Sonar-TS introduces a training-free framework for natural language querying of massive time series databases, defining the new NLQ4TSDB problem, creating the NLQTSBench benchmark, and using a search-then-verify approach with symbolic indexing and code-based verification to find pattern matches across millions of data points.
The paper introduces Sonar‑TS, a training‑free neural‑symbolic system that tackles the newly defined NLQ4TSDB problem—natural‑language queries over database‑scale time‑series—by converting shape intents into searchable symbols and verifying candidates with executable code, achieving up to 3.8× higher scores than the strongest Text‑to‑SQL baseline while highlighting remaining challenges in shape understanding.
The paper defines the NLQ4TSDB problem of letting non‑expert users query massive time‑series data with natural language, builds the large‑scale NLQTSBench benchmark, proposes the neural‑symbolic Sonar‑TS framework that searches then verifies, and shows it outperforms existing baselines while highlighting remaining challenges.
