Week 3 – Hypernyms and Hyponyms

Mar 04, 2019

This week, I started my research on the three different methods to be coded in my project. I read Xiaojin Zhu and Saisuresh Krishnakumaran’s “Hunting Elusive Metaphors Using Lexical Resources,” the first of the methods–beyond its relevance to my project, I found the paper to be a fantastic read!  

Zhu and Krishnakumaran’s algorithm is built around the use of hyponyms and hypernyms, and it is first important to understand what these terms mean. Hyponyms and hypernyms describe the type-of relationships between words; the words salmon and goldfish, for example, are hyponyms of the word fish, which in turn is a hyponym of the term animal. Animal is a hypernym of fish, which is a hypernym of salmon and goldfish, and so on.

This leads us to our next question: how are hypernym and hyponym word relations implemented within the actual research? Zhu and Krishnakumaran found that hyponym and hypernym word relations were indicative of a metaphorical or literal relationship between a subject and an object.


The data used was formatted thusly: [Subject] [Relational Verb] [Object]


Zhu and Krishnakumaran offer the example of “He is a scientist,” which is a literal phrase, and “All the world’s a stage,” a very metaphorical phrase. Within the first phrase, he is synonymous with person. Scientist is a hyponym of person, and thus the subject and the object have a hyponym-hypernym relation, and the sentence is literal because of this relation. For the second phrase, world does not have a hyponym-hypernym relationship with stage, and because of this, the sentence is metaphorical. This is how I’ll be utilizing hyponym-hypernym relationships to determine whether or not a sentence is metaphorical or not. I’m praying for good results!

2 Replies to “Week 3 – Hypernyms and Hyponyms”

  1. Doan Y. says:

    Good luck with your results and looking forward to see the other 2 methods you’ll be using to code this project.

  2. Anish M. says:

    These are some very interesting findings! Would you think an RNN would work to detect this specific kind of metaphors? I’m interested to find out how you will approach more lengthy and understated metaphors.

Leave a Reply

Your email address will not be published.