Despite advancements in Large Language Models, reasoning benchmarks remain centered on high-resource languages, leaving languages like Vietnamese under-evaluated. In this study, we aim to address this gap by evaluating four models: PhoGPT (native), Vistral and VBD-Llama (adapted), and Llama-2 (English-centric), on commonsense reasoning and arithmetic reasoning. As Vietnamese benchmarks for these tasks are lacking, we adapt two analogy datasets from English to Vietnamese and construct two sequence datasets, ensuring a range of structural complexity and difficulty levels. We evaluate diverse prompting strategies, including Chain-of-Thought, role-playing guidance, cross-lingual prompting, and few-shot learning. Our results reveal a baseline proficiency in analogical and arithmetic reasoning among the models, with Vistral and Llama-2 outperforming other models in multiple tasks. The effects of Chain-of-Thought and contextual guidance are limited in Vietnamese, while cross-lingual prompting and few-shot learning show promising performance improvements. The findings underscore the feasibility of adapting benchmarks to less-resourced languages and provide insights into strengths and weaknesses in the performance of Vietnamese LLMs, suggesting directions for model improvements.

We aim to develop a metric of politicization by investigating whether this concept can be operationalized computationally using document embeddings. We are interested in measuring the extent to which foreign aid is politicized. Textual reports of foreign aid projects are often made available by donor governments, but these are large and unstructured. By embedding them in vector space, we can compute similarities between sets of known politicized keywords and the foreign aid reports. We present a pilot study where we apply this metric to USAID reports.