What’s in a Definition? An Investigation of Semantic Features in Lexical Dictionaries

Luigi Di Caro


Encoding and generating word meanings as short definitions for user- and machine-consumption dictionaries is still a usually adopted strategy within interpretable lexical-semantic resources. Definitions have the property of being natural to be created by humans, while several techniques have been proposed for the automatic extraction from large corpora. However, the reversed process of going back to the words (i.e., onomasiological search) is all but simple for both humans and machines. Indeed, definitions show context- and conceptual-based properties which influence their quality. In this contribution, I want to draw the attention to this problem, through a simple content-to-form experimentation with humans in the loop. The results give some first insight on the relevance of the problem from a computational perspective. In addition, I analyzed (both quantitatively and qualitatively) a set of 1,901 word definitions taken from different sources, towards the modeling of features for their generation and automatic extraction.


Paper Citation