Anishinaabemowin (Ojibwe) AI Translation Benchmark Dataset

This dataset contains the raw outputs and metadata from a two-round benchmark study evaluating nine large language models (LLMs) on English-to-Anishinaabemowin (Ojibwe) translation tasks. Five open-source models were tested via Groq.com (January 29, 2026) and four commercial platforms were tested directly (February 12, 2026), using three standardized prompts selected to probe distinct morphological features of Anishinaabemowin. The evaluation framework is adapted from the Multidimensional Quality Metrics (MQM) methodology and implemented in QualCoder qualitative analysis software, with the goal of identifying open-source models suitable for fine-tuning in support of Indigenous language revitalization. The dataset is intended as a replicable methodological baseline for assessing LLM suitability in low-resource and endangered language contexts.