A survey on ordering of text at different granular levels
摘要
A well-ordered text is a crucial need for various models. The text ordering task has both direct and indirect influence on various tasks like concept to text, document modelling, essay scoring, linearization, machine translation, opinion generation of debate, string regeneration, text generation, text summarization, visual referring expression, etc. It contributes both as pre-processing for training data and post-processing for output data. In a document, the text entities include words, sentences, and paragraphs. Words are the basic building blocks of document. The order of words carries the correct grammatical-based syntactic structure of a document. Sentences as a cluster of words carry meaningful information ordering in their ordering. Similarly, the order of the paragraph maintains coherence with the topic of the description. A well-ordered document is the most feasible input to retrace the properties of the document. This survey presents all the basic elements of the task, effective techniques, popular datasets, and performance evaluation benchmarks with their strengths and weaknesses at granular levels.