A Simple but Effective Pluggable Entity Lookup Table for Pre-trained Language Models
This addresses the issue of knowledge recall in PLMs for tasks involving entities, but it is incremental as it builds on existing knowledge-enhanced PLM methods.
The paper tackles the problem of pre-trained language models (PLMs) poorly recalling factual knowledge, especially for rare entities, by proposing a Pluggable Entity Lookup Table (PELT) that aggregates entity representations from corpora to infuse supplemental knowledge, achieving this with only 0.2%-5% pre-computation and demonstrating effectiveness on knowledge-related tasks.
Pre-trained language models (PLMs) cannot well recall rich factual knowledge of entities exhibited in large-scale corpora, especially those rare entities. In this paper, we propose to build a simple but effective Pluggable Entity Lookup Table (PELT) on demand by aggregating the entity's output representations of multiple occurrences in the corpora. PELT can be compatibly plugged as inputs to infuse supplemental entity knowledge into PLMs. Compared to previous knowledge-enhanced PLMs, PELT only requires 0.2%-5% pre-computation with capability of acquiring knowledge from out-of-domain corpora for domain adaptation scenario. The experiments on knowledge-related tasks demonstrate that our method, PELT, can flexibly and effectively transfer entity knowledge from related corpora into PLMs with different architectures.