Open AccessData-to-text Generation via Planning
Author(s) -
Sen Yang,
Yang Liu
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1827/1/012190
Subject(s) - planner , plan (archaeology) , realization (probability) , text generation , sentence , computer science , artificial intelligence , natural language processing , mathematics , statistics , archaeology , history
Planning based text generation from structural data is an emerging approach that manages to plan for “what to say.” Existing methods can not dynamically adjust the planning, because they generate complete plans statically without considering the result of surface realization. To address this issue, we propose a model that conducts dynamic text planning and surface realization alternately. In each sentence’s generation, our model first plans the records according to the generated text and the plan history, then realizes the sentence conditioned on the corresponding plan. Experimental results demonstrate that our model performs better on both text planning and text generation. Our model with the dynamic planner achieves impressive results on the E2E dataset.