To improve deep-learning performance in low-resource settings, many researchers have redesigned model architectures or applied additional data (e.g., external resources, unlabeled samples). However, there have been relatively few discussions on how to make good use of small amounts of labeled samples, although it is potentially beneficial and should be done before applying additional data or redesigning models. In this study, we assume a low-resource setting in which only a few labeled samples (i.e., 30–100 per class) are available, and we discuss how to exploit them without additional data or model redesigns. We explore possible approaches in the following three aspects: training validation splitting, early stopping, and weight initialization. Extensive experiments are conducted on six public sentence classification datasets. Performance on various evaluation metrics (e.g., accuracy, loss, and calibration error) significantly varied depending on the approaches that were combined in the three aspects. Based on the results, we propose an integrated method, which is to initialize the model with a weight averaging method and use a non-validation stop method to train all samples. This simple integrated method consistently outperforms the competitive methods; e.g., the average accuracy of six datasets of this method was 1.8% higher than those of conventional validation-based methods. In addition, the integrated method further improves the performance when adapted to several state-of-the-art models that use additional data or redesign the network architecture (e.g., self-training and enhanced structural models). Our results highlight the importance of the training strategy and suggest that the integrated method can be the first step in the low-resource setting. This study provides empirical knowledge that will be helpful when dealing with low-resource data in future efforts. Our code is publicly available at https://github.com/DMCB-GIST/exploit_all_samples.

Automated Essay Scoring (AES) plays a crucial role in educational assessment, yet building reliable systems often requires large amounts of prompt- and trait-specific labeled data. As a result, many studies rely heavily on a single public dataset, ASAP++, which limits their generalizability. In this paper, we explore diverse zero-shot AES approaches that are both prompt-agnostic and multi-trait, thereby reducing reliance on labeled data. Our study examines large language models (LLMs), BERT-like fine-tuned models, and simple feature-based methods. Furthermore, we propose two novel zero-shot frameworks: CAnUSe (Comparative Assessment and subsequent Uncertainty-aware Self-training) and EASY (Essay Assessment with Simple Yardsticks). CAnUSe combines LLM-based pairwise comparisons with uncertainty-aware self-training using a BERT-like single-essay scoring model, while EASY employs only three intuitive features-essay length, vocabulary diversity, and grammar quality. Evaluation on three benchmark datasets-ASAP++, Ellipse, and TOEFL11-shows that our frameworks achieve state-of-the-art performance, demonstrating strong generalizability across datasets. Remarkably, despite operating in a zero-shot setting, our models approach the performance of fully supervised models trained with approximately 10K labeled samples. Further analysis provides practical insights for building real-world AES systems.

Automated Essay Scoring (AES) plays a crucial role in education by providing scalable and efficient assessment tools. However, in real-world settings, the extreme scarcity of labeled data severely limits the development and practical adoption of robust AES systems. This study proposes a novel approach to enhance AES performance in both limited-data and full-data settings by introducing three key techniques. First, we introduce a Two-Stage fine-tuning strategy that leverages low-rank adaptations to better adapt an AES model to target prompt essays. Second, we introduce a Score Alignment technique to improve consistency between predicted and true score distributions. Third, we employ uncertainty-aware self-training using unlabeled data, effectively expanding the training set with pseudo-labeled samples while mitigating label noise propagation. We implement above three key techniques on DualBERT. We conduct extensive experiments on the ASAP++ dataset. As a result, in the 32-data setting, all three key techniques improve performance, and their integration achieves 91.2% of the full-data performance trained on approximately 1,000 labeled samples. In addition, the proposed Score Alignment technique consistently improves performance in both limited-data and full-data settings: e.g., it achieves state-of-the-art results in the full-data setting when integrated into DualBERT.

In practical customer-support dialogue systems, responses must simultaneously deliver factually grounded information and context-appropriate empathy, yet existing single-stage generation models often exhibit specialization bias, favoring one objective at the expense of the other. To address this limitation, we propose a dual-stage generation framework that explicitly decouples factual grounding from empathetic modulation. Our primary configuration follows a fact-to-empathy order, in which the system first generates a fact-centric draft via structured query interpretation and optional retrieval-augmented generation, then applies empathy-aware tuning conditioned on inferred emotion type, intensity, and empathy necessity. To enable deployment in resource-constrained environments, only the query interpretation module is explicitly trained using knowledge distillation, allowing the overall system to operate with compact 4B–8B backbone language models. Furthermore, we construct a customer-support dialogue dataset designed to reflect realistic interactions involving both informational and emotional demands. Extensive experiments with compact models show that the proposed approach generally improves key dimensions of empathetic response quality while maintaining overall factual performance, thereby helping mitigate the representational entanglement empirically observed in single-stage baselines. Both quantitative metrics and scenario-based analyses confirm that decoupled generation enables a more balanced integration of factuality and empathy than single-stage generation. These results suggest that dual-stage generation provides a practical and extensible foundation for deployable, real-world customer-support dialogue systems.