We explore the caveats of a promising research field, namely neural machine translation, in the context of preserving indigenous languages in Africa. We face the challenge of dealing with low-resource language pairs. Methodically, we employ some literature approaches and frameworks to learn translation models from Bible data in Moore (a major language in Burkina Faso) and French. Our experiments indeed confirm previous findings in the literature that vanilla neural machine translation models are ineffective for low resource language pairs. Surprisingly, however, we also found that we are not able to even remotely match performance recorded by the state of the art adapted methods for low-resource language pair in the literature. Nevertheless, although word alignment, Byte Pair Encoding and Adam optimization did not successfully bring reasonable performance, we note that there are many remaining insightful approaches for low-resource language pairs.