Papers with training method
ERNIE-M: Enhanced Multilingual Representation by Aligning Cross-lingual Semantics with Monolingual Corpora (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for pretraining cross-lingual models are limited in their size due to the limited amount of parallel corpora. |
| Approach: | They propose a method that encourages the model to align multiple languages with monolingual corpora to overcome the constraint of the parallel corpus size. |
| Outcome: | The proposed method outperforms existing cross-lingual models and delivers new state-of-the-art results in various cross-linguistic downstream tasks. |
Contrastive Learning for Many-to-many Multilingual Neural Machine Translation (2021.acl-long)
Copied to clipboard
| Challenge: | Existing multilingual machine translation approaches focus on English-centric directions, while non-English directions lag behind. |
| Approach: | They propose a multilingual machine translation system with an emphasis on non-English directions. |
| Outcome: | The proposed model outperforms existing models on English-centric and non-English directions on multilingual translation benchmarks. |
Toward Building General Foundation Models for Language, Vision, and Vision-Language Understanding Tasks (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing foundation models can only perform the best in one type of understanding tasks. |
| Approach: | They propose a method for training a general foundation model, X-FM, using text, image, and image-text data. |
| Outcome: | The proposed method outperforms existing foundation models on language, vision, and vision-language understanding tasks. |
Balancing Lexical and Semantic Quality in Abstractive Summarization (2023.acl-short)
Copied to clipboard
| Challenge: | Existing methods to reduce exposure bias in sequence-to-sequence models are underexplored. |
| Approach: | They propose a method to re-rank sequence-to-sequence neural models to reduce exposure bias. |
| Outcome: | The proposed method achieves an 89.67 BERTScore on the CNN/DailyMail and XSum datasets. |
Stabilized In-Context Learning with Pre-trained Language Models for Few Shot Dialogue State Tracking (2023.findings-eacl)
Copied to clipboard
| Challenge: | Prompt-based methods with large pre-trained language models have shown impressive unaided performance across many NLP tasks. |
| Approach: | They propose a meta-learning scheme to stabilize the ability of the model to perform well under various prompts and introduce a saliency model to limit dialogue text length. |
| Outcome: | The proposed model improves on large pre-trained language models with labeled in-context exemplars and can be used to generate more exemplar queries. |
Towards Robust Extractive Question Answering Models: Rethinking the Training Methodology (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing models lack robustness against distribution shifts and adversarial attacks when training on unanswerable questions in EQA datasets. |
| Approach: | They propose a novel loss function for the EQA problem to improve the robustness of extractive question answering models by adding adversarial questions to a crowdsourcing process. |
| Outcome: | The proposed method maintains in-domain performance while improving on out-of-domain datasets. |
exBERT: Extending Pre-trained Models with Domain-specific Vocabulary Under Constrained Training Resources (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to train pre-trained models with limited corpus and computational resources are limited by the complexity of the training resources. |
| Approach: | They propose a method to extend BERT pre-trained models from a general domain to a new pre-train model for a specific domain with a different additive vocabulary. |
| Outcome: | The proposed method outperforms existing methods on biomedical benchmark tasks using the MTL-Bioinformatics-2016 dataset. |
Translation-based Supervision for Policy Generation in Simultaneous Neural Machine Translation (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to train simultaneous machine translation agents have been used to find the optimal action sequences for translation quality and lag. |
| Approach: | They propose a supervised learning approach that detects minimum reads required for generating target tokens by comparing simultaneous translations against full-sentence translations. |
| Outcome: | The proposed method produces much higher quality translations while minimizing the average lag in simultaneous translation. |
Improving Named Entity Recognition by External Context Retrieving and Cooperative Learning (2021.acl-long)
Copied to clipboard
| Challenge: | Recent work shows document-level contexts can significantly improve Named Entity Recognition models. |
| Approach: | They propose to find external contexts of a sentence by retrieving and selecting a set of semantically relevant texts through a search engine with the original sentence as the query. |
| Outcome: | The proposed approach can achieve new state-of-the-art performance on 8 NER data sets across 5 domains. |
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. |
GAML-BERT: Improving BERT Early Exiting by Gradient Aligned Mutual Learning (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to improve the early exiting of natural language processing (NLP) are notoriously gigantic and slow in both training and inference. |
| Approach: | They propose a framework for improving the early exiting of BERT by asking each exit to distill knowledge from each other. |
| Outcome: | The proposed framework outperforms the state-of-the-art (SOTA) BERT early exiting methods on the GLUE benchmark. |
Entailment-Preserving First-order Logic Representations in Natural Language Entailment (2025.acl-long)
Copied to clipboard
| Challenge: | First-order logic (FOL) is often used to represent logical entailment, but determining natural language (NL) enanglement using FOL remains a challenge. |
| Approach: | They propose an Entailment-Preserving FOL representations task and a method which trains an NL-to-FOL translator by using the natural language entailment labels as verifiable rewards. |
| Outcome: | The proposed method achieves 1.8–2.7% improvement in EPR and 17.4–20.6% increase in E PR@16 compared to baselines in three datasets. |
Best of Both Worlds: Making High Accuracy Non-incremental Transformer-based Disfluency Detection Incremental (2021.acl-long)
Copied to clipboard
| Challenge: | Currently, Transformer-based text classifiers are not suitable for live incremental processing, operating only on the level of complete sentence inputs. |
| Approach: | They propose to introduce a method for word-by-word left-to-right incremental processing to Transformers such as BERT, models without an intrinsic sense of linear order. |
| Outcome: | The proposed method maintains high non-incremental performance while operating strictly incrementally. |
Distilling the Knowledge of Large-scale Generative Models into Retrieval Models for Efficient Open-domain Conversation (2021.findings-emnlp)
Copied to clipboard
| Challenge: | generative models are less practical for building real-time conversation systems due to high latency and large memory footprint. |
| Approach: | They propose a method that preserves the efficiency of a retrieval model while leveraging the conversational ability of generative models. |
| Outcome: | The proposed method preserves the efficiency of a retrieval model while leveraging the conversational ability of generative models. |
REAR: A Relevance-Aware Retrieval-Augmented Framework for Open-Domain Question Answering (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to extend knowledge scope of large language models (LLMs) lack internal parametric knowledge, resulting in misusing external knowledge. |
| Approach: | They propose a retrieval-augmented approach that provides LLMs with potentially relevant documents through a module. |
| Outcome: | The proposed approach outperforms existing methods on four open-domain QA tasks. |
Turning Trash into Treasure: Accelerating Inference of Large Language Models with Token Recycling (2025.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) generate only one token at each decoding step, leading to high latency. |
| Approach: | They propose a speculative decoding paradigm that stores tokens in an adjacency matrix and employs a breadth-first-search algorithm to construct a draft tree. |
| Outcome: | The proposed method outperforms existing train-free methods by 30% and even a training method by 25%. |
Investigating and Enhancing Vision-Audio Capability in Omnimodal Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent years have witnessed significant advancements in large language models (LLMs) but still struggle with integrating vision and audio. |
| Approach: | They propose a self-knowledge distillation method to improve vision-audio capabilities of OLLMs by learning from the vision-text components. |
| Outcome: | The proposed method improves vision-audio capabilities of OLLMs by learning from vision-text components, which improves interaction between audio and images and results in improved performance on multimodal tasks. |
Can We Achieve High-quality Direct Speech-to-Speech Translation without Parallel Speech Data? (2024.acl-long)
Copied to clipboard
| Challenge: | Existing two-pass direct speech-to-speech translation models require parallel speech data to train, which is challenging to collect. |
| Approach: | They propose a two-pass direct speech-to-speech translation (S2ST) model that decomposes the task into speech- to-text translation (s2TT) and text-tospech (TTS) they propose 'composer' S2ST model that integrates pretrained S2TT and TTS models into a direct S2 ST model. |
| Outcome: | The proposed model integrates pretrained S2TT and TTS models into a direct S2ST model without parallel speech data. |
DDxGym: Online Transformer Policies in a Knowledge Graph Based Natural Language Environment (2024.lrec-main)
Copied to clipboard
Benjamin Winter, Alexei Gustavo Figueroa Rosero, Alexander Loeser, Felix Alexander Gers, Nancy Katerina Figueroa Rosero, Ralf Krestel
| Challenge: | specialized OpenAI Gym environment for clinical differential diagnosis is limited by data access due to privacy concerns. |
| Approach: | They propose a specialized OpenAI Gym environment for clinical differential diagnosis . they frame the task as a natural-language-based reinforcement learning problem . |
| Outcome: | The proposed model improves over baselines and improves on existing models. |
Joint Audio/Text Training for Transformer Rescorer of Streaming Speech Recognition (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies have shown that streaming end-to-end speech recognition models suffer from higher word error rates (WER) compared to non-streaming models, streaming endto-ended ASR models are limited to short audio context or not use future context to satisfy low latency constraints. |
| Approach: | They propose a 2nd-pass rescoring model on top of the 1st-pass streaming model to improve recognition accuracy while keeping latency low. |
| Outcome: | The proposed method improves word error rate significantly compared to the existing model without adding any additional parameters or latency. |
Understanding Reference Policies in Direct Preference Optimization (2025.findings-naacl)
Copied to clipboard
| Challenge: | Direct Preference Optimization (DPO) is a widely used training method for the fine-tuning of large language models (LLMs). |
| Approach: | They explore the dependency on the reference model or policy in DPO and examine the necessity of the KL-constraint from the reference policies in D PO. |
| Outcome: | The proposed model is sensitive to the KL divergence constraint, and the reference policy can penalize deviations from the reference constraint. |
NSL-MT: Linguistically Informed Negative Samples for Efficient Machine Translation in African Low-Resource Languages (2026.findings-acl)
Copied to clipboard
| Challenge: | In low-resource settings, models encounter too few examples to reliably distinguish grammatical patterns from noise. |
| Approach: | They propose a negative space learning machine translation (NSL-MT) method that augments limited parallel data with synthetically generated violations of the target language’s grammar and explicitly penalizes the model when it assigns high probability to these violations. |
| Outcome: | The proposed method delivers 3-12% BLEU gains for well-performing models and 56-89% gains for models lacking decent initial support. |
Towards Formality-Aware Neural Machine Translation by Leveraging Context Information (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Formality is one of the most important linguistic properties to determine the naturalness of translation. |
| Approach: | They propose a method to explicitly inform neural machine translation models by pinpointing key informative tokens using a formality classifier. |
| Outcome: | The proposed method improves translation quality and conforms to the appropriate syntax. |
Universal Sentence Representation Learning with Conditional Masked Language Model (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to learn sentence representations on unlabeled corpora are difficult and expensive to obtain, making it hard to cover many domains and languages. |
| Approach: | They propose a method to train sentence representations on large unlabeled corpora by conditioning on the encoded vectors of adjacent sentences. |
| Outcome: | The proposed method outperforms existing models on SentEval and can be extended to a broad range of languages and domains. |
Measuring and Increasing Context Usage in Context-Aware Machine Translation (2021.acl-long)
Copied to clipboard
| Challenge: | Recent work in neural machine translation has demonstrated the necessity and feasibility of using inter-sentential context, but it is often not clear how much they actually utilize it at translation time. |
| Approach: | They propose a conditional cross-mutual information metric to quantify usage of context by model architectures that can use it at translation time. |
| Outcome: | The proposed method increases context usage and improves translation quality according to BLEU and COMET metrics. |
Training Neural Response Selection for Task-Oriented Dialogue Systems (P19-1)
Copied to clipboard
Matthew Henderson, Ivan Vulić, Daniela Gerz, Iñigo Casanueva, Paweł Budzianowski, Sam Coope, Georgios Spithourakis, Tsung-Hsien Wen, Nikola Mrkšić, Pei-Hao Su
| Challenge: | Despite their popularity, retrieval-based models have had modest impact on task-oriented dialogue systems . main obstacle to their application is the low-data regime of most task-orientated dialogue tasks . e-commerce, banking, and other domains are applications of retrieval models . |
| Approach: | They propose a method which pretrains a retrieval-based model on large general-domain conversational corpora and fine-tunes it for the target dialogue domain. |
| Outcome: | The proposed method is evaluated on five diverse domains, ranging from e-commerce to banking. |
All That Glitters is Not Gold: Improving Robust Retrieval-Augmented Language Models with Fact-Centric Preference Alignment (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to learn adaptive retrieval for noisy documents lack prior filtering and may lead to the loss of crucial information. |
| Approach: | They propose a method to improve retrieval performance without prior filtering . they use LLMs self-generated synthetic data as training data without manual annotation . |
| Outcome: | The proposed method performs positive document mining based on factual consistency and uses LLMs self-generated synthetic data as training data without manual annotation. |
Two Birds One Stone: Dynamic Ensemble for OOD Intent Classification (2023.acl-long)
Copied to clipboard
| Challenge: | Out-of-domain (OOD) intent classification is an active field of natural language understanding . previous studies have suggested that PTMs would be "overthinking" the semantic features of the sample in the open-world scenario . |
| Approach: | They propose a method that allows the model to decide whether to make a decision on OOD classification early during inference. |
| Outcome: | The proposed method can improve inference speed and achieve significant performance improvements. |
Efficient Training for Cross-lingual Speech Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Currently, large language models (LLMs) focus on the text modality, making speech modeling difficult. |
| Approach: | They propose a cross-lingual speech language model that trains on discrete speech tokens to achieve cross-modal and cross-linguistic alignment through continual pre-training. |
| Outcome: | The proposed method achieves cross-modal and cross-lingual alignment through continual pre-training. |
SOTOPIA-π: Interactive Learning of Socially Intelligent Language Agents (2024.acl-long)
Copied to clipboard
Ruiyi Wang, Haofei Yu, Wenxin Zhang, Zhengyang Qi, Maarten Sap, Yonatan Bisk, Graham Neubig, Hao Zhu
| Challenge: | Existing studies on building language agents have not addressed this social learning gap. |
| Approach: | They propose an interactive learning method that improves the social intelligence of language agents by using behavior cloning and self-reinforcement based training on filtered social interaction data. |
| Outcome: | The proposed method allows a 7B LLM to reach the social goal completion ability of an expert model (GPT-4-based agent) without the loss of more generic abilities, such as the ability to answer knowledge-based questions. |
Improve LLM-as-a-Judge Ability as a General Ability (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies focus on generative judges, but only on their judge ability. |
| Approach: | They propose a method that leverages the generative and reasoning capabilities of large language models to evaluate LLM responses across diverse scenarios, providing accurate preference signals. |
| Outcome: | The proposed model performs on RewardBench with only 2% to 40% of the data required by other training frameworks. |
Make Some Noise: Unlocking Language Model Parallel Inference Capability through Noisy Training (2024.emnlp-main)
Copied to clipboard
Yixuan Wang, Xianzhen Luo, Fuxuan Wei, Yijun Liu, Qingfu Zhu, Xuanyu Zhang, Qing Yang, Dongliang Xu, Wanxiang Che
| Challenge: | Existing speculative decoding methods require additional model structure and training processes to assist the model for draft token generation. |
| Approach: | They propose a make some noise training framework that introduces some noise at the input for the model to learn the denoising task. |
| Outcome: | The proposed model improves inference speed by 2.3-2.7x times without compromising model performance. |
Improving the robustness of NLI models with minimax training (2023.acl-long)
Copied to clipboard
| Challenge: | Experimental results show that our method consistently outperforms other robustness enhancement techniques on out-of-distribution adversarial test sets, while maintaining high in-distance accuracy. |
| Approach: | They propose a minimax objective between a learner model being trained for the task and an auxiliary model aiming to maximize the learner's loss by up-weighting underrepresented "hard" examples with patterns that contradict the shortcuts learned from the prevailing "easy" examples. |
| Outcome: | The proposed method outperforms other robustness enhancement techniques on out-of-distribution adversarial test sets while maintaining high in-distance accuracy. |
Exploring Reversal Mathematical Reasoning Ability for Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have been a success in the wide range of natural language understanding and reasoning tasks. |
| Approach: | They propose a training method to improve general and reversal reasoning abilities by using a reversed dataset. |
| Outcome: | The proposed method improves general and reversal reasoning abilities and alleviates the reverse curse. |
Step-by-Step Mastery: Enhancing Soft Constraint Following Ability of Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | In real-world scenarios, user instructions often contain soft constraints, which are semantically related and cannot be rule-based verified, posing challenges for large language models. |
| Approach: | They propose a pipeline to construct datasets with high-quality outputs for instructions containing soft constraints automatically and use Direct Preference Optimization (DPO) as the training method. |
| Outcome: | The proposed model improves the LLMs' soft constraint following ability by using direct preference optimization (DPO) and constraint quantity. |
Decoupled Proxy Alignment: Mitigating Language Prior Conflict for Multimodal Alignment in MLLMs (2025.findings-emnlp)
Copied to clipboard
Chenkun Tan, Pengyu Wang, Shaojun Zhou, Botian Jiang, Zhaowei Li, Dong Zhang, Xinghao Wang, Yaqian Zhou, Xipeng Qiu
| Challenge: | Recent advances in multimodal large language models focus on improving performance . however, language prior conflict leads to suboptimal vision-language alignment . |
| Approach: | They propose a method to decouple the alignment process from language prior interference . they use a proxy LLM to detach from language interference during pretraining . |
| Outcome: | The proposed method improves training performance and generalizes training data. |
Transfer Fine-tuning for Quality Estimation of Text Simplification (2024.lrec-main)
Copied to clipboard
| Challenge: | Experimental results show that quality estimation of text simplification models can be improved on a small labeled corpus. |
| Approach: | They propose a method to train quality estimation of text simplification on a small-scale labeled corpus prior to fine-tuning pre-trained language models. |
| Outcome: | The proposed method improves quality estimation of text simplification on a small-scale labeled corpus. |
KASER: Knowledge-Aligned Student Error Simulator for Open-Ended Coding Tasks (2026.acl-long)
Copied to clipboard
| Challenge: | coding tasks that provide detailed insights into student knowledge are challenging to train . open-ended tasks often suffer from mode collapse and fail to capture student errors . |
| Approach: | They propose a method that aligns errors with student knowledge by using a hybrid reward system. |
| Outcome: | The proposed method outperforms baselines on code and error prediction and error coverage and simulated code diversity on two real-world datasets. |
Outcome Accuracy is Not Enough: Aligning the Reasoning Process of Reward Models (2026.acl-long)
Copied to clipboard
Binghai Wang, Yantao Liu, Yuxuan Liu, Tianyi Tang, Shenzhi Wang, Chang Gao, Chujie Zheng, Yichang Zhang, Le Yu, Shixuan Liu, Tao Gui, Qi Zhang, Xuanjing Huang, Bowen Yu, Fei Huang, Junyang Lin
| Challenge: | Recent studies observe a phenomenon where reward models achieve high accuracy on static datasets but fail to generalize effectively during RLHF. |
| Approach: | They propose a method that combines rationale consistency with outcome accuracy to improve performance on RM-Bench and JudgeBench. |
| Outcome: | The proposed method surpasses baselines on RM-Bench and JudgeBench by an average of 5% and improves creative writing tasks by 7%. |