Hang Hu, Ziyan Liu, Rujie Wen, Ruihui Hou, Xueyan Wu, Mu Zhang, Jianxing Yu, Tong Ruan, Jingping Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Elasticsearch (ES) is a distributed RESTful search engine optimized for large-scale and long-text search scenarios. Recent research on text-to-Query has explored using large language models (LLMs) to convert user query intent to executable code, making it an increasingly popular research topic. To our knowledge, we are the first to introduce the novel semantic parsing task text-to-ES. To bridge the gap between LLM and ES, in detail, we leverage LLMs and employ domain experts to generate ES query bodies, which are Domain-Specific Language (DSL), along with the corresponding post-processing code to support multi-index ES queries. Consequently, we propose the text-to-ES benchmark that consists of two datasets: Large Elasticsearch Dataset (LED), containing 26,207 text-ES pairs derived from a 224.9GB schema-free database, and ElasticSearch (BirdES)with 10,926 pairs sourced from the Bird dataset on a 33.4GB schema-fixed database. Compared with fourteen advanced LLMs and six code-based LLMs, the model we trained outperformed DeepSeek-R1 by 15.64% on the LED dataset, setting a new state-of-the-art, and achieved 78% of DeepSeek-R1’s performance on the BirdES dataset. Additionally, we provide in-depth experimental analyses and suggest future research directions for this task. Our datasets are available at https://huggingface.co/datasets/Barry1915/Text-to-ES.

Medical visual question answering (MedVQA) requires the model to answer natural language questions based on the given medical images. Recently, large vision-language models (LVLMs) have emerged as the dominant paradigm for this task, enabling open-ended predictions and effectively integrating textual and visual information. However, due to the lack of domain-specific medical knowledge, LVLMs often suffer from hallucination issues, leading to inaccurate answers and flawed chain-of-thought (CoT) reasoning. This limits both the accuracy and interpretability of their responses. Hence, we propose MedKCoT, a framework that integrates a multi-modal medical knowledge graph (MMKG) to mitigate LVLM hallucinations during CoT generation and introduces a novel supervision strategy to train an MMKG retriever without requiring manual annotations. Firstly, we define an LVLM preference score as the supervision signal to train an MMKG retriever. Secondly, the retriever searches entity knowledge related to the medical question-answering pair, which assists the LVLM in generating high-quality CoTs. Finally, the obtained CoTs and original question-answering pairs are used to fine-tune a lightweight VLM as the MedVQA model. We evaluate MedKCoT on two public MedVQA datasets Slake and VQA-RAD. Experimental results show that MedKCoT achieves improvements of 2.93 % and 2.58 % over previous methods. The source code is available at https://github.com/EnjoyFailure/MedKCoT.

Decision trees, as a structured representation of medical knowledge, are critical resources for building clinical decision support systems. Their structured decision pathways can be used for retrieval to enhance clinical decision making. Currently, mainstream methods mainly utilize large language models and in-context learning for decision tree extraction. However, these methods often face challenges in understanding the structure of decision trees and accurately extracting the complete content of tree nodes, leading to noise in the extracted trees and ultimately impacting their effectiveness in clinical decision support system. To this end, in this paper, we propose a novel decision tree extraction framework, including two stages. In the first stage, we propose to use the If-Else pseudocode to represent the decision tree structure and design specific constraints on format and content to guide the LLM in generating outputs. In the second stage, we introduce a novel node-filling strategy called PlanSelect to match the extracted triplets with sub-sentences in the generated pseudocode, including four reasoning steps: observation, plan, action, and answer. To evaluate the effectiveness of our proposed method, we construct an English decision tree extraction dataset (EMDT) and conduct extensive experiments on the built and public datasets. Experiments on the Text2DT and EMDT datasets demonstrate that our method outperforms the current state-of-the-art approaches, achieving improvements of 1.37% and 1.54% on the $ER$ metric (which is lower is better), respectively. Furthermore, we use the medical decision trees extracted using our framework to improve the model's performance on clinical decision making tasks, i.e., CMB-Clin and MedQA.

Cross-lingual terminology alignment is an important task in the field of medical terminology. Through cross-lingual alignment, medical terms from different languages can be accurately mapped to their corresponding concepts, thus establishing a unified and multilingual medical terminology fusion system. However, the scarcity of annotated parallel corpora for medical terminology in Chinese and English languages poses a challenge during the training process of Chinese-English terminology alignment models, resulting in decreased alignment accuracy. To address this issue, this paper proposes a hybrid approach that combines Large Language Models(LLMs) and pretrained Language Models(PLMs), leveraging the rich multilingual knowledge embedded in LLMs to obtain more comprehensive Chinese-English terminology information for assisting cross-lingual terminology alignment. However, LLMs require significant computational resources and memory, making them less suitable for large-scale alignment tasks. To tackle this problem, a confidence sampling strategy is introduced, delegating challenging samples to the LLMs for re-ranking, thereby reducing resource costs. Additionally, a prompt strategy tailored for terminology alignment tasks is proposed to enhance the accuracy of predictions made by the LLMs.We evaluate our method on mapping files of two medical open terminologies, and the experimental results demonstrate that our method outperforms baseline methods by 2% in terms of Hits@1 and Hits@10 metrics.Our code and data are available at https://github.com/Bruce-Y12/two-stage-ranking.