Text-to-ES aims to translate natural language into executable Elasticsearch(ES) queries. Unlike more open-ended reasoning tasks, such as mathematical reasoning, this problem is defined over a constrained domain-specific language, where the synergy between forward reasoning and backward reasoning can be more explicitly realized under strict structural constraints. While recent work has explored reasoning-oriented training and prompting in text-to-SQL, forward and backward reasoning have not been systematically studied in text-to-ES. In this paper, we systematically study the bidirectional duality of the text-to-ES task by establishing a bidirectional reasoning perspective across both training and prompting stages. We introduce two large-scale datasets, TERD and ETRD, for forward and backward reasoning, respectively, and propose DualES, a unified framework that combines the two. Experimental results show that our DualES-CoRE strategy sets a new State-of-the-Art(SOTA), achieving 69.23% on LED and 50.14% on BirdES. This corresponds to absolute execution accuracy gains of 6.92% and 7.34% over previous benchmarks, outperforming both open-source and closed-source baselines. We will release the datasets in the future.

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.