Comments on: A Puntitled Framework for Evaluating the Quality of Puns

By: rfong'12

rfong'12 — Thu, 17 Mar 2016 22:51:00 +0000

Slant references seem difficult because they’re simultaneously underconstrained and have high semantic dimensionality.

Mapping “hate accessories” to “kill the messenger” has some pretty expensive steps, which might be reduced by storing word-sense-similarity lookups much looser than synonymity
– very loose semantic association between “hate” and “kill”
– one word sense of “accessories” is a hypernym of a phrase containing “messenger”

By: elplatt

elplatt — Wed, 16 Mar 2016 18:50:00 +0000

Proposal 1: modifier for the number of interpretations that are mutually, semantically consistent.

By: Daniel Schultz

Daniel Schultz — Wed, 16 Mar 2016 18:47:00 +0000

In reply to rfong'12. This is all awesome -- also, FWIW you might be able to generate slant references by analyzing a corpus for commonly repeated phrases.

By: rfong'12

rfong'12 — Wed, 16 Mar 2016 18:42:00 +0000

Ugh I thought my friend invented the n-dimensional pun terminology. Thanks for collating these into one place! I think it’d be funnier (and more NLP-relevant) to refer to “dual definitions” as “word-sense re-ambiguation”. Slant references sound difficult to achieve programmatically.

I’ve been doing a lot of what I call “shittyNLP” recently, which is just simple deterministic messing around with text for amusement purposes.

You might enjoy my Pun Assistinator (phonetic superstring lookup) – http://rfong.github.io/pun/
and this portmanteau generator (or couple name generator for your friends) – https://gist.github.com/rfong/1188f5128d6e443ede98

By: Jonathan Bobrow

Jonathan Bobrow — Wed, 16 Mar 2016 17:58:00 +0000

Just wanted to add a great series of illustrated puns… http://oddowl.com/#/square-dance/ and http://oddowl.com/#/baroque-oli/