Oct 12, 2022 � We can clearly see that the pre-trained bi-encoder models outperform the variant without pre-training in nearly all settings (with the results�...

Oct 6, 2022 � Our experimental results show that the resulting encoders allow us to predict commonsense properties with much higher accuracy than is possible�...

Sep 15, 2022 � We can clearly see that the pre-trained bi-encoder models outperform the variant without pre-training in nearly all settings (with the results�...

Our experimental results show that the resulting encoders allow us to predict commonsense properties with much higher accuracy than is possible by directly fine�...

Our experimental results show that the resulting encoders allow us to predict commonsense properties with much higher accuracy than is possible by directly fine�...

On-demand video platform giving you access to lectures from conferences worldwide.

Modelling commonsense properties using pre-trained bi-encoders. In Proceedings of the 29th International Conference on Computational Linguistics, pages 3971�...

Abstract:Neural language representation models such as BERT, pre-trained on large-scale unstructured corpora lack explicit grounding to real-world commonsense�...

We show that this leads to embeddings which capture a more diverse range of commonsense properties, and consistently improves results in downstream tasks such�...

BERT-base was used as the language model in these experiments. from publication: Modelling Commonsense Properties using Pre-Trained Bi-Encoders | Grasping�...