A Regression Model Of English Text Readability Based On Newconcept3

Recently the importance of automatic text readability assessment hasbeen realized by many researchers because of its wide application in俗such fields as language teaching, language testing and publishing industry.Although the history of text readability (some researchers refer to it astext difficulty) study is nearly one century long, the problem still seems tobe a mystery to us, which allures many to devote to its research. Untilnow, many readability formulas have come out to measure text readability.However, sometimes, there are some problems or weakness in them.In this paper, a progressive assessing model is established on thebasis of the previous achievements. The author first puts forward the sixassumed factors that are assumed to be related with text difficultyaccording to language theories, namely total words, clause number,average sentence length, chunk number, type token ratio and worddifficulty. Then the sixty training texts from New Concept Book Three arechosen and the exact values of each factor for each text are calculated.Next the author tries to certify the hypothesis by conducting theregression analysis, aiming to figure out the function between textdifficulty value and the assumed factors. By doing that, the author findsthat the two factors, total words and word difficulty are collinear. That indicates linear regression analysis is not proper here. Then anothermethod ridge regression is used to explore the function relationship.Finally, the function which is tested by ten texts from New HorizonCollege English Book Two proves that the model is effective, and theproposed elements, that is, total words, number of clause, averagesentence length, number of chunks, TTR (type/token ratio), and worddifficulty all influence text readability，but the elements of total words,clause number and word difficulty weigh more than the other threefactors.Using regression analysis to build the quantitative relationshipbetween readability and its latent influential variables is a bold attempt,although the idea may be mature enough. The paper, which proposes anew way to solve readability problem, is of certain value in measuring thedifficulty level of college English text materials.