Papers with language model pretraining
English-Myanmar Supervised and Unsupervised NMT: NICT’s Machine Translation Systems at WAT-2019 (D19-52)
Copied to clipboard
| Challenge: | NICT participated in the 6th Workshop on Asian Translation (WAT-2019) shared translation task, specifically Myanmar (My) - English task in both translation directions. |
| Approach: | They present the participation of the NICT in the 6th Workshop on Asian Translation (WAT-2019) shared translation task, specifically Myanmar (Burmese) - English task in both translation directions. |
| Outcome: | The proposed systems perform the third in English-to-Myanmar and the second in Myanmar-to English according to BLEU score. |
Jeff Da at COIN - Shared Task: BIG MOOD: Relating Transformers to Explicit Commonsense Knowledge (D19-60)
Copied to clipboard
| Challenge: | Recent studies show that large-scale pre-training models can be effective for large datasets. |
| Approach: | They propose a method of integrating contextual embeddings with commonsense graph embeddINGs by preprocessing knowledge bases and aligning tokens between misaligned tokenization methods. |
| Outcome: | The proposed method achieves higher accuracy than BERT and scores highest without pretraining. |
VILA: Improving Structured Content Extraction from Scientific PDFs Using Visual Layout Groups (2022.tacl-1)
Copied to clipboard
| Challenge: | Recent work has improved extraction accuracy by incorporating elementary layout information, for example, each token’s 2D position on the page, into language model pretraining. |
| Approach: | They propose a method that explicitly models VIsual LAyout (VILA) groups, that is, text lines or text blocks, to further improve extraction accuracy. |
| Outcome: | The proposed methods show that inserting special tokens denoting layout group boundaries can lead to a 1.9% Macro F1 improvement in token classification. |
Entity Extraction in Low Resource Domains with Selective Pre-training of Large Language Models (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to perform named entity recognition (NER) on unlabeled data are difficult to obtain in low-resource domains. |
| Approach: | They propose ways to use unlabeled data for pretraining to improve performance in downstream tasks. |
| Outcome: | The proposed methods outperform models trained on unlabeled data on seven domains. |
K-hop neighbourhood regularization for few-shot learning on graphs: A case study of text classification (2023.eacl-main)
Copied to clipboard
| Challenge: | We show that few-sample word-document graphs can be used for improved learning in low-resource settings. |
| Approach: | They propose a method to utilize word-document graph properties for improved learning in low-resource settings by using a regularizer for heterogeneous graphs. |
| Outcome: | The proposed method outperforms a baseline TextGCN with 17% accuracy over eight languages while performing on par with the state-of-the-art models. |
Linguistic Knowledge and Transferability of Contextual Representations (N19-1)
Copied to clipboard
| Challenge: | Recent work has explored contextual word representations, which assign each word a vector that is a function of the entire input sequence. |
| Approach: | They compare pretrained word representations with 16 diverse probing tasks to examine their transferability. |
| Outcome: | The pretrained representations are successful across a diverse set of NLP tasks . the models are competitive with state-of-the-art models but fail on fine-grained tasks requiring fine-granular knowledge, the study finds . |
Pretraining Sentiment Classifiers with Unlabeled Dialog Data (P18-2)
Copied to clipboard
| Challenge: | Existing methods to train sentiment classifiers with unlabeled data are costly and time-consuming. |
| Approach: | They propose a conditional language model with unlabeled dialog data instead of a language model to pretrain sentiment classifiers. |
| Outcome: | The proposed strategy outperforms state-of-the-art methods with unlabeled dialog data and is simple but effective. |
Investigating Transferability in Pretrained Language Models (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Recent work on deep NLP models has centered on probing, a method that involves training classifiers for different tasks on model representations. |
| Approach: | They propose a method for determining the impact of each pretrained layer on transfer task performance by ablation. |
| Outcome: | The proposed method shows that pretraining models improve performance on downstream tasks . the results highlight the limitations of methods that operate on frozen models or single data samples. |
ChineseBERT: Chinese Pretraining Enhanced by Glyph and Pinyin Information (2021.acl-long)
Copied to clipboard
| Challenge: | ChineseBERT model incorporates glyph and pinyin information of Chinese characters into pretraining . proposed model achieves new performance boost over baseline models with fewer training steps . |
| Approach: | They propose a ChineseBERT model that incorporates glyph and pinyin information into pretraining . the glyph embedding is obtained based on different fonts of a character, and the pinyink embeddment characterizes the pronunciation of Chinese characters. |
| Outcome: | The proposed model achieves new performance boosts over baseline models with fewer training steps. |
LMSOC: An Approach for Socially Sensitive Pretraining (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Large-scale pretraining models have been shown to learn effective linguistic representations for many NLP tasks, but there are many real-world contextual aspects of language that current approaches do not capture. |
| Approach: | They propose to integrate speaker social context into the learned representations of large-scale language models by using graph representation learning algorithms and primed language model pretraining with these social context representations. |
| Outcome: | The proposed approach improves on geographically sensitive language modeling tasks by more than 100% relative lift on MRR compared to baselines. |
Contextual morphologically-guided tokenization for Latin encoder models (2026.eacl-long)
Copied to clipboard
| Challenge: | Existing tokenization methods focus on information-theoretical goals like high compression and low fertility rather than linguistic goals like morphological alignment. |
| Approach: | They propose to incorporate morphological knowledge into tokenization to improve both morphology and downstream performance. |
| Outcome: | The proposed tokenization improves overall performance on four downstream tasks. |
Beyond Random Sampling: Efficient Language Model Pretraining via Curriculum Learning (2026.eacl-long)
Copied to clipboard
| Challenge: | Curriculum learning has improved efficiency across machine learning domains, but remains underexplored for language model pretraining. |
| Approach: | They present a systematic investigation of curriculum learning in LLM pretraining . they use vanilla curriculum learning, pacing-based sampling, and interleaved curricula . |
| Outcome: | The proposed framework accelerates convergence in early and mid-training phases, reducing training steps by 18-45% to reach baseline performance. |
Let’s Stop Incorrect Comparisons in End-to-end Relation Extraction! (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing literature on Relation Extraction (RE) uses multiple evaluation setups to compare performance. |
| Approach: | They propose to quantify the most common comparison mistake and evaluate it leads to overestimating the final RE performance by around 5% on ACE05. |
| Outcome: | The proposed meta-analysis overestimates the final RE performance by around 5% on ACE05. |
Byte Pair Encoding is Suboptimal for Language Model Pretraining (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Subword tokenization is a popular language model that can be used to segment text. |
| Approach: | They analyze differences between byte-pair encoding (BPE) and unigram LM tokenization methods to find subword units that align more closely with morphology. |
| Outcome: | The proposed method recovers subword units that align more closely with morphology and avoids problems stemming from BPE’s greedy construction procedure. |
PreAlign: Boosting Cross-Lingual Transfer by Early Establishment of Multilingual Alignment (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large language models exhibit reasonable multilingual abilities, despite predominantly English-centric pretraining. |
| Approach: | They propose a framework that establishes multilingual alignment prior to language model pretraining and preserves this alignment using a code-switching strategy during pretraining. |
| Outcome: | Experiments in a synthetic English to English-Clone setting show that PreAlign outperforms standard multilingual joint training in language modeling, zero-shot cross-lingual transfer, and cross-linguistic knowledge application. |
Enhancing Distractor Generation for Multiple-Choice Questions with Retrieval Augmented Pretraining and Knowledge Graph Integration (2024.findings-acl)
Copied to clipboard
Han Cheng Yu, Yu An Shih, Kin Man Law, KaiYu Hsieh, Yu Chen Cheng, Hsin Chih Ho, Zih An Lin, Wen-Chuan Hsu, Yao-Chung Fan
| Challenge: | Existing LMs undergo task-agnostic pertaining, but task-specific pretraining has gained prominence. |
| Approach: | They propose retrieval augmented pretraining and task-specific pretraining for DG . they propose to refine language model pretraining to align it more closely with downstream task . |
| Outcome: | The proposed method improves the performance of multiple-choice questions by integrating knowledge graphs and language models. |
Neural Networks Against (and For) Self-Training: Classification with Small Labeled and Large Unlabeled Sets (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing models for text classification suffer from the semantic drift problem, which is a problem for self-training. |
| Approach: | They propose a semi-supervised text classifier based on self-training using one positive and one negative property of neural networks. |
| Outcome: | The proposed model outperforms ten baseline models in five benchmarks and is additive to language model pretraining. |
Towards A Unified View of Sparse Feed-Forward Network in Pretraining Large Language Model (2023.emnlp-main)
Copied to clipboard
| Challenge: | Large and sparse feed-forward layers (S-FFN) have proven effective in scaling up the model size for pretraining large language models. |
| Approach: | They compare S-FFN architectures for language modeling and compare their performance and efficiency . they found a simpler selection method that selects blocks through their mean aggregated hidden states . |
| Outcome: | The proposed model size and selection method achieve lower perplexity in language model pretraining compared to existing MoE architectures. |
Sequence Reducible Holdout Loss for Language Model Pretraining (2024.lrec-main)
Copied to clipboard
| Challenge: | Data selection techniques have shown empirical benefits in reducing the number of gradient steps to train neural models. |
| Approach: | They propose to modify an existing data selection technique to adapt it to the sequence losses typical in language modeling. |
| Outcome: | The proposed technique reduces the number of steps required to train neural models by 4.3% and improves generalization ability on out of domain datasets. |
Studying the Role of Input-Neighbor Overlap in Retrieval-Augmented Language Models Training Efficiency (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have shown that retrieval-augmented language models can perform better than larger models, but the optimal degree of overlap remains unclear. |
| Approach: | They investigate how varying levels of query–context overlap affect model performance . they specifically generate synthetic context through paraphrasing queries . |
| Outcome: | The proposed model can improve test-time perplexity and training time by 40% without compromising performance. |