Jason S Lucas, Matt Murtagh White, Ali Al-Lawati, Uchendu Uchendu, Adaku Uchendu, Dongwon Lee. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Large Language Models (LLMs) have demonstrated remarkable performance in various NLP tasks, including semantic parsing, which translates natural language into formal code representations. However, the reverse operation, translating code into natural language, termed semantic captioning, has received less attention. This task is increasingly important as LLMs are integrated into platforms for code generation, security analysis, and educational purposes. In this paper, we focus on the captioning of SQL query (SQ2Text) to address the critical need for understanding and explaining SQL queries in an era where LLM-generated code poses potential security risks. We repurpose semantic parsing datasets for semantic captioning, specifically SQL2Text. To overcome the limited robustness of Text2SQL datasets for the reverse task, we introduce an iterative ICL prompt leveraging GPT-4o to generate multiple additional utterances. We conduct experiments across multiple in-context learning (ICL) methods, emphasizing smaller, more computationally efficient LLMs. Our findings demonstrate that leveraging the inherent graph properties of SQL for few-shot ICL sample selection significantly outperforms random selection by up to 39% on BLEU score and provides better results than alternative approaches. Dataset and codes are accessible.

Trajectory generation is an important task in movement studies. Generated trajectories augment the training corpus of deep learning applications, facilitate experimental and theoretical research, and mitigate the privacy concerns associated with real trajectories. This is especially significant in the wildlife domain, where trajectories are scarce due to the ethical and technical constraints of the collection process. In this paper, we consider the problem of generating long-horizon trajectories, akin to wildlife migration, based on a small set of real samples. We propose a hierarchical approach to learn the global movement characteristics of the real dataset, and recursively refine localized regions. Our solution, WildGraph discretizes the geographic path into a prototype network of H31 regions and leverages a novel recurrent VAE to probabilistically generate paths over the regions, based on occupancy. Experiments performed on two wildlife migration datasets demonstrate the remarkable capability of WildGraph to generate realistic months-long trajectories using a sample size as small as 60 while improving generalization compared to existing work. Moreover, WildGraph achieves superior or comparable performance on performance measures, including geographic imagery similarity. Our code is published on the following repository: https://github.com/aliwister/wildgraph.