[5E1-GS-6d] Language media processing

10:00 AM - 10:15 AM JST(1:00 AM - 1:15 AM UTC)

[5E1-GS-6d-05] Formulation and Learning of Japanese Temporal Relation Estimation Based on Probability Distributions

〇Remi Oguni¹, Daichi Mochihashi^2,3, Ichiro Kobayashi¹ (1. Ochanomizu University, 2. The Institute of Statistical Mathematics, 3. National Institute for Japanese Language and Linguistics)

Keywords:

Temporal relationship,Normal distribution

Recognising temporal relations is a crucial task in natural language understanding for accurate contextual comprehension. This study estimates temporal relations between events within Japanese sentences based on a framework that represents the occurrence state of events as a probability distribution on a logarithmic time axis and views temporal relations as relationships between probability distributions. We target five temporal relation labels: A<B, A<=B, A=B, A>=B, and A>B. We re-examine probability calculation and the formulation of learning objectives. By considering not only the emphasis of positive examples but also the suppression of inappropriate relations, we aim to improve estimation consistency. Comparative experiments using generated corpora and existing datasets analyse the impact of the proposed method on estimation performance and the interpretability of outputs.