According to the abstract of "Generation-Heavy Hybrid Machine Translation" (arXiv:2610.25107), state-of-the-art machine translation depends on large, symmetric resources from both source and target languages. The authors note this is true regardless of whether the approach is transfer-based, interlingua-based, or symbolic.
The claim is significant because it identifies a common bottleneck across MT paradigms: even advanced systems may struggle when one language has far more data or resources than the other. The abstract does not yet specify how the proposed generation-heavy hybrid approach addresses this limitation.
The source is a single arXiv announcement, so this summary is limited to the opening of the abstract. There are no other sources to compare, and no experimental results or further details are available in the provided text.