EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Prompt-Alignment Transfer for Low-Resource Language Tasks Under Severe Domain Shift

Finn Frank

Journal of Social Sciences, Humanities and Innovation · 03.10.2026 · 2 (2) · 109-120

DOI: 10.66238/jsshi.v2i2.57 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Social Sciences, Humanities and Innovation, online erschienen am 03.10.2026. Verfasst von Finn Frank. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The proliferation of large language models has fundamentally transformed natural language processing, offering unprecedented capabilities in text understanding and generation. However, the equitable distribution of these advancements remains severely constrained by the persistent resource disparity among global languages. When low-resource language tasks are further complicated by severe domain shift, where the target application domain diverges significantly from the high-resource source training data, traditional cross-lingual transfer methodologies exhibit catastrophic performance degradation. This paper investigates a novel framework, termed Prompt-Alignment Transfer, designed to bridge the linguistic and domain gaps simultaneously. By projecting domain-agnostic prompts from a high-resource source language into an aligned latent space tailored for a low-resource target language, our methodology effectively mitigates the negative transfer phenomena commonly observed in conventional fine-tuning paradigms. The framework incorporates a dual-optimization strategy, simultaneously aligning cross-lingual representations while employing adversarial domain adaptation within the prompt continuous space. Extensive theoretical analysis and empirical evaluations across diverse linguistic families and specialized domains, such as medical and legal texts, demonstrate that Prompt-Alignment Transfer significantly outperforms existing zero-shot and few-shot cross-lingual baselines. The methodology establishes a new paradigm for deploying advanced natural language processing solutions in linguistically marginalized and domain-specific contexts, offering a robust pathway toward more inclusive and adaptable artificial intelligence systems.

Zitieren

APA

Frank, F. (2026). Prompt-Alignment Transfer for Low-Resource Language Tasks Under Severe Domain Shift. Journal of Social Sciences, Humanities and Innovation, 2(2), 109-120. https://doi.org/10.66238/jsshi.v2i2.57

Vancouver

Frank F. Prompt-Alignment Transfer for Low-Resource Language Tasks Under Severe Domain Shift. Journal of Social Sciences, Humanities and Innovation. 2026;2(2):109-120. doi:10.66238/jsshi.v2i2.57

BibTeX
@article{Frank2026,
  author = {Frank, Finn},
  title = {Prompt-Alignment Transfer for Low-Resource Language Tasks Under Severe Domain Shift},
  journal = {Journal of Social Sciences, Humanities and Innovation},
  year = {2026},
  volume = {2},
  number = {2},
  pages = {109--120},
  doi = {10.66238/jsshi.v2i2.57}
}
Kostenlos starten →