arXiv cs.AI· Jessica Lundin, Katherine Rosenfeld, David Adelani, Geoffrey Mosoti Nyakiongora, Halfan Ngowo, Sloan Coffin, Ralph Abboud, Guillaume Chabot-Couture·· 3 小时前AI 评分35
Fikiria：用强化学习实现斯瓦希里语母语数学推理
Show Your Work in Swahili: Reinforcement Learning for Native-Language Mathematical Reasoning
AI 导读
arXiv 论文提出 Fikiria，一个面向斯瓦希里语数学推理的两阶段 SFT+GRPO 训练方案，奖励结合格式门槛、答案准确率与 ChrF++ 推理相似度。
正文 · 原文
Abstract:We present Fikiria, a two-stage SFT+GRPO recipe for Swahili mathematical reasoning. Its reward multiplies a format gate by a weighted sum of answer accuracy and the ChrF\texttt{++} similarity of the Swahili reasoning to an in-language reference chain of thought, so that training targets the surface fidelity of the reasoning trace and not only the final answer. We train on a 2.3k-problem Swahili dataset created by translating GSM8K with GPT-4.1 and filtering with SSA-COMET. On AfriMGSM Swahili, Fikiria-AfriqueQwen-14B reaches 79.2\%. Under the same fixed prompt and strict answer-format scoring, GPT-4o reaches 68.0\%, partly because it often fails the required answer format; IrokoBench reports 78.8\% for GPT-4o using a best-per-model prompt, a figure numerically close to ours but not directly comparable. In a small human evaluation with co-author annotators, GPT-4o's outputs were preferred overall. ChrF\texttt{++} measures surface fidelity, not the validity of individual reasoning steps.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.16104 [cs.CL] |
| (or arXiv:2611.16104v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.16104 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jessica Lundin PhD [view email]
[v1]
Wed, 11 Nov 2026 23:25:01 UTC (2,141 KB)
来源：arXiv cs.AI · arxiv.org