WaveTLM improves reliability of time-series language model outputs
WaveTLM: Reliable Time-Series Language Modeling through Task Compilation
Machine Learning
Summary
Time-series language models let people use everyday language to work with data that changes over time, like weather or heart rates. But sometimes these models give answers that sound right but are actually incorrect or inconsistent with the task. The authors of this paper created a way to make these models more reliable by turning user requests into detailed tasks that always produce valid outputs. Their new method, WaveTLM, performs much better at producing correct outputs across many time-series tasks than previous methods. This means users can trust the model’s results more when analyzing complex time-based data.
What this means in practice
- •For data science teams: Generate reliable time-series forecasts and analyses with consistent outputs from natural language queries.
- •For signal processing engineers: Produce trustworthy waveform classifications and anomaly detections using language-driven models with guaranteed output validity.
Authors
Jiahui Chen, Bingke Zhu, Hongyu Pan, Yingying Chen
Abstract
Time-series language models provide a shared natural-language interface across temporal tasks, but plausible text does not guarantee reliable task outputs. Responses may appear reasonable while hallucinating the required object: numerical sequences can violate shape, scale, channel order, or temporal alignment, and textual decisions can fall outside the legal label space. We formulate reliable time-series language modeling, separating task-object reliability from predictive quality. We introduce ExecTS-QA, a contract-grounded benchmark spanning forecasting, imputation, classification, anomaly detection, and waveform analysis. We further propose WaveTLM, a unified compiler-executor model whose task compiler transforms user requests, visible arguments, and wave-grounded evidence into typed task states, while task-native executors construct numerical tensors, legal decisions, or structured records. On ExecTS-QA, a single WaveTLM checkpoint achieves 99.40% contract-valid coverage, compared with 37.83% for the strongest evaluated string-first baseline, while retaining balanced predictive performance across all five task families. Evaluations on SciTS, TSQA, IRTS-ToolBench, and ARFBench provide additional evidence of transfer. The code, construction scripts, and ExecTS-QA dataset will be publicly released upon publication. These results show that task compilation can convert plausible language generation into reliable time-series outputs.