A Unified Model for Reverse Dictionary and Definition Modelling
share
We train a dual-way neural dictionary to guess words from definitions (reverse dictionary), and produce definitions given words (definition modelling). Our method learns the two tasks simultaneously, and handles unknown words via embeddings. It casts a word or a definition to the same representation space through a shared layer, then generates the other form from there, in a multi-task fashion. The model achieves promising automatic scores without extra resources. Human annotators prefer the proposed model's outputs in both reference-less and reference-based evaluation, which indicates its practicality. Analysis suggests that multiple objectives benefit learning.
READ FULL TEXT