Dataset Card for VMC-P

Dataset Summary

VMC-P (VU Amsterdam Metaphor Corpus with Paraphrases) is a dataset, developed for end-to-end metaphor interpretation. It consists of 10,716 textual sequences (6,653 sequences for training, 2,063 sequences for validation, and 2,000 sequences for testing) with manual annotations on the token level. These annotations specify the metaphricity (such as metaphorical or literal) and provide paraphrases of metaphorical lexical units, including both single- and multi-word expressions. The data were sourced from VU Amsterdam Metaphor Corpus, including genres, e.g., fiction, news, academic and conversational text.

Language

English

Dataset Structure

An example sentence and its labels from VMC-P is shown below:

{
  'ID': 'trn_976', 
  'doc_ID': 'ac2-fragment06', 
  'sent_ID': '1465', 
  'sent': "‘ Do n't they realise they 're playing with political dynamite ?", 
  'metaphor_index_list': [[7, 8], [10]], 
  'pos_list': ['underestimating', 'risks'], 
  'neg_list': [['exploiting', 'competing', 'trifling', 'dallying', 'exhausting', 'deploying', 'wagering', 'taking on', 'flirting', 'simulating', 're-creating', 'travelling', 'using', 'roleplaying', 'wreaking', 'manipulating', 'considering', 'toying', 'acting as', 'moving', 'discharging', 'running', 'playacting', 'bringing', 'gambling', 'working', 'making', 'performing', 'recreating', 'spiel', 'sounding', 'representing', 'acting', 'meeting', 'encountering', 'betting', 'diddling', 'making for', 'hitting', 'fiddling', 'take foolish risk', 'having', 'end up benefit their overlord', 'trying', 'producing', 'contributing'], ['explosive compound', 'explode']], 
  'lemma': "' do not they realise they be play with political dynamite ?", 
  'pos_tags': ['``', 'VB', 'RB', 'PRP', 'VB', 'PRP', 'VBP', 'VBG', 'IN', 'JJ', 'NN', '.'], 
  'open_class': ['O', 'VERB', 'ADV', 'O', 'VERB', 'O', 'VERB', 'VERB', 'O', 'ADJ', 'NOUN', 'O'], 
  'genre': 'fiction'
}

Data Fields

• ID: An unique index in the VMC-P dataset
• doc_ID and sent_ID: The indices were inherited from VU Amsterdam Metaphor Corpus.
• sent: The input sentence with tokenization.
• metaphor_index_list: The indices indicate the metaphorical tokens in the given sentence.
• pos_list: The correct paraphrases (ground truth labels).
• neg_list: The incorrect paraphrases (negative samples for contrastive learning paraphrases for metaphors).
• lemma: The lemmatized input sentence.
• pos_tags: Parts-of-speech tags of tokens in the input sentence, following the Universal Dependencies scheme.
• open_class: Open-class word labels, including verb, noun, adjective, adverb and others.
• genre: The genre indicates the origin of the text, including fiction, news, academic, conversation.

Licensing Information

Creative Commons Attribution-ShareAlike 3.0 Unported License

Citation Information

@article{mao2024metapro2,
  title={{MetaPro} 2.0: {Computational} Metaphor Processing on the Effectiveness of Anomalous Language Modeling},
  author={Mao, Rui and He, Kai and Ong, Claudia Beth and Liu, Qian and Cambria, Erik},
  booktitle={Findings of the Association for Computational Linguistics: ACL},
  year={2024},
  address={Bangkok, Thailand},
  publisher={Association for Computational Linguistics}
}