**Summary:**
The paper proposes a novel methodology that combines sample sparsification and sample amplification for improving the efficiency and effectiveness of training machine learning models, particularly in text classification tasks. It aims to reduce the need for extensive labeled data, a common challenge in natural language processing (NLP), by finding a smaller, high-quality subset of the original data (support set) and adding random samples from unknown categories (amplified set) to improve model robustness against out-of-distribution (OOD) samples. The method is evaluated on various text datasets, showing improvements in zero-shot multilingual transfer, open set recognition, and reduced labeling budget without compromising model performance.

**Strengths:**
- **Innovative combination:** The paper innovatively combines sample sparsification and sample amplification, offering a novel approach to tackle data scarcity and OOD sample classification issues.
- **Broad applicability:** The methodology is demonstrated to work across different datasets and languages, showcasing its potential for a wide range of NLP tasks.
- **Improvement in zero-shot multilingual transfer:** The paper presents an improvement in zero-shot multilingual transfer performance, which is a significant contribution to the field of multilingual NLP.

**Weaknesses:**
- **Theoretical underpinning:** The paper lacks a rigorous theoretical framework to explain the effectiveness of the proposed methods. Providing a deeper understanding of why and how the combination of sample sparsification and amplification works could strengthen the paper.
- **Limitations in open set recognition:** While the paper shows improvements in open set recognition, the scope and depth of evaluation could be expanded to cover a broader range of scenarios, including more challenging OOD cases.

**Questions:**
- How does the proposed methodology compare to existing approaches in terms of efficiency and effectiveness in reducing labeling budgets while maintaining or improving model performance?
- What are the theoretical foundations that underpin the effectiveness of combining sample sparsification and amplification for improving robustness against OOD samples?
- How does the performance of the proposed methodology scale with the size of the original dataset and the complexity of the NLP task?

**Soundness:**
**Presentation:**
The paper is well-organized and includes relevant background information. However, there is a lack of clarity in the methodology's theoretical underpinnings, which could benefit from more in-depth analysis.

**Contribution:**
The paper contributes a novel approach to address data scarcity and OOD sample classification in NLP, with demonstrated improvements in zero-shot multilingual transfer and open set recognition.

**Rating:**
3 good

**Paper Decision:**
- **Decision:** Accept
- **Reasons:** The paper presents a novel methodology with clear contributions to the field of NLP, particularly in addressing data scarcity and OOD sample classification. While there are areas for improvement, such as theoretical foundations and experimental scope, the proposed methods show promising results and have the potential for significant impact in the field.