times more parameters on three style-transfer benchmark datasets. |
Tail-to-Tail Non-Autoregressive Sequence Prediction for Chinese Grammatical Error Correction (2021.acl-long)
Copied to clipboard
| Challenge: | Experimental results demonstrate the effectiveness of Tail-to-Tail (TtT) non-autoregressive sequence prediction for Chinese Grammatical Error Correction (CGEC) |
| Approach: | They propose a framework for Chinese Grammatical Error Correction (CGEC) that uses a BERT-initialized Transformer Encoder to model the error positions. |
| Outcome: | The proposed framework solves the problem of Chinese Grammatical Error Correction (CGEC) by modeling the token dependencies. |
Data-Informed Global Sparseness in Attention Mechanisms for Deep Neural Networks (2024.lrec-main)
Copied to clipboard
| Challenge: | Attention pruning techniques have been developed to identify and exploit sparseness . previous work has taken pioneering steps to discover and explain the sparsity in attention patterns . |
| Approach: | They propose a framework that observes attention patterns in a fixed dataset and generates a global sparseness mask. |
| Outcome: | The proposed approach saves 90% of computations and maintains quality of results. |
What’s under the hood: Investigating Automatic Metrics on Meeting Summarization (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing evaluation metrics do not capture meeting-specific errors, leading to ineffective assessment. |
| Approach: | They examine the relationship between established metrics and human evaluations to determine what challenges and errors are captured by correlating metric scores with human evaluation. |
| Outcome: | The proposed measures show weak correlations with human evaluations and a third of the correlations show error masking. |
DC-MBR: Distributional Cooling for Minimum Bayesian Risk Decoding (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing methods for decoding target language are degenerate, hallucinating or empty. |
| Approach: | They propose a method that tunes down the Softmax temperature to reduce autoregressive over-smoothness by label smoothing the output distributions. |
| Outcome: | The proposed method improves MBR in various settings. |
Beyond Error Propagation in Neural Machine Translation: Characteristics of Language Also Matter (D18-1)
Copied to clipboard
| Challenge: | Neural machine translation suffers from exposure bias and error propagation problem. |
| Approach: | They conduct a series of analyses to deeply understand the accuracy drop problem . they find that the left part of the translated sentence is often better than its right part . |
| Outcome: | The results show that the left part of the translated sentence is often better than its right part in left-to-right decoding models. |
Learning from Imperfect Data: Towards Efficient Knowledge Distillation of Autoregressive Language Models for Text-to-SQL (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing text-to-SQL LLMs are computationally expensive and difficult to deploy in real-world applications. |
| Approach: | They propose to distill a larger teacher model into a smaller student model by using imperfect data to improve the KD. |
| Outcome: | The proposed method achieves the best tradeoff between performance and efficiency on 5 text-to-SQL benchmarks. |
Revealing the Attention Floating Mechanism in Masked Diffusion Models (2026.findings-acl)
Copied to clipboard
Xin Dai, Pengcheng Huang, Zhenghao Liu, Shuo Wang, Yukun Yan, Chaojun Xiao, Yu Gu, Ge Yu, Maosong Sun
| Challenge: | Masked diffusion models (MDMs) leverage bidirectional attention and a denoising process. |
| Approach: | They investigate the attention behaviors of Masked diffusion models by revealing the phenomenon of Attention Floating. |
| Outcome: | The proposed model doubles the performance of autoregressive models in knowledge-intensive tasks. |
GeoCoder: Solving Geometry Problems by Generating Modular Code through Vision-Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Various vision-language models (VLMs) have made significant progress in multimodal tasks, but they still struggle with geometry problems. |
| Approach: | They propose a vision-language model that leverages modular code-finetuning to generate and execute code using a predefined geometry function library. |
| Outcome: | The proposed model improves geometric reasoning abilities by 16% on a GeomVerse dataset compared to other methods. |
MARS: Unleashing the Power of Speculative Decoding via Margin-Aware Verification (2026.findings-acl)
Copied to clipboard
Jingwei Song, Xinyu Wang, Hanbin Wang, Xiaoxuan Lei, Tianyu Shi, Shixin Han, Eric Yang, Xiao-Wen Chang, Lynn Ai
| Challenge: | Autoregressive large language models suffer from high inference latency due to memorybandwidth constraints. |
| Approach: | They propose a method that decouples generation and verification by decoupling tokens and a lightweight draft model. |
| Outcome: | The proposed method delivers consistent and significant speedups over state-of-the-art baselines while preserving generation quality across diverse benchmarks. |
Neighbors Are Not Strangers: Improving Non-Autoregressive Translation under Low-Frequency Lexical Constraints (2022.naacl-main)
Copied to clipboard
| Challenge: | Existing approaches to lexically constrained neural machine translation suffer from high latency. |
| Approach: | They propose a plug-in algorithm for non-autoregressive translation for this problem . they propose ACT to familiarize the model with the source-side context of constraints . |
| Outcome: | The proposed model improves over the backbone constrained NAT model in constraint preservation and translation quality, especially for rare constraints. |
Sketch and Refine: Towards Faithful and Informative Table-to-Text Generation (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for table-to-text generation suffer from poor faithfulness and low coverage. |
| Approach: | They propose a method that combines Autoregressive and Non-Autoregressive generation to generate a table-to-text from a key-value table using a skeleton and an edit-based non-autoregressively generation model. |
| Outcome: | The proposed method outperforms the existing methods on WikiPerson and WikiBio datasets on coverage and faithfulness. |
TILGAN: Transformer-based Implicit Latent GAN for Diverse and Coherent Text Generation (2021.findings-acl)
Copied to clipboard
| Challenge: | Existing autoregressive models suffer from the exposure bias problem due to mismatches between training and generation stages. |
| Approach: | They propose a Transformerbased Implicit Latent GAN which combines a transformer autoencoder and GAN in the latent space with a novel design and distribution matching based on the Kullback-Leibler divergence. |
| Outcome: | The proposed model improves local and global coherence and quality-diversity trade-off on three benchmark datasets. |
Multilingual Nonce Dependency Treebanks: Understanding how Language Models Represent and Process Syntactic Structure (2024.naacl-long)
Copied to clipboard
| Challenge: | a number of studies have focused on making explicit the linguistic information encoded in language models (LMs) however, this method has been criticized for various reasons. |
| Approach: | They introduce a framework for creating nonce treebanks for multilingual UD corpora . they investigate word co-occurrence statistics and show how nonce data affects the performance of syntactic dependency probes. |
| Outcome: | The proposed framework satisfies syntactic argument structure and ensures grammaticality via language-specific rules. |
Non-autoregressive Text Editing with Copy-aware Latent Alignments (2023.emnlp-main)
Copied to clipboard
| Challenge: | Seq2Edit approaches still face several challenges such as inflexibility in generation and difficulty in generalizing to other languages. |
| Approach: | They propose a non-autoregressive text editing method that models the edit process with latent CTC alignments and introduces the copy operation into the edit space. |
| Outcome: | The proposed method outperforms existing Seq2Edit models and achieves similar or even better results than Seq1Edit with over 4 speedup. |
VELA: An LLM-Hybrid-as-a-Judge Approach for Evaluating Long Image Captions (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing evaluation metrics for image captioning are primarily designed for short captions and are not suitable for long captions. |
| Approach: | They propose an automatic evaluation metric for long captions developed within a novel LLM-Hybrid-as-a-Judge framework. |
| Outcome: | The proposed metric outperforms existing metrics and achieves superhuman performance on LongCap-Arena. |
“My Answer is C”: First-Token Probabilities Do Not Match Text Answers in Instruction-Tuned Language Models (2024.findings-acl)
Copied to clipboard
Xinpeng Wang, Bolei Ma, Chengzhi Hu, Leon Weber-Genzel, Paul Röttger, Frauke Kreuter, Dirk Hovy, Barbara Plank
| Challenge: | Multiple choice questions are one of the most popular evaluation formats for understanding the capabilities of autoregressive large language models (LLMs). |
| Approach: | They evaluated how aligned first-token evaluation is with the text output along several dimensions, namely final option choice, refusal rate, choice distribution and robustness under prompt perturbation. |
| Outcome: | The proposed evaluation methods are misaligned on all dimensions, reaching mismatch rates over 60%. |
Amphista: Bi-directional Multi-head Decoding for Accelerating LLM Inference (2025.naacl-long)
Copied to clipboard
Zeping Li, Xinlong Yang, Ziheng Gao, Ji Liu, Guanchen Li, Zhuang Liu, Dong Li, Jinzhang Peng, Lu Tian, Emad Barsoum
| Challenge: | Existing methods such as Medusa lack adequate information interaction between different drafting heads. |
| Approach: | They propose an enhanced speculative decoding framework that builds upon Medusa and integrates a drafting block capable of parallel inference. |
| Outcome: | The proposed framework outperforms Medusa in terms of head accuracy and latency. |
Generative Aspect-Based Sentiment Analysis with Contrastive Learning and Expressive Structure (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods struggle with implicit aspects and opinions, such as "it took an hour to be seated". |
| Approach: | They propose two techniques for improved structured generation for ACOS quadruple extraction. |
| Outcome: | The proposed model improves on three ACOS datasets and shows significant improvement on implicit aspect and opinion splits. |
What Have We Achieved on Non-autoregressive Translation? (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have shown that non-autoregressive (NAT) methods underperform autoregressive methods (AT) however, their evaluation using BLEU has been shown to weakly correlate with human annotations. |
| Approach: | They propose to evaluate four representative NAT methods using BLEU to narrow the performance gap between autoregressive and autoregressive translations. |
| Outcome: | The proposed methods underperform NAT and autoregressive methods under more reliable evaluation metrics. |
Non-Autoregressive Translation by Learning Target Categorical Codes (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive text generation models still fall behind in translation quality . authors propose a model that learns implicitly categorical codes as latent variables . |
| Approach: | They propose a non-autoregressive Transformer model that implicitly categorizes latent variables into decoding . they find it improves translation quality by introducing more informative decoder inputs . |
| Outcome: | The proposed model achieves comparable or better performance in machine translation tasks than strong baselines. |
Con-NAT: Contrastive Non-autoregressive Neural Machine Translation (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Neural machine translation models are autoregressive, which means they predict tokens one by one based on source tokens and previously predicted tokens. |
| Approach: | They propose a conditional masked language model which incorporates contrastive learning into the conditional language model. |
| Outcome: | The proposed model improves on WMT’16 Ro-En translation directions with different data sizes. |
Inference Strategies for Machine Translation with Conditional Masking (2020.emnlp-main)
Copied to clipboard
| Challenge: | Conditional masked language model training has proven successful for non-autoregressive and semi-auto-regressively sequence generation tasks. |
| Approach: | They propose a conditional masked language model (CMLM) that is a factorization of conditional probabilities of partial sequences and propose heuristics to improve performance. |
| Outcome: | The proposed algorithm is more efficient than the standard “mask-predict” algorithm on machine translation tasks. |
TOP-Training: Target-Oriented Pretraining for Medical Extractive Question Answering (2025.coling-main)
Copied to clipboard
| Challenge: | e-health records underscore the growing significance of information extraction (IE) from these datasets. |
| Approach: | They propose a target-oriented pre-training paradigm for extractive question-answering in the medical domain . TOP-Training moves one step further than popular domain-oriented fine-tuning . |
| Outcome: | The proposed method improves on the Medical-EQA benchmarks. |
Don’t Parse, Choose Spans! Continuous and Discontinuous Constituency Parsing via Autoregressive Span Selection (2023.acl-long)
Copied to clipboard
| Challenge: | Constituency parsing is a fundamental task in natural language processing, having many applications in downstream tasks such as language modeling. |
| Approach: | They propose a simple and unified approach for both continuous and discontinuous constituency parsing via autoregressive span selection. |
| Outcome: | The proposed model can predict all possible continuous and discontinuous constituency trees without sacrificing data coverage and without expensive chart-based parsing algorithms. |
HYDRA: A Multi-Head Encoder-only Architecture for Hierarchical Text Classification (2025.emnlp-main)
Copied to clipboard
| Challenge: | State-of-the-art approaches rely on complex components like graph encoders, label semantics, and autoregressive decoders. |
| Approach: | They propose a multi-head encoder-only architecture for hierarchical text classification that treats each level as a separate classification task with its own label space. |
| Outcome: | The proposed architecture matches or exceeds state-of-the-art methods on four benchmarks. |
From Past To Path: Masked History Learning for Next-Item Prediction in Generative Recommendation (2026.acl-long)
Copied to clipboard
Kaiwen Wei, Kejun he, Xiaomian Kang, Jie Zhang, null Ymyang, Li Jin, Zhenyang Li, Jiang Zhong, Richard He Bai, Junnan Zhu
| Challenge: | Generative recommendation models inherently bias towards local contexts, failing to capture deeper historical dependencies necessary for understanding complex user intents. |
| Approach: | They propose a training framework that shifts the objective from simple next-step prediction to deep comprehension of history by entropy-guided masking policy and a curriculum learning scheduler to enhance the framework. |
| Outcome: | The proposed framework outperforms state-of-the-art generative models on three public datasets and shows that it is more accurate than current models. |
Shall We Pretrain Autoregressive Language Models with Retrieval? A Comprehensive Study (2023.emnlp-main)
Copied to clipboard
Boxin Wang, Wei Ping, Peng Xu, Lawrence McAfee, Zihan Liu, Mohammad Shoeybi, Yi Dong, Oleksii Kuchaiev, Bo Li, Chaowei Xiao, Anima Anandkumar, Bryan Catanzaro
| Challenge: | a recent study shows that retrieval-augmented LMs can improve text generation quality and accuracy. |
| Approach: | They propose a model that reproduces RETRO parameters while retrieving a text corpus . they find RETRO outperforms GPT on text generation with less repetition . |
| Outcome: | The proposed model outperforms standard retrieval-augmented GPT and retrieval augmented GTP on text generation and accuracy tasks. |
Jakiro: Boosting Speculative Decoding via Decoupled MoE (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods to accelerate large language model inference have a fundamental limitation: candidates at the same tree layer share identical feature representations, constraining diversity and diminishing overall effectiveness. |
| Approach: | They propose a decoupled mixture of experts (MoE) into a draft model to generate diverse tokens from distinct feature spaces. |
| Outcome: | The proposed approach achieves significant speedups over strong baselines, with notable improvements in non-greedy scenarios where token diversity is crucial. |
An Efficient Dialogue Policy Agent with Model-Based Causal Reinforcement Learning (2025.coling-main)
Copied to clipboard
| Challenge: | Existing models for dialogue policy training consider one-step dialogues, leading to inaccurate simulations. |
| Approach: | They propose a framework for dialogue policy learning that trains an agent to select dialogue actions via deep reinforcement learning. |
| Outcome: | The proposed framework achieves state-of-the-art performance on three dialogue datasets . it uses model-based reinforcement learning with automatically constructed causal chains . |
Autoregressive Knowledge Distillation through Imitation Learning (2020.emnlp-main)
Copied to clipboard
| Challenge: | Autoregressive models are ubiquitous in natural language processing due to the sequential nature of text generation. |
| Approach: | They propose a compression technique for autoregressive models driven by an imitation learning perspective on knowledge distillation. |
| Outcome: | The proposed method outperforms other distillation algorithms on translation and summarization tasks while increasing inference speed 14 times. |
Just Pass Twice: Efficient Token Classification with LLMs for Zero-Shot NER (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods to name entity recognition use autoregressive decoding, hallucinated entities, and formatting errors. |
| Approach: | They propose a method that allows causal LLMs to perform discriminative token classification with full bidirectional context. |
| Outcome: | The proposed method surpasses the previous best method on zero-shot NER benchmarks by +7.9 F1 on average across CrossNER and MIT benchmarks. |
MultiVerse: Efficient and Expressive Zero-Shot Multi-Task Text-to-Speech (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Text-to-speech systems that scale up the amount of training data have certain limitations: they require a large amount of data, which increases costs, and overlook prosody similarity. |
| Approach: | They propose a zero-shot multi-task TTS system that can perform TTS or speech style transfer in zero- shot and cross-lingual conditions. |
| Outcome: | The proposed system outperforms other TTS systems trained with the same small amount of data and achieves zero-shot performance comparable to data-driven systems. |
SC2: Towards Enhancing Content Preservation and Style Consistency in Long Text Style Transfer (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods for short TST are difficult to implement and can cause content degradation. |
| Approach: | They propose a method to vary the style polarity of text while preserving semantic content. |
| Outcome: | The proposed method improves over baselines and is highly efficient. |
Improving Scheduled Sampling with Elastic Weight Consolidation for Neural Machine Translation (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Autoregressive models trained with maximum likelihood estimation suffer from exposure bias, i.e. the discrepancy between ground-truth prefixes used during training and model-generated prefix at inference time. |
| Approach: | They propose to use Elastic Weight Consolidation to better balance mitigating exposure bias with retaining performance. |
| Outcome: | The proposed method significantly outperforms maximum likelihood estimation and scheduled sampling baselines on four translation datasets. |
Recurrent Event Network: Autoregressive Structure Inferenceover Temporal Knowledge Graphs (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for reasoning over temporal knowledge graphs focus on past timestamps and are not able to predict future interactions. |
| Approach: | They propose a novel autoregressive architecture for predicting future interactions using a recurrent event encoder and a neighborhood aggregator. |
| Outcome: | The proposed method achieves state-of-the-art on five public datasets. |
CTC-based Non-autoregressive Textless Speech-to-Speech Translation (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing direct speech-to-speech translation models require text supervision during training, which is not feasible for numerous unwritten languages. |
| Approach: | They propose a non-autoregressive (NAR) model that generates discrete units from the source speech and employs a unit-based vocoder to synthesize the target. |
| Outcome: | The proposed model achieves translation quality comparable to the autoregressive model while preserving up to 26.81 decoding speedup. |
Mirror: A Universal Framework for Various Information Extraction Tasks (2023.emnlp-main)
Copied to clipboard
Tong Zhu, Junfei Ren, Zijian Yu, Mengsong Wu, Guoliang Zhang, Xiaoye Qu, Wenliang Chen, Zhefeng Wang, Baoxing Huai, Min Zhang
| Challenge: | Recent studies often formulate IE tasks as a triplet extraction problem, but this paradigm does not support multi-span and n-ary extraction, leading to weak versatility. |
| Approach: | They propose a multi-span cyclic graph extraction problem and a non-autoregressive graph decoding algorithm to extract all spans in a single step. |
| Outcome: | The proposed model outperforms or reaches competitive performance with SOTA systems under few-shot and zero-shot settings and it is compatible with 57 datasets. |
Bilingual Synchronization: Restoring Translational Relationships with Editing Operations (2022.emnlp-main)
Copied to clipboard
| Challenge: | Neural Machine Translation (NMT) is a one-shot process that generates the target language equivalent of some source text from scratch. |
| Approach: | They propose a machine translation task which assumes an initial target sequence, that must be transformed into a valid translation of the source. |
| Outcome: | The proposed system outperforms other systems trained for similar tasks. |
OWSM-CTC: An Open Encoder-Only Speech Foundation Model for Speech Recognition, Translation, and Language Identification (2024.acl-long)
Copied to clipboard
| Challenge: | Autoregressive models can be slower during inference and have potential risks of hallucination. |
| Approach: | They propose an encoder-only speech foundation model based on Connectionist Temporal Classification. |
| Outcome: | The proposed model improves on 180k hours of public audio data for multilingual speech recognition, speech translation, and language identification. |
JANUS: Joint Autoregressive and Non-autoregressive Training with Auxiliary Loss for Sequence Generation (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to train autoregressive and non-autoregressive models only consider relevance of model parameters, ignoring correlations between the two manners. |
| Approach: | They propose a joint autoregressive and non-autoregressive training method using aUxiliary losS to enhance the model performance in both AR and NAR manners simultaneously. |
| Outcome: | The proposed method improves the model performance in both AR and NAR manners and reduces the inference latency. |
Diffusion-CAM: Faithful Visual Explanations for dMLLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Existing Class Activation Mapping methods are ill-suited for interpreting non-autoregressive behaviors of diffusion-based architectures. |
| Approach: | They propose to use a method to generate parallel activation maps by probing intermediate representations in the transformer backbone to capture latent features and their class-specific gradients. |
| Outcome: | Experiments show that Diffusion-CAM significantly outperforms SoTA methods in localization accuracy and visual fidelity. |
DP-MLM: Differentially Private Text Rewriting Using Masked Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for text privatization using Differential Privacy rely on autoregressive models which lack a mechanism to contextualize the private rewriting process. |
| Approach: | They propose a method for differentially private text rewriting using masked language models to rewrite a text one token at a time. |
| Outcome: | The proposed method preserves utility at lower levels, compared to previous methods relying on autoregressive models with a decoder. |
Focus-dLLM: Accelerating Long-Context Diffusion LLM Inference via Confidence-Guided Context Focusing (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for estimating attention importance for tokens are ineffective . dLLMs require bidirectional attention, which limits inference efficiency . |
| Approach: | They propose a training-free attention sparsification framework for efficient long-context inference . they propose 'sink-aware pruning strategy' to accurately estimate and remove redundant computation . |
| Outcome: | The proposed approach offers 29 lossless speedup under 32K context length. |
PDR: A Plug-and-Play Positional Decay Framework for LLM Pre-training Data Detection (2026.acl-long)
Copied to clipboard
| Challenge: | Existing likelihood-based methods for detecting pretraining data are limited in black-box, zero-shot settings. |
| Approach: | They propose a training-free and plug-and-play framework that reweights token-level scores to amplify distinct signals from early positions while suppressing noise from later ones. |
| Outcome: | The proposed framework amplifys signals from early positions while suppressing noise from later positions. |
Flexible Generation from Fragmentary Linguistic Input (2022.acl-long)
Copied to clipboard
| Challenge: | dominant paradigm for high-performance models in novel language tasks is direct specialization via training from scratch or fine-tuning large pre-trained models. |
| Approach: | They propose a new model that makes it possible to infer human behavior through basic computational motifs. |
| Outcome: | The proposed model outperforms direct-specialization models in three evaluations and performs comparable to human models. |
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. |
Quick Back-Translation for Unsupervised Machine Translation (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Unsupervised machine translation models are limited by the run-time of autoregressive inference during back-translation and lack of synthetic data efficiency. |
| Approach: | They propose a two-for-one improvement to Transformer back-translation: Quick Back-Translation (QBT). QBT re-purposes the encoder as a generative model, and uses encoder-generated sequences to train the decoder. |
| Outcome: | Experiments on various WMT benchmarks show that QBT dramatically outperforms standard back-translation only method in terms of training efficiency for comparable translation qualities. |
latent-GLAT: Glancing at Latent Variables for Parallel Text Generation (2022.acl-long)
Copied to clipboard
| Challenge: | Recent advances in text generation have limited applications due to multimodality problem. |
| Approach: | They propose a method which uses latent variables to capture word categorical information and invoke an advanced curriculum learning technique to overcome multi-modality problem. |
| Outcome: | The proposed method outperforms strong baselines without an autoregressive model, which further broadens the application scenarios of the parallel decoding paradigm. |
Revisiting Knowledge Distillation for Autoregressive Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Autoregressive language models (LMs) are expensive and memory intensive, preventing the development of industrial applications. |
| Approach: | They propose an adaptive teaching approach to improve the KD of autoregressive language models by distilling knowledge into a small student model. |
| Outcome: | The proposed method can achieve consistent and significant performance gains across all model types and sizes. |
ARXSA: A General Negative Feedback Control Theory in Vision-Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | a new approach to the self-attention mechanism is proposed for integrating data from multiple batches. |
| Approach: | They propose an autoregressive with exogenous inputs approach for the Transformer model . the proposed method transforms the Encoder block into a negative feedback predictive control system . |
| Outcome: | The proposed method is validated through comparative evaluations. |
Learning Contextual Retrieval for Robust Conversational Search (2025.emnlp-main)
Copied to clipboard
| Challenge: | Effective conversational search requires a deep understanding of user intent across multiple dialogue turns. |
| Approach: | They propose a novel LLM-based retriever that directly incorporates conversational context into the retrieval process. |
| Outcome: | The proposed method outperforms existing methods while incurring no additional inference overhead. |
Highly Parallel Autoregressive Entity Linking with Discriminative Correction (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to EL have been shown to be effective for both Entity Disambiguation and Entity Linking, but they suffer from high computational cost due to a complex (deep) decoder and the need for training on a large amount of data. |
| Approach: | They propose a method that parallelizes autoregressive linking across all potential mentions and relies on a shallow and efficient decoder. |
| Outcome: | The proposed method outperforms state-of-the-art approaches on the English dataset AIDA-CoNLL and is >70 times faster and more accurate than the previous generative method. |
Mask-Predict: Parallel Decoding of Conditional Masked Language Models (D19-1)
Copied to clipboard
| Challenge: | a masked language model is used to train a model to predict subsets of mangled words . a parallel decoding algorithm can be used to generate translations in a constant number of iterations. |
| Approach: | They propose a model and a parallel decoding algorithm which train a machine to predict any subset of target words . they introduce conditional masked language models (CMLMs) which are trained with a mangled language model objective . |
| Outcome: | The proposed model improves state-of-the-art performance levels for non-autoregressive and parallel decoding models by over 4 BLEU on average. |
How do autoregressive transformers solve full addition? (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large pre-trained language models have demonstrated impressive capabilities, but there is still much to learn about how they operate. |
| Approach: | They investigate the ability of the autoregressive transformer to perform basic addition operations by using causal analysis to find that a few different attention heads in the middle layers control the addition carry . they found that due to the lack of global focus on the sequence within these attention heads, the model struggles to handle long-sequence addition tasks. |
| Outcome: | The model performs basic addition tasks, but it still faces challenges with length generalization. |
SSD-LM: Semi-autoregressive Simplex-based Diffusion Language Model for Text Generation and Modular Control (2023.acl-long)
Copied to clipboard
| Challenge: | Existing diffusion models for continuous-valued domains have not been adopted for text data. |
| Approach: | They propose a diffusion-based language model with two key design choices . semi-autoregressive model generates blocks of text and allows local context updates . they evaluate it on unconstrained text generation benchmarks . |
| Outcome: | The proposed model outperforms autoregressive models on unconstrained text generation benchmarks on uncontrolled text generation. |
ITER: Iterative Transformer-based Entity Recognition and Relation Extraction (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent advances in NLP generate structured information in an autoregressive manner, causing low throughput . authors propose an efficient encoder-based relation extraction model that performs the task in three parallelizable steps. |
| Approach: | They propose an efficient encoder-based relation extraction model that performs the task in three parallelizable steps. |
| Outcome: | The proposed model achieves state-of-the-art on two datasets and is faster than existing models. |
Language Modeling, Lexical Translation, Reordering: The Training Process of NMT through the Lens of Classical SMT (2021.emnlp-main)
Copied to clipboard
| Challenge: | Statistical MT decomposes the translation task into distinct components that are learned separately. |
| Approach: | They show that neural machine translation models acquire different competences over the course of training . previous work shows how to improve some of the competences in NMT by using lexical translation probabilities, phrase memories, alignment information. |
| Outcome: | The proposed model improves translation quality and word-by-word translation, while learning complex reordering patterns. |
Mechanistic Interpretability of Emotion Inference in Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on large language models (LLMs) show promising capabilities in predicting human emotions from text. |
| Approach: | They investigate how autoregressive LLMs infer emotions by focusing on appraisal theory . they show that emotion representations are functionally localized to specific regions in the model . |
| Outcome: | The proposed model is functionally localized to specific regions in the model, and the results align with theoretical and intuitive expectations. |
PALM: Pre-training an Autoencoding&Autoregressive Language Model for Context-conditioned Generation (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing techniques for natural language understanding and generation use autoencoding and/or autoregressive objectives to train models. |
| Approach: | They propose a self-supervised pre-training scheme that pre-trains an autoencoding and autoregressive language model on a large unlabeled corpus for generating new text conditioned on context. |
| Outcome: | The proposed scheme achieves state-of-the-art results on a variety of language generation benchmarks covering generative question answering, abstractive summarization and conversational response generation. |
MRRL: Modifying the Reference via Reinforcement Learning for Non-Autoregressive Joint Multiple Intent Detection and Slot Filling (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing non-autoregressive models for multiple intent detection and slot filling have limited overall accuracy due to multi-modality problem and lack of alignment between correct predictions. |
| Approach: | They propose a method for multiple intent detection and slot filling that introduces a modifier and employs reinforcement learning to modify the reference. |
| Outcome: | The proposed method outperforms the previous best approach by 3.6 overall accuracy on MixATIS dataset. |
Mask the Correct Tokens: An Embarrassingly Simple Approach for Error Correction (2022.emnlp-main)
Copied to clipboard
| Challenge: | Text error correction methods usually use the source (incorrect) sentence as encoder input and generate the target (correct) sentences through the decoder. |
| Approach: | They propose a method to correct errors in text sequences by randomly masking out the correct tokens in the source sentence. |
| Outcome: | The proposed method improves accuracy on Mandarin and English datasets with autoregressive and non-autoregressive generation models. |
Retrieval Heads are Dynamic (2026.acl-long)
Copied to clipboard
Yuping Lin, Zitao Li, Yue Xing, Pengfei He, Yingqian Cui, Yaliang Li, Bolin Ding, Jingren Zhou, Jiliang Tang
| Challenge: | Recent studies have identified "retrieval heads" in Large Language Models responsible for extracting information from input contexts. |
| Approach: | They propose to examine retrieval heads from a dynamic perspective . they establish that retrieval head activation is highly dynamic and functionally irreplaceable . |
| Outcome: | The proposed model's hidden state encodes a predictive signal for future retrieval head patterns, indicating an internal planning mechanism. |
Mitigating Sequential Dependencies: A Survey of Algorithms and Systems for Generation-Refinement Frameworks in Autoregressive Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Sequential dependencies present a fundamental bottleneck in deploying large-scale autoregressive models . |
| Approach: | They analyze methods based on generation strategies and refinement mechanisms . they examine deployment strategies across computing environments and explore applications spanning text, images, and speech generation. |
| Outcome: | The proposed frameworks can be used to improve the quality of autoregressive models. |
Maverick: Efficient and Accurate Coreference Resolution Defying Recent Trends (2024.acl-long)
Copied to clipboard
| Challenge: | Large autoregressive generative models have emerged as the cornerstone for achieving the highest performance across several Natural Language Processing tasks. |
| Approach: | They propose a pipeline that trains a state-of-the-art Coreference Resolution system within the constraints of an academic budget and trains with up to 0.006x the memory resources. |
| Outcome: | The proposed framework outperforms encoder-based discriminative systems on the CoNLL-2012 benchmark, training with up to 0.006x the memory resources and obtaining 170x faster inference compared to previous state-of-the-art systems. |
How Speculative Can Speculative Decoding Be? (2024.lrec-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) have a largely increased latency due to their ability to autoregressively model . speculative decoding is a technique that trades generation quality for speed . |
| Approach: | They propose to use a draft model to draft tokens autoregressively and then verify them in parallel. |
| Outcome: | The proposed model could draft tokens autoregressively and then verify them in parallel . the proposed model trades quality for speed and could fail in verification stage . |
Prediction-Augmented Generation for Automatic Diagnosis Tasks (2025.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) adopt autoregressive architecture, predicting the next word token based on the preceding context. |
| Approach: | They propose a method that integrates task-specific predictive models as external tools to improve model generation quality and accuracy. |
| Outcome: | The proposed method improves the generation quality and predictive accuracy of large language models in inference-driven tasks. |
Language Models are Few-Shot Butlers (2021.emnlp-main)
Copied to clipboard
| Challenge: | Pretrained language models demonstrate strong performance in most NLP tasks when fine-tuned on small task-specific datasets. |
| Approach: | They propose a two-stage procedure to learn from a small set of demonstrations and a simple reinforcement learning algorithm to improve by interacting with an environment. |
| Outcome: | The proposed method improves with only 1.2% of the demonstrations and a simple reinforcement learning algorithm over existing methods in the ALFWorld environment. |
CART: A Generative Cross-Modal Retrieval Framework With Coarse-To-Fine Semantic Modeling (2025.acl-long)
Copied to clipboard
Minghui Fang, Shengpeng Ji, Jialong Zuo, Hai Huang, Yan Xia, Jieming Zhu, Xize Cheng, Xiaoda Yang, Wenrui Liu, Gang Wang, Zhenhua Dong, Zhou Zhao
| Challenge: | Cross-modal retrieval tasks are used to retrieve data from one modality or another based on a query from another modality. |
| Approach: | They propose a generative cross-modal retrieval framework based on coarse-to-fine semantic modeling . they propose combining K-Means and RQ-VAE to discretize multimodal data into token sequences that support autoregressive generation. |
| Outcome: | The proposed framework achieves excellent performance and efficiency in multimodal retrieval tasks. |
Diffusion Language Model with Query-Document Relevance for Query-Focused Summarization (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Query-Focused Summarization (QFS) aims to generate summaries that address specific queries by extracting crucial information from source documents. |
| Approach: | They propose a non-autoregressive diffusion language model that incorporates query-document fragment relevance and query-doctoral global relevance to enhance the adaptability of QFS tasks. |
| Outcome: | The proposed model achieves state-of-the-art performance on Debatepedia and PubMedQA datasets in ROUGE scores, GPT-4, and human evaluations. |
Boosting Inference Efficiency: Unleashing the Power of Parameter-Shared Pre-trained Language Models (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Parameter-shared pre-trained language models (PLMs) have emerged as a successful approach in resource-constrained environments. |
| Approach: | They propose a method to enhance the inference efficiency of parameter-shared PLMs by pre-training models that can achieve even greater acceleration. |
| Outcome: | The proposed method improves inference efficiency on autoregressive and autoencoding models. |
CTC-based Non-autoregressive Speech Translation (2023.acl-long)
Copied to clipboard
Chen Xu, Xiaoqian Liu, Xiaowen Liu, Qingxuan Sun, Yuhao Zhang, Murun Yang, Qianqian Dong, Tom Ko, Mingxuan Wang, Tong Xiao, Anxiang Ma, Jingbo Zhu
| Challenge: | End-to-end speech translation (E2E ST) and non-autoregressive (NAR) generation are promising in language and speech processing for their advantages of less error propagation and low latency. |
| Approach: | They develop a model that uses connectionist temporal classification to predict the source and target texts. |
| Outcome: | The proposed model achieves an average BLEU score of 29.5 with a speed-up of 5.67. |
Deep Equilibrium Non-Autoregressive Sequence Learning (2023.findings-acl)
Copied to clipboard
| Challenge: | et al., 2017) is the most prevailing neural architecture for sequence-to-sequence learning. |
| Approach: | They propose to solve for the equilibrium state of NAR models with black-box root-finding solvers and back-propagate through the equilibrium point via implicit differentiation with constant memory. |
| Outcome: | The proposed framework can converge to a more accurate prediction on four WMT benchmarks. |
A Generative Framework for Personalized Sticker Retrieval (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing relevance-based generative retrieval methods lack personalization, leading to a mismatch between diverse user expectations and the retrieved results. |
| Approach: | They propose a representation learning model that learns discriminative user representations to encode user-specific sticker preferences. |
| Outcome: | The proposed framework outperforms state-of-the-art methods in generating relevant stickers for queries. |
Set Generation Networks for End-to-End Knowledge Base Population (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing knowledge base population systems require a machine translation task to generate multiple facts, but the fact order is not considered. |
| Approach: | They propose a knowledge base population task that aims to discover facts about entities from texts and expand a KB with these facts. |
| Outcome: | The proposed networks achieve state-of-the-art (SoTA) performance on two benchmark datasets. |
Improving Autoregressive Grammatical Error Correction with Non-autoregressive Models (2023.findings-acl)
Copied to clipboard
| Challenge: | Autoregressive models assign low probabilities to tokens that need corrections . grammatical error correction (GEC) is widely applied to natural language processing tasks . |
| Approach: | They propose to use a non-autoregressive model as an auxiliary model to train GEC models to correct grammatical errors in sentences. |
| Outcome: | The proposed method outperforms baselines on English and Chinese GEC tasks significantly. |
Maximal Clique Based Non-Autoregressive Open Information Extraction (2021.emnlp-main)
Copied to clipboard
| Challenge: | Open Information Extraction (OpenIE) aims to discover textual facts from a given sentence. |
| Approach: | They propose a non-autoregressive framework that generates a fact graph and a graph with an edge linking two nodes that belong to the same fact. |
| Outcome: | The proposed framework outperforms current state-of-the-art methods on two benchmark datasets and significantly outperformed the existing ones. |
Temperature-Centric Investigation of Speculative Decoding with Knowledge Distillation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Speculative decoding is a novel method to expedite inference in autoregressive (large) language models. |
| Approach: | They propose to use a smaller model as a draft model to speculate a block of tokens, which the target model then evaluates for acceptance. |
| Outcome: | The proposed method can be used to accelerate inference in autoregressive (large) language models by using smaller models as draft models to speculate tokens for multiple inference steps. |
STARS: A Unified Framework for Singing Transcription, Alignment, and Refined Style Annotation (2025.findings-acl)
Copied to clipboard
Wenxiang Guo, Yu Zhang, Changhao Pan, Zhiyuan Zhu, Ruiqi Li, ZheTao Chen, Wenhao Xu, Fei Wu, Zhou Zhao
| Challenge: | Existing automated singing annotation (ASA) methods tackle isolated aspects of the annotation pipeline. |
| Approach: | They propose a framework that addresses transcription, alignment, and refined style annotations. |
| Outcome: | The proposed framework delivers comprehensive multi-level annotations encompassing: (1) precise phoneme-audio alignment, (2) robust note transcription and temporal localization, (3) expressive vocal technique identification, and (4) global stylistic characterization including emotion and pace. |
RecBase: Generative Foundation Model Pretraining for Zero-Shot Recommendation (2025.emnlp-main)
Copied to clipboard
Sashuai Zhou, Weinan Gan, Qijiong Liu, Ke Lei, Jieming Zhu, Hai Huang, Yan Xia, Ruiming Tang, Zhenhua Dong, Zhou Zhao
| Challenge: | Existing methods for addressing item-level user interests are lacking in cross-domain generalization . RecBase model is domain-agnostic and can be used to enhance recommender systems' effectiveness . |
| Approach: | They propose a domain-agnostic foundational model pretrained with a recommendation-oriented objective that leverages a large-scale, heterogeneous, cross-domain corpus with unified textual representations and feature mappings to enhance cross- domain generalization. |
| Outcome: | The proposed model matches or surpasses baselines in zero-shot and cross-domain recommendation tasks on eight real-world datasets. |
SpecBound: Adaptive Bounded Self-Speculation with Layer-wise Confidence Calibration (2026.findings-acl)
Copied to clipboard
| Challenge: | Speculative decoding has emerged as a promising approach to accelerate autoregressive inference in large language models. |
| Approach: | They propose a self-draft framework that suppresses spurious confidence via layer-wise temperature annealing in early-exit decision and adaptively bounds speculation length based on token-wise decoding difficulty. |
| Outcome: | The proposed framework suppresses spurious confidence and bounds speculation length based on token-wise decoding difficulty. |
Efficient Large Scale Language Modeling with Mixtures of Experts (2022.emnlp-main)
Copied to clipboard
Mikel Artetxe, Shruti Bhosale, Naman Goyal, Todor Mihaylov, Myle Ott, Sam Shleifer, Xi Victoria Lin, Jingfei Du, Srinivasan Iyer, Ramakanth Pasunuru, Giridharan Anantharaman, Xian Li, Shuohui Chen, Halil Akin, Mandeep Baines, Louis Martin, Xing Zhou, Punit Singh Koura, Brian O’Horo, Jeffrey Wang, Luke Zettlemoyer, Mona Diab, Zornitsa Kozareva, Veselin Stoyanov
| Challenge: | Mixture of Experts layers (MoEs) enable efficient scaling of language models . large autoregressive language models such as GPT-3 can be adapted to a wide range of tasks . |
| Approach: | They propose to use Mixture of Experts layers to enable efficient scaling of language models . they find that MoEs are substantially more compute efficient than dense models compared to MoE models - but only when they are more modestly trained . |
| Outcome: | The proposed model outperforms dense models in a wide range of tasks and domains. |
PRoDeliberation: Parallel Robust Deliberation for End-to-End Spoken Language Understanding (2024.findings-emnlp)
Copied to clipboard
Trang Le, Daniel Lazar, Suyoun Kim, Shan Jiang, Duc Le, Adithya Sagar, Aleksandr Livshits, Ahmed Aly, Akshat Shrivastava
| Challenge: | End-to-end models for Spoken Language Understanding have been autoregressive, resulting in higher latencies. |
| Approach: | They propose a method that uses Connectionist Temporal Classification to train robust non-autoregressive deliberation models. |
| Outcome: | The proposed method achieves 10x latency reduction over autoregressive models while preserving ability to correct ASR mistranscriptions. |
TextGenSHAP: Scalable Post-Hoc Explanations in Text Generation with Long Documents (2024.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are difficult to explain and understand due to long input contexts and autoregressive output generation. |
| Approach: | They propose a post-hoc explanation method which incorporates LLM-specific techniques. |
| Outcome: | The proposed method improves retrieval recall and prediction accuracy significantly on open-domain question answering benchmarks. |
CONTESTS: a Framework for Consistency Testing of Span Probabilities in Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Language model scores are often treated as probabilities, but their reliability as probability estimators has mainly been studied through calibration, overlooking other aspects. |
| Approach: | They propose a framework to assess model reliability across interchangeable completion and conditioning orders by performing statistical tests on real and synthetic data to eliminate training effects. |
| Outcome: | The proposed framework assesses the consistency of model predictions across interchangeable completion and conditioning orders on real and synthetic data to eliminate training effects. |
Lost in Diffusion: Uncovering Hallucination Patterns and Failure Modes in Diffusion Large Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | dLLMs have emerged as a promising non-autoregressive paradigm for text generation, but their hallucination mechanisms remain underexplored. |
| Approach: | They present the first controlled comparative study to evaluate hallucination patterns in Diffusion Large Language Models. |
| Outcome: | The proposed model exhibits higher propensity for hallucination than AR counterparts controlled for architecture, scale, and pre-training weights. |
Chain-of-Thought Compression Should Not Be Blind: V-Skip for Efficient Multimodal Reasoning via Dual-Path Anchoring (2026.acl-long)
Copied to clipboard
| Challenge: | Existing efforts to mitigate this via token compression fail due to its autoregressive nature . linguistically redundant tokens are erroneously pruned, leading to hallucinations . |
| Approach: | They propose a method that reformulates token pruning as a Visual-Anchored Information Bottleneck (VA-IB) optimization problem. |
| Outcome: | Experiments on Qwen2-VL and Llama-3.2 families show that the proposed model achieves a speedup with negligible accuracy loss. |
Rarely a problem? Language models exhibit inverse scaling in their predictions following few-type quantifiers (2023.findings-acl)
Copied to clipboard
| Challenge: | Current work suggests that language models deal poorly with quantifiers-they struggle to predict which quantifier is used in a given context and also perform poorly at generating appropriate continuations following logical quantifier. |
| Approach: | They propose to use 960 English sentence stimuli to build 22 autoregressive transformer models of different sizes to test their performance on ‘few’-type quantifiers. |
| Outcome: | The proposed models perform poorly on ‘few’-type quantifiers, and the larger the model, the worse its performance. |
Acquisition and Application of Novel Knowledge in Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for constructing new datasets rely on timestamps or simple templates that do not accurately reflect the real world. |
| Approach: | They propose a knowledge dataset construction approach that simulates biological evolution using knowledge graphs to generate synthetic entities with diverse attributes. |
| Outcome: | The proposed framework outperforms knowledge augmentation methods by 3.3%-38%. |
Faster In-Context Learning for LLMs via N-Gram Trie Speculative Decoding (2025.emnlp-main)
Copied to clipboard
| Challenge: | In-Context Learning (ICL) is a key method in prompt engineering, but its long retrieved contexts and limited token throughput will slow reasoning speeds. |
| Approach: | They propose a method that leverages the overlap between context and model output to generate drafts from the context. |
| Outcome: | The proposed method achieves the highest mean speedup on Vicuna-7B, Llama2-7B-Chat, and Llma3-8B-Instruct tasks. |
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. |
InfoDiffusion: Information Entropy Aware Diffusion Process for Non-Autoregressive Text Generation (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing text diffusion models have failed to capture the difference between the “easy-first” process and the “keyword-first" process of humans. |
| Approach: | They propose a non-autoregressive text diffusion model that incorporates a "keyinfo-first" generation strategy and a noise schedule based on the amount of text information. |
| Outcome: | The proposed model outperforms the baseline model in terms of generation quality and diversity, and higher sampling efficiency. |
Combining Constrained and Unconstrained Decoding via Boosting: BoostCD and Its Application to Information Extraction (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent approaches to structured NLP tasks use autoregressive models trained on pairs of unstructured input text and structured output targets. |
| Approach: | They propose a model that combines constrained and unconstrained decoding in two phases to achieve two weak predictions. |
| Outcome: | The proposed model outperforms previous approaches both in and out of distribution, addressing several common errors identified in those approaches. |
FFN-SkipLLM: A Hidden Gem for Autoregressive Decoding with Adaptive Feed Forward Skipping (2024.emnlp-main)
Copied to clipboard
| Challenge: | Autoregressive Large Language Models (LLMs) are omnipresent but typically come with a substantial model size. |
| Approach: | They propose a novel fine-grained skip strategy for autoregressive large language models . they observe the saturation of computationally expensive feed-forward blocks of LLMs . |
| Outcome: | The proposed method can skip 25-30% of FFN blocks with marginal change in performance on knowledge-intensive generation tasks. |
Breaking Block Boundaries: Anchor-based History-stable Decoding for Diffusion Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Semi-autoregressive (Semi-AR) decoding suffers from inherent block constraints . naive lookahead decoding is unreliable, token stability closely correlates with convergence trend, and historical information is isolated. |
| Approach: | They propose a training-free, plug-and-play dynamic decoding strategy that monitors the stability of tokens in real time through dynamic anchors. |
| Outcome: | The proposed approach reduces decoding steps by 80% while improving performance by 3.67% on the BBH benchmark. |
Fixing Semantic Blind Spots in Anchor Tokens of dMLLMs (2026.findings-acl)
Copied to clipboard
| Challenge: | Autoregressive models (ARMs) are prone to hallucinations due to their sequential text generation and high latency. |
| Approach: | They propose a training-free decoding strategy that augments the attention key space with a static, distance-aware matrix to reduce the attention sink effect on semantic anchors. |
| Outcome: | The proposed method reduces the attention sink effect on semantic anchors while enhancing their ability to aggregate global visual information. |
Speculative Streaming: Efficient and Scalable Speculative Decoding with Multi-Stream Attention (2025.emnlp-main)
Copied to clipboard
Nikhil Bhendawade, Irina Belousova, Qichen Fu, Henry Mason, Antonie Lin, Mohammad Rastegari, Mahyar Najibi
| Challenge: | Speculative decoding is a prominent technique for accelerating LLM inference by leveraging an auxiliary draft model, but its effectiveness is limited by the autoregressive nature of draft generation. |
| Approach: | They propose a method that integrates speculative draft generation directly within the target model using multi-stream attention. |
| Outcome: | The proposed method improves acceptance but also latency and speculation latency, limiting overall speedup. |
RACER: Retrieval-Augmented Contextual Rapid Speculative Decoding (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for decoding large language models generate one token per step, causing high inference latency. |
| Approach: | They propose a method that integrates retrieved exact patterns with logit-driven future cues. |
| Outcome: | Experiments on Spec-Bench, HumanEval, and MGSM-ZH show that RACER outperforms training-free methods and accelerates inference. |
SelFusion: Self-distillation for Diffusion Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing knowledge distillation methods for autoregressive large language models (LLMs) are not effective for reducing generation quality, but they can be useful for real-time applications. |
| Approach: | They propose a self-distillation framework that allows for effective KD without external teacher . they propose to use two modes of knowledge distillation to determine distillation direction . |
| Outcome: | The proposed framework outperforms existing methods with external teachers on instruction-following tasks. |
Using Shapley interactions to understand how models use structure (2025.acl-long)
Copied to clipboard
| Challenge: | Language models are intricately structured systems, and attribution measures are important for understanding how they combine features to influence outputs. |
| Approach: | They use Shapley Taylor interaction indices to examine how language and speech models internally relate and structure their inputs. |
| Outcome: | The proposed methods show that language models encode phonetic interactions . they show that the inputs are more entangled for pairs where a consonant influences a vowel or approximant . |
Exploring Layer-wise Representations of English and Chinese Homonymy in Pre-trained Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | lexical ambiguity can arise due to the misunderstanding of its multiple senses. |
| Approach: | They propose to use part of speech to examine homonyms in Chinese and English . they find no universal layer depth excels in differentiating homnomial representations . |
| Outcome: | The proposed model improves contextualization of homonym representations in Chinese . the results challenge the simplistic understanding of their inner workings, the authors say . |
TESS 2: A Large-Scale Generalist Diffusion Language Model (2025.acl-long)
Copied to clipboard
| Challenge: | Existing instruction-following diffusion models are predominantly trained using an autoregressive paradigm. |
| Approach: | They propose a general instruction-following diffusion language model that outperforms contemporary instruction-tuned diffusion models and matches and sometimes exceeds strong autoregressive (AR) models. |
| Outcome: | The proposed model outperforms and sometimes exceeds existing autoregressive (AR) models on a number of tasks. |
LLMVoX: Autoregressive Streaming Text-to-Speech Model for Any LLM (2025.findings-acl)
Copied to clipboard
Sambal Shikhar, Mohammed Irfan Kurpath, Sahal Shaji Mullappilly, Jean Lahoud, Fahad Shahbaz Khan, Rao Muhammad Anwer, Salman Khan, Hisham Cholakkal
| Challenge: | Existing speech-enabled LLMs degrade conversational quality by modifying the LLM, compromising its linguistic capabilities. |
| Approach: | They propose a lightweight 30M-parameter, LLM-agnostic, autoregressive streaming TTS system that generates high-quality speech with low latency. |
| Outcome: | The proposed system achieves a significantly lower word error rate compared to speech-enabled LLMs while operating at comparable latency. |
EZ-VC: Easy Zero-shot Any-to-Any Voice Conversion (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Current voice conversion methods struggle in zero-shot cross-lingual settings . authors develop a method that can be used in zero shot cross-linguistic settings despite advances in technology . |
| Approach: | They propose a voice-conversion model that combines discrete speech representations with a non-autoregressive speech decoder. |
| Outcome: | The proposed approach excels in zero-shot cross-lingual settings even for unseen languages and accents. |
SDAR: A Synergistic Diffusion-AutoRegression Paradigm for Scalable Sequence Generation (2026.findings-acl)
Copied to clipboard
Shuang Cheng, Yihan Bian, Dawei Liu, Yuhua Jiang, Yihao Liu, Linfeng Zhang, Qian Yao, Zhongbo Tian, Wenhai Wang, Qipeng Guo, Kai Chen, Biqing Qi, Bowen Zhou
| Challenge: | Autoregressive (AR) language models are a dominant paradigm in the field of parallelism and non-causal modeling. |
| Approach: | They propose a blockwise discrete diffusion model that preserves AR-compatible serving while enabling parallel intra-block generation. |
| Outcome: | The proposed model achieves theoretical speedups over 5 and wall-clock speedup of 2.3 on H200 GPUs in latency-critical regimes. |
Prediction Hubs are Context-Informed Frequent Tokens in LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Hubness is a tendency for a few points to be among the nearest neighbours of a disproportionate number of other points. |
| Approach: | They show that only large-scale representation comparisons are not characterized by hubness . they show that hubs are the result of context-modulated frequent tokens . |
| Outcome: | The results show that the comparison between context and unembedding vectors does not result in hubness . the findings suggest that hubness is not a negative property that needs to be mitigated when LLMs are being used for next token prediction. |
DiTReducio: A Training-Free Acceleration for DiT-Based TTS via Progressive Calibration (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing training-free acceleration approaches for text-to-speech models are constrained by training costs. |
| Approach: | They propose a training-free acceleration framework that compresses computations in DiT-based TTS models . they propose Temporal Skipping and Branch Skipping to eliminate redundant computations . |
| Outcome: | Experimental results show that the proposed framework reduces FLOPs and improves RTF by 37.1%. |
Exploring the Hidden Capacity of LLMs for One-Step Text Generation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) can reconstruct surprisingly long texts via autoregressive generation from just one trained input embedding. |
| Approach: | They show that large language models can reconstruct surprisingly long texts via autoregressive generation from just one trained input embedding. |
| Outcome: | The proposed model can generate hundreds of accurate tokens in one token-parallel forward pass, when provided with only two learned embeddings. |
LLM-ForcedAligner: A Non-Autoregressive and Accurate LLM-Based Forced Aligner for Multilingual and Long-Form Speech (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for forcing alignment are language-specific and prone to temporal shifts. |
| Approach: | They propose a slot-filling paradigm that uses time indices to predict slot positions. |
| Outcome: | The proposed method reduces accumulated temporal shifts by 69% compared with prior methods. |
Local Normalization Distortion and the Thermodynamic Formalism of Decoding Strategies for Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing decoding strategies for language models are based on heuristics and are difficult to improve. |
| Approach: | They propose a theory of decoding strategies for language models by expressing popular decoding algorithms as equilibrium states and stating objective functions they optimize. |
| Outcome: | The proposed method is based on ergodic theory and is difficult to improve . it shows that local normalization distortion is a fundamental defect of decoding strategies . |
Releasing the Capacity of GANs in Non-Autoregressive Image Captioning (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing non-autoregressive (NAR) models suffer from their inherent multi-modality problem. |
| Approach: | They propose an Adversarial Non-autoregressive Transformer for Image Captioning that improves model performance by modifying model structure to be compatible with contrastive learning. |
| Outcome: | The proposed model achieves 26.72 times faster than the autoregressive model on the MSCOCO dataset. |
Sprout: Green Generative AI with Carbon-Efficient LLM Inference (2024.emnlp-main)
Copied to clipboard
| Challenge: | Sprout reduces the carbon footprint of inference in large language models by over 40% in real-world evaluations. |
| Approach: | Sprout introduces "generation directives" to guide autoregressive generation process . et al. cites Llama model and global electricity grid data as examples . |
| Outcome: | Sprout reduces the carbon footprint of generative AI models by over 40% in real-world evaluations using the Llama model and global electricity grid data. |
Scale-VAE: Preventing Posterior Collapse in Variational Autoencoder (2024.lrec-main)
Copied to clipboard
| Challenge: | Variational autoencoder (VAE) is a widely used generative model . but when employing strong autoregressive generation networks, VAE tends to converge to a degenerate local optimum known as posterior collapse. |
| Approach: | They propose a model called Scale-VAE to solve a posterior collapse problem . they use a factor to keep the posterior dimension discriminative across data instances . |
| Outcome: | The proposed model outperforms state-of-the-art models in density estimation and representation learning. |
Preference Optimization for Review Question Generation Improves Writing Quality (2026.findings-acl)
Copied to clipboard
| Challenge: | Peer reviewers are overloaded and face tight deadlines, leading some to rely on large language models (LLMs) to draft questions and comments. |
| Approach: | They use open-review review datasets to train a human preference model based on human reviewer questions . human evaluations show IntelliAsk generates more grounded, substantive and effortful questions than strong baselines . |
| Outcome: | The proposed model predicts reviewer-question quality better than API-based SFT baselines and provides scalable evaluation. |
Hetero-Designer: Automated Design of Multi-Agent Systems with Heterogeneous LLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to design LLM-based Multi-agent systems are constrained by homogeneous LLMs. |
| Approach: | They propose an automated design of heterogeneous-LLMs-based MAS with a binary-star transformer and an autoregressive graph generation pipeline. |
| Outcome: | The proposed pipeline is high-performing on various benchmarks and extensible to unseen LLMs and roles. |
Predicting Through Generation: Why Generation Is Better for Prediction (2025.acl-long)
Copied to clipboard
Md Kowsher, Nusrat Jahan Prottasha, Prakash Bhat, Chun-Nam Yu, Mojtaba Soltanalian, Ivan Garibay, Ozlem Garibay, Chen Chen, Niloofar Yousefi
| Challenge: | Large Language Models (LLMs) are increasingly used for predictive tasks such as classification and regression. |
| Approach: | They propose a framework that generates output tokens from mas-sive text corpora and a task adapter to ensure consistency between token generation and final prediction. |
| Outcome: | The proposed framework outperforms baseline models on classification and regression benchmarks and the proposed framework consistently outperformed standard baseline models. |
LLM-Codec: Neural Audio Codec Meets Language Model Objectives (2026.findings-acl)
Copied to clipboard
| Challenge: | Neural audio codecs are optimized for waveform reconstruction rather than autoregressive prediction. |
| Approach: | They propose to augment codec training with language-model-facing objectives while keeping both codec and LLM architectures unchanged. |
| Outcome: | The proposed model improves speech coherence and predictability by preserving the semantic alignment between audio and text representations. |
Cautious Next Token Prediction (2025.findings-acl)
Copied to clipboard
Yizhou Wang, Lingzhi Zhang, Yue Bai, Mang Tik Chiu, Zhengmian Hu, Mingyuan Zhang, Qihua Dong, Yu Yin, Sohrab Amirghodsi, Yun Fu
| Challenge: | Existing methods for decoding autoregressive models are temperature scaling and nucleus sampling to balance diversity and coherence. |
| Approach: | They propose a training-free decoding strategy that uses a model with a low perplexity score to select the trial with the lowest perplexities as the most probable and reliable path. |
| Outcome: | The proposed approach outperforms existing standard decoding strategies consistently by a clear margin. |
Step-by-Step: Controlling Arbitrary Style in Text with Large Language Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing methods for autoregressive text generation have low controllability and accumulating errors. |
| Approach: | They propose a three-stage prompt-based approach to express autoregressive text in a specific region editing task using a word frequency-based strategy. |
| Outcome: | Experiments on publicly competitive datasets confirm that the proposed approach achieves state-of-the-art performance. |
SDAR-VL: Stable and Efficient Block-wise Diffusion for Vision-Language Understanding (2026.acl-long)
Copied to clipboard
| Challenge: | Existing block-wise discrete diffusion models lack robust autoregressive (AR) decoders. |
| Approach: | They propose a block-wise discrete diffusion framework for large-scale vision-language understanding with a progressive beta noise curriculum. |
| Outcome: | The proposed framework improves training efficiency, convergence stability, and task performance over conventional block diffusion. |
Safety Guardrails of Large Language Models Are Vulnerable to Value-Driven Adversarial Prompting (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing jailbreak attacks against large language models (LLMs) can be divided into white-box attacks and black-box attack. |
| Approach: | They propose a value-driven jailbreak attack that exploits the phenomenon that large language models agree with humans to induce LLMs to affirm the moral value of harmful tasks. |
| Outcome: | Extensive experiments on five state-of-the-art (SOTA) LLMs show the value-driven jailbreak attack achieves an average attack success rate (ASR) of 91.8% on JailbreakBench and 95.2% on the AdvBench subset. |
Spec-VLA: Speculative Decoding for Vision-Language-Action Models with Relaxed Acceptance (2025.emnlp-main)
Copied to clipboard
| Challenge: | Visual Language Models (VLMs) have significant parameter size and autoregressive (AR) decoding nature impose considerable computational demands on VLA models. |
| Approach: | They propose a framework to relax acceptance utilizing the relative distances represented by the action tokens of the VLA model. |
| Outcome: | Empirical results show that the proposed framework improves the speed of the prediction task by 44%. |
Masked Diffusion Captioning for Visual Feature Learning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Unlike autoregressive captioning, the strength of the visual learning signal in MDC does not depend on each token’s position in the sequence, reducing the need for auxiliary objectives. |
| Approach: | a decoder conditioned on visual features is trained to reconstruct the original text. |
| Outcome: | masked diffusion captioning (MDC) is a form of image-conditioned captioning that can be applied to visual tasks. |
Reviving Cultural Heritage: A Novel Approach for Comprehensive Historical Document Restoration (2025.acl-long)
Copied to clipboard
Yuyi Zhang, Peirong Zhang, Zhenhua Yang, Pengyu Yan, Yongxin Shi, Pengwei Liu, Fengjun Guo, Lianwen Jin
| Challenge: | Existing methods for historical document restoration focus on single modality or limited-size restoration, failing to meet practical needs. |
| Approach: | They propose a full-page HDR dataset and an automated HDR solution to replace manual restoration methods. |
| Outcome: | The proposed solution improves OCR accuracy from 46.83% to 84.05% when processing severely damaged documents, with enhancement to 94.25% through human-machine collaboration. |
An Empirical Study of Iterative Refinements for Non-autoregressive Translation (2025.acl-long)
Copied to clipboard
| Challenge: | Iterative non-autoregressive (NAR) models have recently demonstrated impressive performance in varied generation tasks, surpassing the autoregressive Transformer. |
| Approach: | They propose a strategy to conduct efficient refinements without performance declines by using two simple metrics to identify potential problems existing in current refinement processes. |
| Outcome: | The proposed model outperforms the autoregressive Transformer by around one BLEU on average. |
Do All Autoregressive Transformers Remember Facts the Same Way? A Cross-Architecture Analysis of Recall Mechanisms (2025.emnlp-main)
Copied to clipboard
| Challenge: | Prior work has identified MLP modules in early layers as key contributors to factual recall. |
| Approach: | They conduct a comprehensive evaluation of factual recall across several models . they find that Qwen-based models behave differently from previous patterns . |
| Outcome: | The results show that attention modules in the earliest layers contribute more to factual recall than MLP modules. |
MENTOR: Efficient Autoregressive Image Generation with Balanced Multimodal Control (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent text-to-image models achieve impressive visual quality but still face challenges in precise controllability, balancing multimodal inputs, and high training cost for multimodal image generation. |
| Approach: | They propose an autoregressive framework with a two-stage training paradigm for controllable multimodal image generation. |
| Outcome: | Extensive experiments on DreamBench++ and DreamBech show that the proposed framework achieves a strong balance between textual and visual guidance for controllable image generation. |
TiC-LM: A Web-Scale Benchmark for Time-Continual LLM Pretraining (2025.acl-long)
Copied to clipboard
Jeffrey Li, Mohammadreza Armandpour, Seyed Iman Mirzadeh, Sachin Mehta, Vaishaal Shankar, Raviteja Vemulapalli, Samy Bengio, Oncel Tuzel, Mehrdad Farajtabar, Hadi Pouransari, Fartash Faghri
| Challenge: | Large language models (LLMs) trained on historical web data inevitably become outdated. |
| Approach: | They propose a web-scale dataset for time-continual pretraining of LLMs derived from 114 dumps of Common Crawl (CC) they also design time-stratified evaluations to assess how well various continual learning methods adapt to new data while retaining past knowledge. |
| Outcome: | The proposed benchmarks show that autoregressive meta-schedules combined with a fixed-ratio replay of older data can achieve comparable held-out loss to re-training from scratch, while requiring significantly less computation (2.6x). |
Anchored Cyclic Generation: A Novel Paradigm for Long-Sequence Symbolic Music Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Autoregressive sequence modeling has been successful in many domains, but maintaining long-term coherence and structural integrity remains a challenge. |
| Approach: | They propose an ACG paradigm that relies on anchor features from previously generated musical content to guide subsequent generation during the autoregressive process. |
| Outcome: | The proposed framework outperforms existing methods in symbolic music generation tasks. |
Mapping 1,000+ Language Models via the Log-Likelihood Vector (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods to compare autoregressive language models are based on log-likelihoods . a model map is constructed using coordinates that capture the geometric structure of probability distributions based upon text-generation probabilities. |
| Approach: | They propose to use log-likelihood vectors to compare autoregressive language models . when treated as model features, their squared Euclidean distance approximates KL divergence . |
| Outcome: | The proposed method is highly scalable and easy to implement. |
Flexible-length Text Infilling for Discrete Diffusion Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing discrete diffusion models lack flexibility for text infilling without ground-truth positional data. |
| Approach: | They propose a discrete diffusion model that jointly denoises token values and token positions using a novel sample-level Optimal Transport coupling. |
| Outcome: | The proposed method outperforms existing methods on infilling benchmarks such as One-Billion-Word and Yelp. |
Memorization, Emergence, and Explaining Reversal Failures: A Controlled Study of Relational Semantics in LLMs (2026.acl-long)
Copied to clipboard
Yihua Zhu, Qianying Liu, Jiaxin Wang, Fei Cheng, Chaoran Liu, Akiko Aizawa, Sadao Kurohashi, Hidetoshi Shimodaira
| Challenge: | Autoregressive LLMs perform well on relational tasks that require linking entities via relational words, but it is unclear whether they learn the logical semantics of such relations or whether left-to-right order bias is involved. |
| Approach: | They propose a framework that generates text from symmetric/inverse triples and trains autoregressive models from scratch. |
| Outcome: | The proposed framework generates text from symmetric/inverse triples, trains autoregressive models from scratch, and evaluates memorization, logical inference, and in-context generalization to unseen entities. |
Reward-Weighted Sampling: Enhancing Non-Autoregressive Characteristics in Masked Diffusion LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | Masked diffusion models (MDMs) are promising non-autoregressive alternatives for large language modeling. |
| Approach: | They propose a method that leverages an external reward model to provide a principled global signal during the iterative diffusion process. |
| Outcome: | The proposed method improves non-autoregressive generation orders and performance across evaluation metrics. |
Fast-Decoding Diffusion Language Models via Progress-Aware Confidence Schedules (2026.findings-acl)
Copied to clipboard
| Challenge: | *SchED* is a training-free, model-agnostic early-exit algorithm that terminates diffusion decoding using a progress-aware confidence threshold. |
| Approach: | They propose a training-free, model-agnostic early-exit algorithm that terminates diffusion decoding using a progress-aware confidence threshold. |
| Outcome: | The proposed algorithm achieves 4 speedups on instruction-tuned models while maintaining baseline performance on average. |
Amadeus: Autoregressive Model with Bidirectional Attribute Modelling for Symbolic Music (2026.acl-long)
Copied to clipboard
| Challenge: | Existing symbolic music generation models represent musical notes as a sequence of attribute tokens with fixed unidirectional dependencies. |
| Approach: | They propose a symbolic music generation framework that adopts a autoregressive and a discrete diffusion architectures for note attributes. |
| Outcome: | The proposed framework improves state-of-the-art models across objective and subjective metrics. |
Autoregressive Semantic Visual Reconstruction Helps VLMs Understand Better (2026.findings-acl)
Copied to clipboard
| Challenge: | Typical large vision-language models emphasize vision-to-language alignment while overlooking fine-grained visual information. |
| Approach: | They introduce autoregressive semantic visual reconstruction (ASVR) that enables joint learning of visual and textual modalities within a unified autoregression framework. |
| Outcome: | The proposed model improves baselines and multimodal understanding benchmarks by 2-3%. |
Towards Efficient Large Language Model Serving: A Survey on System-Aware KV Cache Optimization (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) are a powerful tool for high-performance inference serving. |
| Approach: | They focus on system-aware KV infrastructure for serving LLMs . they analyze cross-behavior co-design affinity and behavior-objective links . |
| Outcome: | The proposed key-value (KV) cache is crucial for low-latency, high-throughput LLM inference serving. |
Paramanu: Compact and Competitive Monolingual Language Models for Low-Resource Morphologically Rich Indian Languages (2026.acl-long)
Copied to clipboard
| Challenge: | Multilingual large language models are expensive to pretrain and suffer from imbalances across languages and datasets. |
| Approach: | They propose a family of Indian language-only autoregressive language models trained on open-source language-specific data for the five most spoken Indian languages. |
| Outcome: | The proposed model outperforms most larger models up to 8B across all five languages. |
ZipVoice-Dialog: Non-Autoregressive Spoken Dialogue Generation with Flow Matching (2026.findings-acl)
Copied to clipboard
Han Zhu, Wei Kang, Liyong Guo, Zengwei Yao, Fangjun Kuang, Weiji Zhuang, Zhaoqing Li, Zhifeng Han, Dong Zhang, Xin Zhang, Xingchen Song, Lingxuan Ye, Long Lin, Daniel Povey
| Challenge: | Existing autoregressive models for dialogue generation suffer from high latency and stability issues. |
| Approach: | They propose a non-autoregressive (NAR) zero-shot spoken dialogue generation model based on flow-matching. |
| Outcome: | The proposed model outperforms existing models in speech generation due to poor speech intelligibility and turn-taking precision. |
Advancing Reasoning in Diffusion Language Models with Denoising Process Rewards (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for improving reasoning in diffusion language models rely on outcome-based rewards that provide no direct supervision over the denoising process. |
| Approach: | They propose a method that provides a process-level reinforcement signal over denoising trajectory of diffusion language models. |
| Outcome: | Experiments on challenging reasoning benchmarks show that the proposed model improves reasoning stability, interpretability and overall performance. |
Beyond Fully Random Masking: Attention-Guided Denoising and Optimization for Diffusion Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for full-attention dLLMs rely on random masking strategies that overlook intrinsic token dependencies. |
| Approach: | They propose an attention-guided denoising and optimization framework that aligns training and optimization with attention-derived dependencies. |
| Outcome: | The proposed framework outperforms state-of-the-art methods on mathematical and coding benchmarks. |
SceneLM: 3D-Aware Language Models for Editable 3D Scene Synthesis (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for synthesising 3D scenes from a single image are text-driven and lack precise metric understanding from images. |
| Approach: | They propose a language-model-based framework that grounds 3D scene synthesis in visual evidence by recovering an executable metric 3D layout directly from a single image. |
| Outcome: | The proposed framework recovers an executable metric 3D layout directly from an RGB image and instantiates, places, and edits objects for iterative refinement. |
Reinforcement Learning for Diffusion LLMs via Energy-Based Gibbs Alignment (2026.acl-long)
Copied to clipboard
| Challenge: | Diffusion Large Language Models (dLLMs) offer parallel decoding and bidirectional context modeling . aligning dLLms with reinforcement learning (RL) remains a challenge . |
| Approach: | They propose a variational framework that reformulates RL for dLLMs as a distribution matching problem. |
| Outcome: | The proposed framework reformulates RL for dLLMs as a distribution matching problem. |
The Confidence Paradox: Unveiling the Latent Discriminative Power of Diffusion Large Language Models in Mathematical Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Diffusion large language models (DLLMs) are a promising alternative to autoregressive (AR) generation, offering token-level probabilities under bidirectional context. |
| Approach: | They propose to use diffusion large language models to generate token-level probabilities under bidirectional context and to examine the calibration paradox inherent to their native uncertainty estimates. |
| Outcome: | The proposed model outperforms AR baselines on mathematical reasoning benchmarks and is highly miscalibrated on reasoning benchmark. |
EDSD: Entropy-Driven Design for Faster Speculative Decoding (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for speculative decoding incur substantial training overhead to mitigate information misalignment between autoregressive draft model training and decoding. |
| Approach: | They propose an Entropy-Driven Speculative Decoding framework that uses entropy as a unified, interpretable signal for both draft model training and architectural design. |
| Outcome: | Experiments on seven large language models show that EDSD improves training efficiency by 24.8% and increases acceptance length by 4.0% compared to state-of-the-art methods. |
Attention Sinks in Diffusion Language Models (2026.findings-acl)
Copied to clipboard
Maximo Eduardo Rulli, Simone Petruzzi, Edoardo Michielon, Fabrizio Silvestri, Simone Scardapane, Alessio Devoto
| Challenge: | Masked Diffusion Language Models (DLMs) employ transformer encoders with bidirectional attention, enabling parallel token generation while maintaining competitive performance. |
| Approach: | They conduct an empirical analysis of DLM attention patterns focusing on the attention sinking phenomenon . they find that DLMs also exhibit attention sinks, but with distinct characteristics . |
| Outcome: | The proposed models employ transformer encoders with bidirectional attention, enabling parallel token generation while maintaining competitive performance. |
LR-DWM: Efficient Watermarking for Diffusion Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Current methods for large language models rely on tokens being generated sequentially . left-right Diffusion watermarking uses a fixed, deterministic left-to-right order . |
| Approach: | They propose a scheme that biases tokens based on both left and right neighbors . left-Right Diffusion Watermarking is a low-latency alternative to autoregressive models . |
| Outcome: | The proposed method can be watermarked efficiently with minimal runtime and memory overhead. |