Automatic Readability Evaluation Using a Neural Network 2009 - 2010

Many formulas and methods for assessing the readability of a text or determining the appropriate grade level are inaccurate and based only on surface features of text. In automatic assessment of a text’s reading level, computers can easily run more sophisticated models than simple algebraic formulas; the goal of this project is to create such a model. Indexes and statistics will be computed with respect to various features of a text such as sentence length and lexical density. A combination of these textual features is necessary to accurately capture the readability of a text. A neural network will be used to implement a model for readability using said features as inputs. After being trained the model will be useful for determining approximately what U.S. grade level corresponds to a given text, for use in educational or other settings to assess writing for a certain audience.