Papers with training approach
Generation-Distillation for Efficient Natural Language Understanding in Low-Data Settings (D19-61)
Copied to clipboard
| Challenge: | Recent research points to knowledge distillation as a potential solution for NLU tasks. |
| Approach: | They propose a training approach that distills large finetuned LMs into a small network using unlabeled training examples. |
| Outcome: | The proposed approach outperforms BERT training approaches while using 300 times fewer parameters. |
Training for Diversity in Image Paragraph Captioning (D18-1)
Copied to clipboard
| Challenge: | Existing image captioning models have a lack of diversity between sentences . current models have limited their effectiveness due to repetitive paragraphs . |
| Approach: | They propose to apply sequence-level training to image paragraph captioning models . they find that standard self-critical training produces poor results . |
| Outcome: | The proposed training improves on the Visual Genome dataset with no architectural changes. |
2*n is better than n2: Decomposing Event Coreference Resolution into Two Tractable Problems (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for training coreference systems sample from a largely skewed distribution, making it difficult to learn coreference beyond surface matching. |
| Approach: | They propose a heuristic to efficiently filter out a large number of non-coreferent pairs and a training approach on a balanced set of coreferent and non- coreferente mention pairs. |
| Outcome: | The proposed approach significantly reduces compute requirements on two popular ECR datasets while reducing the computational complexity. |
An Empirical Investigation of Global and Local Normalization for Recurrent Neural Sequence Models Using a Continuous Relaxation to Beam Search (N19-1)
Copied to clipboard
| Challenge: | Neural encoder-decoder models have been successful at a variety of NLP tasks, including machine translation, parsing, and dialog generation. |
| Approach: | They propose a method for search-aware training via a continuous relaxation of beam search to enable global normalization. |
| Outcome: | The proposed approach is able to train globally normalized recurrent sequence models through simple backpropagation. |
HistAlign: Improving Context Dependency in Language Generation by Aligning with History (2023.emnlp-main)
Copied to clipboard
| Challenge: | Language models (LMs) can generate hallucinations and incoherent outputs due to their weak context dependency. |
| Approach: | They propose a training approach to ensure good cache alignment so that the model receives useful signals from the history. |
| Outcome: | The proposed approach improves text coherence and faithfulness on diverse language generation tasks. |
What do we expect from Multiple-choice QA Systems? (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Recent work has shown that good performance on a dataset might not correlate well with human’s expectations from models that “understand” language. |
| Approach: | They propose to train a top performing multiple choice question answering model against expectations from models that "understand" language. |
| Outcome: | The proposed training paradigm leads to a model that performs on par with the original model while better satisfying our expectations. |
Entity Linking via Explicit Mention-Mention Coreference Modeling (2022.naacl-main)
Copied to clipboard
| Challenge: | Using a learning approach for entity mentions is a key component of modern entity linking systems for both candidate generation and making linking predictions. |
| Approach: | They propose a training approach that builds minimum spanning arborescences over mentions and entities to explicitly model mention coreference relationships. |
| Outcome: | The proposed approach improves candidate generation recall and link accuracy on the biomedical dataset and on MedMentions, setting a new SOTA result in linking accuracy. |
Training Language Models with Memory Augmentation (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for training memory-augmented language models only introduce mem-ories at testing time or represent them using a separately trained encoder. |
| Approach: | They propose a training approach that directly takes in-batch examples as accessible memory and new methods for memory construction and data batching that are used for adapting to different sets of memories at testing time. |
| Outcome: | The proposed approach reduces perplexity from 18.70 to 15.37 on multiple language modeling and machine translation benchmarks. |
Leveraging Discourse Rewards for Document-Level Neural Machine Translation (2020.coling-main)
Copied to clipboard
| Challenge: | Document-level machine translation models are often not trained to explicitly ensure discourse quality. |
| Approach: | They propose a method that explicitly optimizes lexical cohesion and coherence metrics by using a reinforcement learning objective. |
| Outcome: | The proposed approach improves document translations over four different languages and three translation domains while maintaining faithfulness to the reference translation. |
SoRFT: Issue Resolving with Subtask-oriented Reinforced Fine-Tuning (2025.acl-long)
Copied to clipboard
| Challenge: | Existing issue-resolving frameworks rely on commercial models, leading to high costs and privacy concerns. |
| Approach: | They propose a training approach to enhance issue resolving capability of LLMs by decomposing issue reasolving into subtasks. |
| Outcome: | The proposed approach improves issue-resolving performance and generalizes model . it is cost-effective and provides a cost-efficient alternative to commercial models . |
Meta-Reinforced Multi-Domain State Generator for Dialogue Systems (2020.acl-main)
Copied to clipboard
| Challenge: | Existing methods to train a multi-domain dialogue state tracker are lacking in accuracy. |
| Approach: | They propose a Meta-Reinforced Multi-Domain State Generator to train a DST meta-learning model with a few domains as source domains and a new domain as target domain. |
| Outcome: | The proposed system outperforms the traditional training approach with extremely little training data in target domain. |
Context Consistency between Training and Inference in Simultaneous Machine Translation (2024.acl-long)
Copied to clipboard
| Challenge: | Simultaneous machine translation (SiMT) aims to yield a partial translation with a monotonically growing source-side context. |
| Approach: | They propose a training approach that encourages consistent context usage between training and inference by optimizing translation quality and latency as bi-objectives and exposing the predictions to the model during the training. |
| Outcome: | The proposed system outperforms existing SiMT systems with context inconsistency for the first time. |
LegoMT2: Selective Asynchronous Sharded Data Parallel Training for Massive Neural Machine Translation (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to train a single model for massive languages have huge communication overheads and parameter interference. |
| Approach: | They propose an efficient training approach with an asymmetric multi-way model architecture for massive multilingual neural machine translation. |
| Outcome: | The proposed model is 16.2 faster than the distributed training method for M2M-100-12B while improving the translation performance by an average of 2.2 BLEU on Flores-101. |
Beneficial Reasoning Behaviors in Agentic Search and Effective Training Methods to Obtain Them (2026.findings-acl)
Copied to clipboard
| Challenge: | Agentic search requires large language models to perform multi-step searches to solve complex information needs. |
| Approach: | They propose a training approach that equips agentic search models with reasoning behaviors before reinforcement learning (RL) they compare successful and failed trajectories and propose supervised fine-tuning and standard RL . |
| Outcome: | The proposed approach outperforms direct RL by 37.2% on three web benchmarks and 6.2% on seven multi-hop QA benchmarks. |