Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Translation or Recitation? Calibrating Evaluation Scores for Machine Translation of Extremely Low-Resource Languages

Danlu Chen, Ka Sing He, Jiahe Tian, Chenghao Xiao, Zhaofeng Wu, Taylor Berg-Kirkpatrick, Freda Shi

Abstract

The landscape of extremely low-resource machine translation (MT) is characterized by perplexing variability in reported performance, often making results across different language pairs difficult to contextualize. For researchers focused on specific language groups -- such as ancient languages -- it is nearly impossible to determine if breakthroughs reported in other contexts (e.g., native African or American languages) result from superior methodologies or are merely artifacts of benchmark collection. To address this problem, we introduce the FRED Difficulty Metrics, which include the Fertility Ratio (F), Retrieval Proxy (R), Pre-training Exposure (E), and Corpus Diversity (D) and serve as dataset-intrinsic metrics to contextualize reported scores. These metrics reveal that a significant portion of result variability is explained by train-test overlap and pre-training exposure rather than model capability. Additionally, we identify that some languages -- particularly extinct and non-Latin indigenous languages -- suffer from poor tokenization coverage (high token fertility), highlighting a fundamental limitation of transferring models from high-resource languages that lack a shared vocabulary. By providing these indices alongside performance scores, we enable more transparent evaluation of cross-lingual transfer and provide a more reliable foundation for the XLR MT community.

Translation or Recitation? Calibrating Evaluation Scores for Machine Translation of Extremely Low-Resource Languages

Abstract

The landscape of extremely low-resource machine translation (MT) is characterized by perplexing variability in reported performance, often making results across different language pairs difficult to contextualize. For researchers focused on specific language groups -- such as ancient languages -- it is nearly impossible to determine if breakthroughs reported in other contexts (e.g., native African or American languages) result from superior methodologies or are merely artifacts of benchmark collection. To address this problem, we introduce the FRED Difficulty Metrics, which include the Fertility Ratio (F), Retrieval Proxy (R), Pre-training Exposure (E), and Corpus Diversity (D) and serve as dataset-intrinsic metrics to contextualize reported scores. These metrics reveal that a significant portion of result variability is explained by train-test overlap and pre-training exposure rather than model capability. Additionally, we identify that some languages -- particularly extinct and non-Latin indigenous languages -- suffer from poor tokenization coverage (high token fertility), highlighting a fundamental limitation of transferring models from high-resource languages that lack a shared vocabulary. By providing these indices alongside performance scores, we enable more transparent evaluation of cross-lingual transfer and provide a more reliable foundation for the XLR MT community.

Paper Structure

This paper contains 24 sections, 3 equations, 2 figures, 13 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Establishing High-Resource Reference Baselines
  4. FRED Difficulty Metric Definition
  5. Experiments and Analysis
  6. Dataset Collection and Analysis
  7. Interpreting Metrics Against Baselines
  8. High-resource baselines establish expected ranges.
  9. Not all XLR languages can match baseline diversity.
  10. Metric Correlation Analysis
  11. Specific Findings for Language Groups
  12. Conclusion
  13. Regression Correlation Analysis
  14. Tables
  15. Low resource Machine Translation survey
  16. ...and 9 more sections

Figures (2)

  • Figure 1: Performance distribution of extremely low-resource (XLR) languages in machine translation, shown as scattered outliers. The green line represents the mean performance of mBART when fine-tuned on varying sizes of training data from five high-resource languages (see Table \ref{['tb:high_res']} and \ref{['ch:high_res']} for details), with the shaded region indicating $\pm1$ standard deviation.
  • Figure 2: Scores of PBSMT versus R-BLEU by language groups.