Effective Distant Supervision for Temporal Relation Extraction
This addresses the challenge of limited high-quality data for temporal relation extraction in new domains, though it is incremental as it builds on existing distant supervision and Transformer methods.
The paper tackles the problem of training temporal relation extraction models in new domains by introducing a method to automatically collect distantly-supervised examples, where explicit temporal cues are masked to force learning of other signals. It demonstrates that a pre-trained Transformer model transfers effectively to human-annotated benchmarks in zero-shot and few-shot settings, with the masking scheme improving generalization.
A principal barrier to training temporal relation extraction models in new domains is the lack of varied, high quality examples and the challenge of collecting more. We present a method of automatically collecting distantly-supervised examples of temporal relations. We scrape and automatically label event pairs where the temporal relations are made explicit in text, then mask out those explicit cues, forcing a model trained on this data to learn other signals. We demonstrate that a pre-trained Transformer model is able to transfer from the weakly labeled examples to human-annotated benchmarks in both zero-shot and few-shot settings, and that the masking scheme is important in improving generalization.