\textit{Versteasch du mi?} Computational and Socio-Linguistic Perspectives on GenAI, LLMs, and Non-Standard Language

Dimensions overall score 4.0

• Large Language Models (LLMs) and generative AI are 'unfair' to less-spoken languages and deepen the digital language divide.limitation 95%
  Evidencepartial

  The design of Large Language Models (LLMs) and generative artificial intelligence (GenAI) has been shown to be “unfair” to less-spoken languages (Petrov et al., 2023) and to deepen the digital language divide (Bella et al., 2023).

  Implicationpartial

  Explicitly stated in the abstract with citations to supporting literature.

  Verificationpartial

  partial

• LLMs and GenAI technologies are made possible by and exacerbate historical processes of linguistic standardization rooted in European nationalist and colonial projects.limitation 90%
  Evidencepartial

  Critical sociolinguistic work has also argued that these technologies are not only made possible by prior socio-historical processes of linguistic standardisation, often grounded in European nationalist and colonial projects, but also exacerbate epistemologies of language as 'monolithic, monolingual, syntactically standardized systems of meaning'.

  Implicationpartial

  Directly stated in the abstract as a core argument from critical sociolinguistic work.

  Verificationpartial

  partial

• South Tyrolean dialects, despite being the primary medium of everyday interaction in South Tyrol, Italy, remain largely absent from LLM training data.limitation 85%
  Evidencepartial

  South Tyrolean dialect comprises non-standard forms of German widely used in informal communication in South Tyrol, Italy (Alber et al., 2024), yet remains largely absent from LLM training data despite being the primary medium of everyday interaction.

  Implicationpartial

  Directly stated in the analysis with a specific case study reference.

  Verificationpartial

  partial

• The orthographic diversity of Kurdish (Latin, Arabic, and Cyrillic scripts) creates significant barriers to cross-dialectal communication and pan-Kurdish linguistic unity.limitation 90%
  Evidencepartial

  This orthographic diversity, while enabling written expression within political boundaries, creates significant barriers to cross-dialectal communication, pan-Kurdish linguistic unity.

  Implicationpartial

  Directly stated in the analysis with a clear description of the barrier.

  Verificationpartial

  partial

• The MMLU benchmark for evaluating LLMs is highly US-centric and Western-focused, with a disproportionate number of questions on North America or Europe.limitation 95%
  Evidencepartial

  The MMLU benchmark is highly US-centric... with approximately 28% of the questions requiring specific knowledge of Western cultures and a staggering 84.9% of geographic questions focusing exclusively on North America or Europe (Faisal et al., 2025).

  Implicationpartial

  Explicitly stated with specific numeric evidence (84.9%).

  Verificationpartial

  partial

• Civil society organizations (CSOs) can act as algorithmic auditors through socio-pragmatic red-teaming to identify where models fail to respect local norms or enforce standardization.method 80%
  Evidencepartial

  CSOs can serve as algorithmic auditors, performing socio-pragmatic red-teaming to identify where models fail to respect local norms or inadvertently enforce standardization.

  Implicationpartial

  Directly stated as a proposed role and method, though presented as a potential strategy rather than a completed result.

  Verificationpartial

  partial

• Academic institutions and national language institutes are crucial for providing the sociolinguistic granularity needed to prevent the erasure of non-standard language varieties in AI development.method 85%
  Evidencepartial

  While Big Tech prioritizes computational scale, universities provide the sociolinguistic granularity necessary to prevent the erasure of non-standard varieties.

  Implicationpartial

  Directly stated as a comparative strength of these institutions versus Big Tech's focus on scale.

  Verificationpartial

  partial

• There is current development of models fine-tuned for automated transcription and subtitling of South Tyrolean audio into standard German writing.technical 90%
  Evidencepartial

  Models are currently being fine-tuned specifically for automated transcription and subtitling of audio(visual) material (Ducceschi and Franzini, 2025). The use case being addressed is that of transposing non-standard audio into standard German writing.

  Implicationpartial

  Directly stated as a current technical development with a specific use case and citation.

  Verificationpartial

  partial

Competitive landscape

Developing LLMs that can understand and process non-standard dialects to bridge the digital language divide.