Metaphor Detection Dataset

Metaphor Detection Dataset

We developed this dataset in the scope of a preliminary experimentation aimed at assessing the quality of a metaphor detection algorithm based on concept abstractness scores [1].

In this simple corpus we collected 150 sentences in the format:

<id> <type> <label> <sentence>


  • id is an unique identifier;
  • type refers to the type of metaphorical expression of the sentence. T1 sentences contains two nouns and one verb used as copula (e.g., lawyers are reale sharks), while T2 sentences indicate a relationship between a noun and a verb (e.g., warm water evaporated in few minutes);
  • label indicates if the sentence is metaphorical or not (M, N);
  • sentence is the sentence.

Download “Metaphor Detection Dataset” MetaphorDetection.tar.bz2 – Downloaded 82 times – 7 KB

Reference papers

[1] [pdf] “Grasping metaphors: lexical semantics in metaphor analysis,” in The Semantic Web: ESWC 2018 Satellite Events, Cham, 2018, p. 192–195.
editor={Gangemi, Aldo
and Gentile, Anna Lisa
and Nuzzolese, Andrea Giovanni
and Rudolph, Sebastian
and Maleshkova, Maria
and Paulheim, Heiko
and Pan, Jeff Z
and Alam, Mehwish},
title={Grasping Metaphors: Lexical Semantics in Metaphor Analysis},
booktitle= {{The Semantic Web: ESWC 2018 Satellite Events}},
publisher={Springer International Publishing},
abstract={Metaphors represent to date an extraordinary challenge for computational linguistics. Dealing with metaphors has relevant consequences on our ability to build agents and systems that understand Natural Language and text documents: annotating metaphoric constructions by linking the metaphor elements to existing resources is a crucial step to make text documents more easily accessible by machines. Our approach tackles metaphors by considering concepts and their abstractness. We report the encouraging results obtained in a preliminary experimentation; we elaborate on present limitations, and individuate the needed improvements, which will be the base for future work.},
pdf = {}