MR-BERT Metaphor-Relation BERT

This repository provides the codes and data for the paper published at ACL 2021: Verb Metaphor Detection via Contextual Relation Learning.

The main idea

The main idea of this paper is to view metaphor detection as a relation classification problem, i.e., the relation between a target verb and its various contexts. This idea is consistent with metaphor related theories, and can exploit techniques used in relation modeling.

Dataset

-	# tokens	# unique sent.	% metaphor
VUA-train	116,622	6,323	11.2
VUA-dev	38,628	1,550	11.6
VUA-Test	50,175(5,873)	2,694	12.4
MOH-X	647	647	48.7
TroFi	3,737	3,737	43.5

We mainly conduct experiments on the VUA dataset. We use the preprocessed version of the VUA dataset provided by (Gao et al., 2018). We also report the results on MOH-X and TroFi in three settings: zero-shot transfer, re-training and fine-tuning.

Requirements

• pip install -r requirements.txt

• You need to download the bert base model--'bert-base-uncased'

Run

• Get the embeddings from here and put it to the folder 'data'

• Training

  • python main_vua.py --bert_base_model_dir model_dir(the bert base model dir you downloaded)
  • e.g. python main_vua.py --bert_base_model_dir /home/model/bert-base-uncased

• Test

  • python test_vua.py

• You can get the model we saved from here

Update

• We update the integration strategy 'Average' to 'attention'. See the following two files for details: main_vua_extend.py, cosine_model.py

• python main_vua_extend.py --bert_base_model_dir model_dir

Citation Information

@inproceedings{song-etal-2021-verb,
    title = "Verb Metaphor Detection via Contextual Relation Learning",
    author = "Song, Wei  and
      Zhou, Shuhui  and
      Fu, Ruiji  and
      Liu, Ting  and
      Liu, Lizhen",
    booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)",
    month = aug,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.acl-long.327",
    doi = "10.18653/v1/2021.acl-long.327",
    pages = "4240--4251",
    abstract = "Correct natural language understanding requires computers to distinguish the literal and metaphorical senses of a word. Recent neu- ral models achieve progress on verb metaphor detection by viewing it as sequence labeling. In this paper, we argue that it is appropriate to view this task as relation classification between a verb and its various contexts. We propose the Metaphor-relation BERT (Mr-BERT) model, which explicitly models the relation between a verb and its grammatical, sentential and semantic contexts. We evaluate our method on the VUA, MOH-X and TroFi datasets. Our method gets competitive results compared with state-of-the-art approaches.",
}