Uncovering inequalities in new knowledge learning by large language models across different languages
Bibliographic record
Abstract
As large language models (LLMs) gradually demonstrate their potential to boost productivity and become integral tools for problem-solving in daily life worldwide, understanding the linguistic inequalities they introduce is becoming increasingly important. Prior research has primarily focused on static analyses of disparities in existing knowledge and capabilities of LLMs across languages. However, LLMs are continuously evolving, acquiring new knowledge to provide current, relevant responses and deliver precise, expert-level answers in specific domains. Investigating linguistic inequalities within this dynamic learning process is, therefore, also essential. In this paper, we explore inequalities in new knowledge learning by LLMs across different languages and four key dimensions: effectiveness, transferability, prioritization, and robustness. Through extensive experiments in both in-context learning and fine-tuning settings, with proprietary and open-source models, we reveal four key findings: 1) LLMs face greater challenges in efficiently and accurately learning new knowledge in lower-resource languages; 2) knowledge learned by LLMs tends to be more easily transferred to higher-resource languages than to lower-resource ones; 3) new knowledge in higher-resource languages is more likely to be retained and prioritized; and 4) LLMs are more robust against incorrect or misleading information in higher-resource languages. We further analyze the underlying causes of these inequalities from linguistic perspectives, pretraining characteristics, and tokenizer design, and propose a preliminary mitigation strategy through the lens of linguistic neurons. This work highlights the urgent need to recognize and address emerging linguistic inequalities in the development of LLMs.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.013 | 0.066 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.003 |
| Scholarly communication | 0.003 | 0.012 |
| Open science | 0.002 | 0.007 |
| Research integrity | 0.001 | 0.005 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".