pith. sign in

arxiv: 2606.18389 · v1 · pith:VXAM2E2Vnew · submitted 2026-06-16 · 💻 cs.CL

Want Better Synthetic Data? Steer It: Activation Steering for Low-Resource Language Generation

Jan Cegin , Daniil Gurgurov , Yusser Al Ghussin , Simon Ostermann This is my paper

Pith reviewed 2026-06-27 00:43 UTC · model grok-4.3

classification 💻 cs.CL
keywords activation steeringsynthetic data generationlow-resource languageslanguage modelsdata diversitydownstream performancesentiment classificationtopic classification
0
0 comments X

The pith

Steering activations in early layers of LLMs produces more diverse synthetic data for low-resource languages and often improves downstream classifier performance.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests activation steering as a way to generate synthetic training data for languages that have little natural text, comparing it to standard few-shot prompting. It defines two steering approaches: one that shifts the model toward a target language's identity through activation contrasts, and another that favors human-written text over back-translated text. Experiments across four models and eleven languages generate sentiment and topic data, then measure both lexical diversity and the accuracy of small classifiers trained on the output. Results indicate that early-layer steering raises diversity and frequently lifts task performance, with the largest gains appearing for the lowest-resource languages.

Core claim

Deriving steering vectors from language-identity contrasts and from human-versus-backtranslated text contrasts, then adding those vectors to hidden states during generation, yields synthetic datasets whose diversity exceeds that of non-steered baselines and whose downstream classifiers often reach higher accuracy, especially when steering is applied at early layers.

What carries the argument

Steering vectors computed from activation differences in language-identity and quality contrasts, added to the model's hidden states at chosen layers during text generation.

If this is right

  • Early-layer steering consistently raises the diversity of the generated sentences.
  • Downstream classifiers trained on the steered data often reach higher accuracy than those trained on non-steered data.
  • The gains appear most clearly for languages with the least available training data.
  • The benefit holds in both zero-shot and few-shot generation settings across four open-source models and eleven typologically diverse languages.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The method could lower inference cost if it reduces the number of in-context examples needed.
  • The same contrast-based steering approach might extend to generating data for other scarce domains such as technical or domain-specific text.
  • Future experiments could test whether the steering vectors transfer across different generation tasks or model families.

Load-bearing premise

The steering vectors built from language contrasts and human-versus-backtranslated contrasts will shift generated text toward higher-quality examples of the target language without narrowing coverage of its natural distribution or introducing new biases.

What would settle it

A head-to-head comparison in which classifiers trained on steered synthetic data achieve equal or lower accuracy than classifiers trained on ordinary few-shot data, or in which diversity metrics show no gain, would falsify the central claim.

Figures

Figures reproduced from arXiv: 2606.18389 by Daniil Gurgurov, Jan Cegin, Simon Ostermann, Yusser Al Ghussin.

Figure 1
Figure 1. Figure 1: Example showing how Quality steering vec￾tors are created. This example shows a contrastive col￾lection of activations for German, from which a steering vector is created. This steering vector is used before generating data from the target LLM and language. language. These are then often used for fine-tuning of downstream encoder models. Recent works have extended these approaches to low-resource languages… view at source ↗
Figure 2
Figure 2. Figure 2: Linear probing results for human-authored vs. backtranslated textual pairs aggregated over languages. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Cosine similarity between quality and language vectors for different LLMs used in this study. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
read the original abstract

Large language models (LLMs) have become an effective tool for synthetic data generation, including for low-resource languages, where generated data can improve downstream task performance. Current best-performing approaches typically rely on few-shot prompting with target-language examples, which increases inference costs and may reduce diversity through lexical anchoring. In this work, we investigate activation steering as an alternative for low-resource synthetic data generation. We study two steering strategies: Language Steering, which targets the linguistic identity of a language, and Quality Steering, which captures well-formedness by contrasting human-written and backtranslated text representations. We evaluate these methods across four open-source LLMs, multiple layers, and 11 typologically diverse languages by generating sentiment and topic classification data and finetuning smaller classifiers. Steering is applied in both zero-shot and few-shot prompting settings and compared against non-steered counterparts. Our results show that steering on early layers consistently improves the diversity of generated data while often yielding stronger downstream model performance, particularly for low-resource languages.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes activation steering as an alternative to few-shot prompting for synthetic data generation in low-resource languages. It defines Language Steering (via language-identity contrasts) and Quality Steering (via human-written vs. backtranslated text contrasts), applies them at various layers of four open-source LLMs across 11 typologically diverse languages, generates sentiment and topic classification data in zero- and few-shot settings, and evaluates the outputs by measuring diversity and downstream classifier performance after fine-tuning.

Significance. If the central empirical claim holds after addressing the noted gaps, the work offers a potentially lower-cost method for improving synthetic data diversity without lexical anchoring from few-shot examples, with particular relevance for low-resource settings. The multi-model and multi-language scope is a positive feature of the evaluation design.

major comments (2)
  1. [Methods] Methods section (steering vector construction): The central claim that early-layer steering improves diversity and downstream performance rests on the premise that the two contrast-derived vectors enhance data quality without introducing new biases or reducing target-language coverage. No explicit validation (e.g., distribution overlap metrics, artifact analysis, or coverage checks on generated outputs) is described for this premise, which is load-bearing for interpreting the results as higher-quality synthetic data.
  2. [Results] Results section: The abstract states that early-layer steering 'consistently improves' diversity and 'often' yields stronger downstream performance, yet the provided text supplies no quantitative effect sizes, error bars, exact layer ranges, or details on post-hoc layer selection. This absence prevents verification of whether the reported gains are robust or merely artifacts of selective reporting.
minor comments (1)
  1. [Abstract] Abstract: Including at least one concrete quantitative result (e.g., diversity delta or accuracy lift with layer index) would strengthen the summary of findings.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We provide point-by-point responses to the major comments and outline the revisions we will make.

read point-by-point responses

  1. Referee: [Methods] Methods section (steering vector construction): The central claim that early-layer steering improves diversity and downstream performance rests on the premise that the two contrast-derived vectors enhance data quality without introducing new biases or reducing target-language coverage. No explicit validation (e.g., distribution overlap metrics, artifact analysis, or coverage checks on generated outputs) is described for this premise, which is load-bearing for interpreting the results as higher-quality synthetic data.

    Authors: We agree that more direct validation of the steering vectors would strengthen the interpretation. The diversity and downstream performance metrics provide indirect evidence, as significant biases or coverage loss would likely manifest as poorer classifier results. To address this, we will add explicit checks including language identification accuracy on generated texts and token distribution comparisons in the revised manuscript. revision: yes

  2. Referee: [Results] Results section: The abstract states that early-layer steering 'consistently improves' diversity and 'often' yields stronger downstream performance, yet the provided text supplies no quantitative effect sizes, error bars, exact layer ranges, or details on post-hoc layer selection. This absence prevents verification of whether the reported gains are robust or merely artifacts of selective reporting.

    Authors: The results section contains per-layer, per-language tables with performance metrics. We will update the abstract to include specific quantitative details on the improvements observed and ensure error bars are clearly reported. Details on layer ranges (early layers 1-8 depending on model) and selection criteria are in the experimental setup; we will make these more prominent to avoid any perception of selective reporting. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical comparison of steered vs. baseline generation

full rationale

The paper is an empirical study that generates synthetic data with and without activation steering, then measures diversity and downstream classifier performance across models and languages. No equations, fitted parameters, or derivations are present that could reduce reported improvements to inputs by construction. Steering vectors are derived from explicit contrasts (language identity, human vs. backtranslated) and applied as an intervention; results are evaluated against non-steered controls rather than being tautological. Self-citations, if any, are not load-bearing for the central empirical claims. The work is self-contained against external benchmarks (downstream accuracy, diversity metrics).

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests on the domain assumption that LLM activations contain separable directions for language identity and text quality. No free parameters or invented entities are described in the abstract.

axioms (1)
  • domain assumption LLM hidden states contain directions that can be isolated to control language identity and output quality
    This is the core premise behind defining Language Steering and Quality Steering vectors.

pith-pipeline@v0.9.1-grok · 5715 in / 1327 out tokens · 48451 ms · 2026-06-27T00:43:52.929168+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

96 extracted references · 32 canonical work pages · 1 internal anchor

  1. [1]

    2022 , howpublished =

    TransformerLens , author =. 2022 , howpublished =

    2022

  2. [2]

    2023 , eprint=

    Steering Language Models With Activation Engineering , author=. 2023 , eprint=

    2023

  3. [3]

    Steering Llama 2 via Contrastive Activation Addition

    Rimsky, Nina and Gabrieli, Nick and Schulz, Julian and Tong, Meg and Hubinger, Evan and Turner, Alexander. Steering Llama 2 via Contrastive Activation Addition. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024. doi:10.18653/v1/2024.acl-long.828

    work page doi:10.18653/v1/2024.acl-long.828 2024

  4. [4]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Language-specific neurons: The key to multilingual capabilities in large language models , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  5. [5]

    Language Arithmetics: Towards Systematic Language Neuron Identification and Manipulation

    Gurgurov, Daniil and Trinley, Katharina and Al Ghussin, Yusser and Baeumel, Tanja and Genabith, Josef Van and Ostermann, Simon. Language Arithmetics: Towards Systematic Language Neuron Identification and Manipulation. Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of...

    work page doi:10.18653/v1/2025.ijcnlp-long.156 2025

  6. [6]

    2026 , eprint=

    DFKI-MLT at SemEval-2026 TASK 7: Steering Multilingual Models Towards Cultural Knowledge , author=. 2026 , eprint=

    2026

  7. [7]

    2026 , eprint=

    Multilingual Steering by Design: Multilingual Sparse Autoencoders and Principled Layer Selection , author=. 2026 , eprint=

    2026

  8. [8]

    2025 , eprint=

    Causal Language Control in Multilingual Transformers via Sparse Feature Steering , author=. 2025 , eprint=

    2025

  9. [9]

    Advances in Neural Information Processing Systems , volume=

    How do large language models handle multilingualism? , author=. Advances in Neural Information Processing Systems , volume=

  10. [10]

    Advances in Neural Information Processing Systems , volume=

    Refusal in language models is mediated by a single direction , author=. Advances in Neural Information Processing Systems , volume=

  11. [11]

    arXiv preprint arXiv:2312.06681 , year=

    Steering llama 2 via contrastive activation addition , author=. arXiv preprint arXiv:2312.06681 , year=

    Pith/arXiv arXiv

  12. [12]

    Advances in Neural Information Processing Systems , volume=

    Inference-time intervention: Eliciting truthful answers from a language model , author=. Advances in Neural Information Processing Systems , volume=

  13. [13]

    2024 , eprint=

    Style Vectors for Steering Generative Large Language Model , author=. 2024 , eprint=

    2024

  14. [14]

    The Twelfth International Conference on Learning Representations , year=

    Sparse Autoencoders Find Highly Interpretable Features in Language Models , author=. The Twelfth International Conference on Learning Representations , year=

  15. [15]

    2024 , url=

    Scaling and evaluating sparse autoencoders , author=. 2024 , url=

    2024

  16. [16]

    Forty-second International Conference on Machine Learning , year=

    AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders , author=. Forty-second International Conference on Machine Learning , year=

  17. [17]

    Manning and Christopher Potts , year=

    Zhengxuan Wu and Aryaman Arora and Zheng Wang and Atticus Geiger and Dan Jurafsky and Christopher D. Manning and Christopher Potts , year=. arXiv:2404.03592 , archivePrefix=

    arXiv

  18. [18]

    2023 , eprint=

    The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets , author=. 2023 , eprint=

    2023

  19. [19]

    arXiv preprint arXiv:2310.01405 , year=

    Representation engineering: A top-down approach to ai transparency , author=. arXiv preprint arXiv:2310.01405 , year=

    Pith/arXiv arXiv

  20. [20]

    2024 , eprint=

    Steering Language Models With Activation Engineering , author=. 2024 , eprint=

    2024

  21. [21]

    Unveiling Language-Specific Features in Large Language Models via Sparse Autoencoders

    Deng, Boyi and Wan, Yu and Yang, Baosong and Zhang, Yidan and Feng, Fuli. Unveiling Language-Specific Features in Large Language Models via Sparse Autoencoders. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025. doi:10.18653/v1/2025.acl-long.229

    work page doi:10.18653/v1/2025.acl-long.229 2025

  22. [22]

    Proceedings of the 58th annual meeting of the association for computational linguistics , pages=

    The state and fate of linguistic diversity and inclusion in the NLP world , author=. Proceedings of the 58th annual meeting of the association for computational linguistics , pages=

  23. [23]

    arXiv preprint arXiv:2604.14090 , year=

    From Weights to Activations: Is Steering the Next Frontier of Adaptation? , author=. arXiv preprint arXiv:2604.14090 , year=

    Pith/arXiv arXiv

  24. [24]

    arXiv preprint arXiv:2601.05062 , year=

    Compositional Steering of Large Language Models with Steering Tokens , author=. arXiv preprint arXiv:2601.05062 , year=

    Pith/arXiv arXiv

  25. [25]

    arXiv preprint arXiv:2601.08331 , year=

    CLaS-Bench: A Cross-Lingual Alignment and Steering Benchmark , author=. arXiv preprint arXiv:2601.08331 , year=

    arXiv

  26. [26]

    arXiv preprint arXiv:2604.03532 , year=

    LangFIR: Discovering Sparse Language-Specific Features from Monolingual Data for Language Steering , author=. arXiv preprint arXiv:2604.03532 , year=

    Pith/arXiv arXiv

  27. [27]

    Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles , year=

    Efficient Memory Management for Large Language Model Serving with PagedAttention , author=. Proceedings of the ACM SIGOPS 29th Symposium on Operating Systems Principles , year=

  28. [28]

    BERT : Pre-training of Deep Bidirectional Transformers for Language Understanding

    Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina. BERT : Pre-training of Deep Bidirectional Transformers for Language Understanding. Proceedings of the 2019 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). 2019. doi:10.18653/v...

    work page doi:10.18653/v1/n19-1423 2019

  29. [29]

    Do Llamas Work in E nglish? On the Latent Language of Multilingual Transformers

    Wendler, Chris and Veselovsky, Veniamin and Monea, Giovanni and West, Robert. Do Llamas Work in E nglish? On the Latent Language of Multilingual Transformers. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024. doi:10.18653/v1/2024.acl-long.820

    work page doi:10.18653/v1/2024.acl-long.820 2024

  30. [30]

    Gemma 3 , url=

    Gemma Team , year=. Gemma 3 , url=

  31. [31]

    2024 , url =

    Llama 3 Model Card , author=. 2024 , url =

    2024

  32. [32]

    Unsupervised Cross-lingual Representation Learning at Scale , journal =

    Alexis Conneau and Kartikay Khandelwal and Naman Goyal and Vishrav Chaudhary and Guillaume Wenzek and Francisco Guzm. Unsupervised Cross-lingual Representation Learning at Scale , journal =. 2019 , url =. 1911.02116 , timestamp =

    Pith/arXiv arXiv 2019

  33. [33]

    Cross-lingual Transfer Learning with P ersian

    Mollanorozy, Sepideh and Tanti, Marc and Nissim, Malvina. Cross-lingual Transfer Learning with P ersian. Proceedings of the 5th Workshop on Research in Computational Linguistic Typology and Multilingual NLP. 2023. doi:10.18653/v1/2023.sigtyp-1.9

    work page doi:10.18653/v1/2023.sigtyp-1.9 2023

  34. [34]

    G r E m LI n: A Repository of Green Baseline Embeddings for 87 Low-Resource Languages Injected with Multilingual Graph Knowledge

    Gurgurov, Daniil and Kumar, Rishu and Ostermann, Simon. G r E m LI n: A Repository of Green Baseline Embeddings for 87 Low-Resource Languages Injected with Multilingual Graph Knowledge. Findings of the Association for Computational Linguistics: NAACL 2025. 2025

    2025

  35. [35]

    and Mao, Yanke and Gao, Haonan and Lee, En-Shiun Annie

    Adelani, David Ifeoluwa and Liu, Hannah and Shen, Xiaoyu and Vassilyev, Nikita and Alabi, Jesujoba O. and Mao, Yanke and Gao, Haonan and Lee, En-Shiun Annie. SIB -200: A Simple, Inclusive, and Big Evaluation Dataset for Topic Classification in 200+ Languages and Dialects. Proceedings of the 18th Conference of the European Chapter of the Association for Co...

    2024

  36. [36]

    MASSIVE : A 1 M -Example Multilingual Natural Language Understanding Dataset with 51 Typologically-Diverse Languages

    FitzGerald, Jack and Hench, Christopher and Peris, Charith and Mackie, Scott and Rottmann, Kay and Sanchez, Ana and Nash, Aaron and Urbach, Liam and Kakarala, Vishesh and Singh, Richa and Ranganath, Swetha and Crist, Laurie and Britan, Misha and Leeuwis, Wouter and Tur, Gokhan and Natarajan, Prem. MASSIVE : A 1 M -Example Multilingual Natural Language Und...

    work page doi:10.18653/v1/2023.acl-long.235 2023

  37. [37]

    Aho and Jeffrey D

    Alfred V. Aho and Jeffrey D. Ullman , title =. 1972

    1972

  38. [38]

    Publications Manual , year = "1983", publisher =

    1983

  39. [39]

    Chandra and Dexter C

    Ashok K. Chandra and Dexter C. Kozen and Larry J. Stockmeyer , year = "1981", title =. doi:10.1145/322234.322243

    work page doi:10.1145/322234.322243 1981

  40. [40]

    Scalable training of

    Andrew, Galen and Gao, Jianfeng , booktitle=. Scalable training of

  41. [41]

    Dan Gusfield , title =. 1997

    1997

  42. [42]

    Tetreault , title =

    Mohammad Sadegh Rasooli and Joel R. Tetreault , title =. Computing Research Repository , volume =. 2015 , url =

    2015

  43. [43]

    A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =

    Ando, Rie Kubota and Zhang, Tong , Issn =. A Framework for Learning Predictive Structures from Multiple Tasks and Unlabeled Data , Volume =. Journal of Machine Learning Research , Month = dec, Numpages =

  44. [44]

    Findings of the Association for Computational Linguistics: NAACL 2024 , pages=

    CoDa: Constrained Generation based Data Augmentation for Low-Resource NLP , author=. Findings of the Association for Computational Linguistics: NAACL 2024 , pages=

    2024

  45. [45]

    2024 , eprint=

    Use Random Selection for Now: Investigation of Few-Shot Selection Strategies in LLM-based Text Augmentation for Classification , author=. 2024 , eprint=

    2024

  46. [46]

    2023 , eprint=

    ZeroShotDataAug: Generating and Augmenting Training Data with ChatGPT , author=. 2023 , eprint=

    2023

  47. [47]

    People Make Better Edits: Measuring the Efficacy of LLM -Generated Counterfactually Augmented Data for Harmful Language Detection

    Sen, Indira and Assenmacher, Dennis and Samory, Mattia and Augenstein, Isabelle and Aalst, Wil and Wagner, Claudia. People Make Better Edits: Measuring the Efficacy of LLM -Generated Counterfactually Augmented Data for Harmful Language Detection. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023. doi:10.18653/v1/...

    work page doi:10.18653/v1/2023.emnlp-main.649 2023

  48. [48]

    Effects of diversity incentives on sample diversity and downstream model performance in LLM -based text augmentation

    Cegin, Jan and Pecher, Branislav and Simko, Jakub and Srba, Ivan and Bielikova, Maria and Brusilovsky, Peter. Effects of diversity incentives on sample diversity and downstream model performance in LLM -based text augmentation. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024. doi:10.186...

    work page doi:10.18653/v1/2024.acl-long.710 2024

  49. [49]

    2023 , eprint=

    AugGPT: Leveraging ChatGPT for Text Data Augmentation , author=. 2023 , eprint=

    2023

  50. [50]

    2025 , eprint=

    Enhancing NER Performance in Low-Resource Pakistani Languages using Cross-Lingual Data Augmentation , author=. 2025 , eprint=

    2025

  51. [51]

    and Gangal, Varun and Wei, Jason and Chandar, Sarath and Vosoughi, Soroush and Mitamura, Teruko and Hovy, Eduard

    Feng, Steven Y. and Gangal, Varun and Wei, Jason and Chandar, Sarath and Vosoughi, Soroush and Mitamura, Teruko and Hovy, Eduard. A Survey of Data Augmentation Approaches for NLP. Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021. 2021. doi:10.18653/v1/2021.findings-acl.84

    work page doi:10.18653/v1/2021.findings-acl.84 2021

  52. [52]

    arXiv preprint arXiv:2501.19314 , year=

    An Efficient Approach for Machine Translation on Low-resource Languages: A Case Study in Vietnamese-Chinese , author=. arXiv preprint arXiv:2501.19314 , year=

    arXiv

  53. [53]

    2024 6th International Conference on Natural Language Processing (ICNLP) , pages=

    Generative-Adversarial Networks for Low-Resource Language Data Augmentation in Machine Translation , author=. 2024 6th International Conference on Natural Language Processing (ICNLP) , pages=. 2024 , organization=

    2024

  54. [54]

    Synthetic Data Generation for Multilingual Domain-Adaptable Question Answering Systems

    Kramchaninova, Alina and Defauw, Arne. Synthetic Data Generation for Multilingual Domain-Adaptable Question Answering Systems. Proceedings of the 23rd Annual Conference of the European Association for Machine Translation. 2022

    2022

  55. [55]

    GeMQuAD: Generating Multilingual Question Answering Datasets from Large Language Models using Few Shot Learning , author=. 2023

    2023

  56. [56]

    arXiv preprint arXiv:2502.15419 , year=

    Beyond Translation: LLM-Based Data Generation for Multilingual Fact-Checking , author=. arXiv preprint arXiv:2502.15419 , year=

    arXiv

  57. [57]

    M ul DA : A Multilingual Data Augmentation Framework for Low-Resource Cross-Lingual NER

    Liu, Linlin and Ding, Bosheng and Bing, Lidong and Joty, Shafiq and Si, Luo and Miao, Chunyan. M ul DA : A Multilingual Data Augmentation Framework for Low-Resource Cross-Lingual NER. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1:...

    work page doi:10.18653/v1/2021.acl-long.453 2021

  58. [58]

    Jetsons at the F in NLP -2023: Using Synthetic Data and Transfer Learning for Multilingual ESG Issue Classification

    Glenn, Parker and Gon, Alolika and Kohli, Nikhil and Zha, Sihan and Dakle, Parag Pravin and Raghavan, Preethi. Jetsons at the F in NLP -2023: Using Synthetic Data and Transfer Learning for Multilingual ESG Issue Classification. Proceedings of the Fifth Workshop on Financial Technology and Natural Language Processing and the Second Multimodal AI For Financ...

    2023

  59. [59]

    SRL-ACO: A text augmentation framework based on semantic role labeling and ant colony optimization , journal =

    Aytuğ Onan , keywords =. SRL-ACO: A text augmentation framework based on semantic role labeling and ant colony optimization , journal =. 2023 , issn =. doi:https://doi.org/10.1016/j.jksuci.2023.101611 , url =

    work page doi:10.1016/j.jksuci.2023.101611 2023

  60. [60]

    Is C hat GPT the ultimate Data Augmentation Algorithm?

    Piedboeuf, Fr \'e d \'e ric and Langlais, Philippe. Is C hat GPT the ultimate Data Augmentation Algorithm?. Findings of the Association for Computational Linguistics: EMNLP 2023. 2023. doi:10.18653/v1/2023.findings-emnlp.1044

    work page doi:10.18653/v1/2023.findings-emnlp.1044 2023

  61. [61]

    2024 , isbn =

    Liu, Qijiong and Chen, Nuo and Sakai, Tetsuya and Wu, Xiao-Ming , title =. 2024 , isbn =. doi:10.1145/3616855.3635845 , booktitle =

    work page doi:10.1145/3616855.3635845 2024

  62. [62]

    GPT 3 M ix: Leveraging Large-scale Language Models for Text Augmentation

    Yoo, Kang Min and Park, Dongju and Kang, Jaewook and Lee, Sang-Woo and Park, Woomyoung. GPT 3 M ix: Leveraging Large-scale Language Models for Text Augmentation. Findings of the Association for Computational Linguistics: EMNLP 2021. 2021. doi:10.18653/v1/2021.findings-emnlp.192

    work page doi:10.18653/v1/2021.findings-emnlp.192 2021

  63. [63]

    Data Augmentation for Intent Classification with Off-the-shelf Large Language Models

    Sahu, Gaurav and Rodriguez, Pau and Laradji, Issam and Atighehchian, Parmida and Vazquez, David and Bahdanau, Dzmitry. Data Augmentation for Intent Classification with Off-the-shelf Large Language Models. Proceedings of the 4th Workshop on NLP for Conversational AI. 2022. doi:10.18653/v1/2022.nlp4convai-1.5

    work page doi:10.18653/v1/2022.nlp4convai-1.5 2022

  64. [64]

    2023 , eprint=

    Using GPT-4 to Augment Unbalanced Data for Automatic Scoring , author=. 2023 , eprint=

    2023

  65. [65]

    DALE: Generative Data Augmentation for Low-Resource Legal NLP

    Sreyan Ghosh and Chandra Kiran Evuru and Sonal Kumar and S Ramaneswaran and S Sakshi and Utkarsh Tyagi and Dinesh Manocha. DALE: Generative Data Augmentation for Low-Resource Legal NLP. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023

    2023

  66. [66]

    Semi-automatic generation of multilingual datasets for stance detection in Twitter , journal =

    Elena Zotova and Rodrigo Agerri and German Rigau , keywords =. Semi-automatic generation of multilingual datasets for stance detection in Twitter , journal =. 2021 , issn =. doi:https://doi.org/10.1016/j.eswa.2020.114547 , url =

    work page doi:10.1016/j.eswa.2020.114547 2021

  67. [67]

    2025 , eprint=

    Synthetic Data Generation for Culturally Nuanced Commonsense Reasoning in Low-Resource Languages , author=. 2025 , eprint=

    2025

  68. [68]

    LL atrieval: LLM -Verified Retrieval for Verifiable Generation

    Li, Xiaonan and Zhu, Changtai and Li, Linyang and Yin, Zhangyue and Sun, Tianxiang and Qiu, Xipeng. LL atrieval: LLM -Verified Retrieval for Verifiable Generation. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024. doi:10.18653/v1/202...

    work page doi:10.18653/v1/2024.naacl-long.305 2024

  69. [69]

    2025 , eprint=

    Prompt Selection Matters: Enhancing Text Annotations for Social Sciences with Large Language Models , author=. 2025 , eprint=

    2025

  70. [70]

    2024 , issue_date =

    Gu, Xu and Chen, Xiaoliang and Lu, Peng and Li, Zonggen and Du, Yajun and Li, Xianyong , title =. 2024 , issue_date =. doi:10.1016/j.engappai.2024.107907 , journal =

    work page doi:10.1016/j.engappai.2024.107907 2024

  71. [71]

    First International Workshop on Linked Science 2011-In conjunction with the International Semantic Web Conference (ISWC 2011) , year=

    Glottolog/Langdoc: Defining dialects, languages, and language families as collections of resources , author=. First International Workshop on Linked Science 2011-In conjunction with the International Semantic Web Conference (ISWC 2011) , year=

    2011

  72. [72]

    The annals of mathematical statistics , pages=

    On a test of whether one of two random variables is stochastically larger than the other , author=. The annals of mathematical statistics , pages=. 1947 , publisher=

    1947

  73. [73]

    2024 , eprint=

    The Linear Representation Hypothesis and the Geometry of Large Language Models , author=. 2024 , eprint=

    2024

  74. [74]

    C hat GPT to Replace Crowdsourcing of Paraphrases for Intent Classification: Higher Diversity and Comparable Model Robustness

    Cegin, Jan and Simko, Jakub and Brusilovsky, Peter. C hat GPT to Replace Crowdsourcing of Paraphrases for Intent Classification: Higher Diversity and Comparable Model Robustness. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023. doi:10.18653/v1/2023.emnlp-main.117

    work page doi:10.18653/v1/2023.emnlp-main.117 2023

  75. [75]

    arXiv preprint arXiv:2402.13116 , year=

    A survey on knowledge distillation of large language models , author=. arXiv preprint arXiv:2402.13116 , year=

    Pith/arXiv arXiv

  76. [76]

    Adapting Multilingual LLM s to Low-Resource Languages with Knowledge Graphs via Adapters

    Gurgurov, Daniil and Hartmann, Mareike and Ostermann, Simon. Adapting Multilingual LLM s to Low-Resource Languages with Knowledge Graphs via Adapters. Proceedings of the 1st Workshop on Knowledge Graphs and Large Language Models (KaLLM 2024). 2024. doi:10.18653/v1/2024.kallm-1.7

    work page doi:10.18653/v1/2024.kallm-1.7 2024

  77. [77]

    arXiv preprint arXiv:2502.10140 , year=

    Small Models, Big Impact: Efficient Corpus and Graph-Based Adaptation of Small Multilingual Language Models for Low-Resource Languages , author=. arXiv preprint arXiv:2502.10140 , year=

    arXiv

  78. [78]

    LLM s vs Established Text Augmentation Techniques for Classification: When do the Benefits Outweight the Costs?

    Cegin, Jan and Simko, Jakub and Brusilovsky, Peter. LLM s vs Established Text Augmentation Techniques for Classification: When do the Benefits Outweight the Costs?. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025

    2025

  79. [79]

    Language-agnostic BERT Sentence Embedding

    Feng, Fangxiaoyu and Yang, Yinfei and Cer, Daniel and Arivazhagan, Naveen and Wang, Wei. Language-agnostic BERT Sentence Embedding. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022. doi:10.18653/v1/2022.acl-long.62

    work page doi:10.18653/v1/2022.acl-long.62 2022

  80. [80]

    Data Mining , booktitle=

    Rajaraman, Anand and Ullman, Jeffrey David , year=. Data Mining , booktitle=

Showing first 80 references.