Adjective Phrases in PNL and its Application to Reverse Dictionary

Abstract

Owing to the large information content of adjectives in natural language-based data, we have dealt with the subject of adjective phrases in the paradigm of PNL. We have proposed a GCR form for the same and demonstrated its application in building a specific instance of Reverse Dictionary. A Reverse Dictionary takes a natural language description of a concept in the user's mind as input and generates semantically appropriate word/s as output. For example, it could accept descriptions like 'to run slowly,' 'to run but not very fast,' etc., for the word 'jog'. As Reverse Dictionary deals with natural language, handling perception-based information is important in its context for capturing the user intent. The D-RD, Reverse Dictionary for Descriptive words, designed in this paper, takes a natural language description of a human as input and generates appropriate adjectives for usage in text/speech as output. As a part of the design, we have proposed a novel CWW based similarity measure for calculating semantic similarity between adjective phrases. The D-RD is evaluated against two benchmarks, Onelook.com and WantWords Online RD, and is reported to outperform both for the test set under consideration.