WISTERIA: Weak Implicit Signal-based Temporal Relation Extraction with Attention
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912980461944832 |
|---|---|
| author | Do, Duy Dao Halftermeyer, Anaïs Dao, Thi-Bich-Hanh |
| author_facet | Do, Duy Dao Halftermeyer, Anaïs Dao, Thi-Bich-Hanh |
| contents | Temporal Relation Extraction (TRE) requires identifying how two events or temporal expressions are related in time. Existing attention-based models often highlight globally salient tokens but overlook the pair-specific cues that actually determine the temporal relation. We propose WISTERIA (Weak Implicit Signal-based Temporal Relation Extraction with Attention), a framework that examines whether the top-K attention components conditioned on each event pair truly encode interpretable evidence for temporal classification. Unlike prior works assuming explicit markers such as before, after, or when, WISTERIA considers signals as any lexical, syntactic, or morphological element implicitly expressing temporal order. By combining multi-head attention with pair-conditioned top-K pooling, the model isolates the most informative contextual tokens for each pair. We conduct extensive experiments on TimeBank-Dense, MATRES, TDDMan, and TDDAuto, including linguistic analyses of top-K tokens. Results show that WISTERIA achieves competitive accuracy and reveals pair-level rationales aligned with temporal linguistic cues, offering a localized and interpretable view of temporal reasoning. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_23319 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | WISTERIA: Weak Implicit Signal-based Temporal Relation Extraction with Attention Do, Duy Dao Halftermeyer, Anaïs Dao, Thi-Bich-Hanh Computation and Language Artificial Intelligence Temporal Relation Extraction (TRE) requires identifying how two events or temporal expressions are related in time. Existing attention-based models often highlight globally salient tokens but overlook the pair-specific cues that actually determine the temporal relation. We propose WISTERIA (Weak Implicit Signal-based Temporal Relation Extraction with Attention), a framework that examines whether the top-K attention components conditioned on each event pair truly encode interpretable evidence for temporal classification. Unlike prior works assuming explicit markers such as before, after, or when, WISTERIA considers signals as any lexical, syntactic, or morphological element implicitly expressing temporal order. By combining multi-head attention with pair-conditioned top-K pooling, the model isolates the most informative contextual tokens for each pair. We conduct extensive experiments on TimeBank-Dense, MATRES, TDDMan, and TDDAuto, including linguistic analyses of top-K tokens. Results show that WISTERIA achieves competitive accuracy and reveals pair-level rationales aligned with temporal linguistic cues, offering a localized and interpretable view of temporal reasoning. |
| title | WISTERIA: Weak Implicit Signal-based Temporal Relation Extraction with Attention |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2603.23319 |