Carpe Diem: On the Evaluation of World Knowledge in Lifelong Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kim, Yujin, Yoon, Jaehong, Ye, Seonghyeon, Bae, Sangmin, Ho, Namgyu, Hwang, Sung Ju, Yun, Se-young
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910415438479360
author Kim, Yujin
Yoon, Jaehong
Ye, Seonghyeon
Bae, Sangmin
Ho, Namgyu
Hwang, Sung Ju
Yun, Se-young
author_facet Kim, Yujin
Yoon, Jaehong
Ye, Seonghyeon
Bae, Sangmin
Ho, Namgyu
Hwang, Sung Ju
Yun, Se-young
contents The dynamic nature of knowledge in an ever-changing world presents challenges for language models trained on static data; the model in the real world often requires not only acquiring new knowledge but also overwriting outdated information into updated ones. To study the ability of language models for these time-dependent dynamics in human language, we introduce a novel task, EvolvingQA, a temporally evolving question-answering benchmark designed for training and evaluating LMs on an evolving Wikipedia database. The construction of EvolvingQA is automated with our pipeline using large language models. We uncover that existing continual learning baselines suffer from updating and removing outdated knowledge. Our analysis suggests that models fail to rectify knowledge due to small weight gradients. In addition, we elucidate that language models particularly struggle to reflect the change of numerical or temporal information. Our work aims to model the dynamic nature of real-world information, suggesting faithful evaluations of the evolution-adaptability of language models.
format Preprint
id arxiv_https___arxiv_org_abs_2311_08106
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Carpe Diem: On the Evaluation of World Knowledge in Lifelong Language Models
Kim, Yujin
Yoon, Jaehong
Ye, Seonghyeon
Bae, Sangmin
Ho, Namgyu
Hwang, Sung Ju
Yun, Se-young
Computation and Language
The dynamic nature of knowledge in an ever-changing world presents challenges for language models trained on static data; the model in the real world often requires not only acquiring new knowledge but also overwriting outdated information into updated ones. To study the ability of language models for these time-dependent dynamics in human language, we introduce a novel task, EvolvingQA, a temporally evolving question-answering benchmark designed for training and evaluating LMs on an evolving Wikipedia database. The construction of EvolvingQA is automated with our pipeline using large language models. We uncover that existing continual learning baselines suffer from updating and removing outdated knowledge. Our analysis suggests that models fail to rectify knowledge due to small weight gradients. In addition, we elucidate that language models particularly struggle to reflect the change of numerical or temporal information. Our work aims to model the dynamic nature of real-world information, suggesting faithful evaluations of the evolution-adaptability of language models.
title Carpe Diem: On the Evaluation of World Knowledge in Lifelong Language Models
topic Computation and Language
url https://arxiv.org/abs/2311.08106