- Rating
- 1
- Confidence
- 3
- Informativeness
- 1
- Presentation
- 2
- Soundness
- 1
Read review
Paper summary
ANCHOR combines DAPO-based reinforcement learning, language rewards, and a vocabulary-margin anchoring loss applied to the final three transformer layers. The goal is to support mixed-language reasoning while providing an explanation in the question language. Experiments with Qwen3-8B-Instruct show improved accuracy and language alignment on MGSM, accuracy gains on MMLU-ProX-Lite, and little accuracy change on Global-MMLU-Lite.
Critical strengths
The comparison with R4, which retains the behavioral rewards but removes anchoring, provides useful evidence that the auxiliary loss contributes beyond reward design. On MGSM, adding anchoring raises accuracy from 84.76% to 86.07% and public-language alignment from 69.1 to 74.2. Improving both measures is a meaningful result for the problem the paper addresses.
Critical weaknesses
- The public-explanation requirement largely fails on the knowledge benchmarks. Tables 17–18 report a 97.5% empty-public-segment rate on both Lite benchmarks under the reported sampling setting, with alignment scores of only 2.2% and 1.9%. Providing an explanation that question-language speakers can read is a central motivation of the paper. These results substantially weaken the claim that ANCHOR addresses the accuracy–language trade-off across tasks.
- The evidence does not establish why the proposed recovery band is important. Figures 3–4 show improvements in the vocabulary margin that the loss directly optimizes. The R4 comparison establishes the benefit of adding the loss, but does not distinguish recovery-band anchoring from a more general language regularizer. There is no final-layer-only control or comparison with another layer band. The transfer of thresholds from question tokens to generated reasoning is also insufficiently validated. I find the empirical gain more convincing than the proposed explanation for it.
Questions for authors
- Evaluation clarification: What causes the 97.5% empty-public rate: the evaluation prompt, response parsing, or the learned behavior? Does explicitly requesting a public explanation preserve the reported accuracy while substantially improving alignment?
- Evidence for the central mechanism: How does the proposed recovery band compare with final-layer-only anchoring or a matched alternative layer band under the same training setup? This comparison would materially affect my assessment of the technical contribution.
5 comments
Response to Q2: Evidence for the Central Mechanism
We sincerely thank you for pointing out the need for a more comprehensive ablation of the Recovery Band. To address this concern, we have conducted three groups of additional experiments: shifted bands outside the identified recovery region, a reduced band, and an expanded band. The only variable in all experiments is the selection of the anchor layer, and the rest of the training and inference settings are exactly the same as the Anchor.The first two groups have been completed, while the expanded band experiment is still running and is expected to finish within approximately three days. We will provide its results in the updated response.
| Config | en | es | fr | de | zh | ja | sw | bn | ru | th | te | AVG | bo | |||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ANCHOR (L34–L36) | 96.00 | 93.20 | 86.40 | 88.80 | 88.40 | 86.00 | 62.00 | 85.20 | 90.80 | 89.20 | 80.80 | 86.07 | — | 74.18 | — | 10.04 | — | 8.40 |
| L36 only | 93.60 | 90.00 | 85.60 | 85.20 | 88.00 | 83.60 | 55.60 | 84.00 | 92.00 | 86.80 | 77.20 | 83.78 | −2.29 | 74.45 | +0.27 | 9.72 | −0.33 | 9.60 |
| L29–L31 (shifted 5 layers earlier) | 94.40 | 91.20 | 87.20 | 88.00 | 88.00 | 86.80 | 56.80 | 85.20 | 93.20 | 88.80 | 80.40 | 85.45 | −0.62 | 69.61 | −4.57 | 7.84 | −2.20 | 6.40 |
| L24–L26 (shifted 10 layers earlier) | 94.80 | 91.20 | 86.00 | 88.00 | 84.80 | 84.40 | 56.00 | 84.40 | 91.60 | 89.20 | 79.60 | 84.55 | −1.53 | 69.13 | −5.05 | 8.12 | −1.92 | 8.00 |
For the reduced-band setting, we retain only L36. For the shifted-band controls, we move the three-layer band five and ten layers earlier, resulting in L29–L31 and L24–L26, respectively. Compared with the original L34–L36 configuration, all three controls yield lower average accuracy, with decreases of 2.29, 0.62, and 1.53 points, respectively. The two shifted bands also exhibit notably lower and , while the L36-only setting achieves a slightly higher with only a marginal difference, but shows a decrease in . Overall, these results suggest that both the location and coverage of the constrained layers matter, and that the recovery band identified by our initial layer-wise analysis provides the best overall performance among the tested configurations. We promise to add this important experiment to the main text in the revised version.
We sincerely hope you can reconsider our work. We are readily available for further discussion until all your concerns are fully resolved. We commit that all key supplementary materials from this discussion period will be fully incorporated into the final version of the paper.
We sincerely thank you for your time and constructive feedback on our work. We provide point-by-point responses to all your questions and concerns below.
Response to Q1: Evaluation Clarification
We appreciate your attention to this interesting phenomenon. We first clarify that the Lite benchmarks were introduced to evaluate cross-task generalization, rather than to serve as the primary setting for method development. To maintain consistency with the MGSM experiments, we initially used the same response structure and language requirements across these benchmarks. However, unlike MGSM, both Lite benchmarks consist of multiple-choice questions with option letters as final answers. Under this prompt, the model often outputs the answer directly and omits the public explanation, as we also observed when inspecting the raw responses. This phenomenon reflects a mismatch between the prompt structure and the multiple-choice task format, rather than necessarily indicating a limitation of ANCHOR.
To directly address your concern, we conducted an additional comparison. Specifically, we revised the prompt to explicitly require the model to provide a complete explanation in the question language before giving its final answer, while keeping the model and all other evaluation conditions unchanged. The results are presented below.
Results on Global-MMLU-Lite:
Results on MMLU-ProX-Lite:
The revised prompt substantially improves language behavior on both benchmarks. On Global-MMLU-Lite, the language alignment score (A) increases from 2.46 to 88.02, while the empty-public rate decreases from 97.29% to 2.51%. On MMLU-ProX-Lite, A increases from 2.07 to 75.96, and the empty-public rate decreases from 97.32% to 14.75%. Therefore, the original results should not be interpreted as evidence that ANCHOR fails to achieve its intended language behavior. Rather, they highlight the importance of adapting the output requirements to the task format when evaluating cross-task generalization. We promise to add this important experiment to the main text in the revised version.