Salvato in:
Dettagli Bibliografici
Autori principali: Virgo, Felix, Cheng, Fei, Pereira, Lis Kanashiro, Asahara, Masayuki, Kobayashi, Ichiro, Kurohashi, Sadao
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:https://arxiv.org/abs/2403.18504
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917624093343744
author Virgo, Felix
Cheng, Fei
Pereira, Lis Kanashiro
Asahara, Masayuki
Kobayashi, Ichiro
Kurohashi, Sadao
author_facet Virgo, Felix
Cheng, Fei
Pereira, Lis Kanashiro
Asahara, Masayuki
Kobayashi, Ichiro
Kurohashi, Sadao
contents We propose a voting-driven semi-supervised approach to automatically acquire the typical duration of an event and use it as pseudo-labeled data. The human evaluation demonstrates that our pseudo labels exhibit surprisingly high accuracy and balanced coverage. In the temporal commonsense QA task, experimental results show that using only pseudo examples of 400 events, we achieve performance comparable to the existing BERT-based weakly supervised approaches that require a significant amount of training examples. When compared to the RoBERTa baselines, our best approach establishes state-of-the-art performance with a 7% improvement in Exact Match.
format Preprint
id arxiv_https___arxiv_org_abs_2403_18504
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle AcTED: Automatic Acquisition of Typical Event Duration for Semi-supervised Temporal Commonsense QA
Virgo, Felix
Cheng, Fei
Pereira, Lis Kanashiro
Asahara, Masayuki
Kobayashi, Ichiro
Kurohashi, Sadao
Computation and Language
We propose a voting-driven semi-supervised approach to automatically acquire the typical duration of an event and use it as pseudo-labeled data. The human evaluation demonstrates that our pseudo labels exhibit surprisingly high accuracy and balanced coverage. In the temporal commonsense QA task, experimental results show that using only pseudo examples of 400 events, we achieve performance comparable to the existing BERT-based weakly supervised approaches that require a significant amount of training examples. When compared to the RoBERTa baselines, our best approach establishes state-of-the-art performance with a 7% improvement in Exact Match.
title AcTED: Automatic Acquisition of Typical Event Duration for Semi-supervised Temporal Commonsense QA
topic Computation and Language
url https://arxiv.org/abs/2403.18504