Analyzing Text Complexity and Text Simplification: Connecting Linguistics, Processing and Educational Applications

Reading plays an important role in the process of learning and knowledge acquisition for both children and adults. However, not all texts are accessible to every prospective reader. Reading difficulties can arise when there is a mismatch between a reader’s language proficiency and the linguistic complexity of the text they read. In such cases, simplifying the text in its linguistic form while retaining all the content could aid reader comprehension. In this thesis, we study text complexity and simplification from a computational linguistic perspective. We propose a new approach to automatically predict the text complexity using a wide range of word level and syntactic features of the text. We show that this approach results in accurate, generalizable models of text readability that work across multiple corpora, genres and reading scales. Moving from documents to sentences, We show that our text complexity features also accurately distinguish different versions of the same sentence in terms of the degree of simplification performed. This is useful in evaluating the quality of simplification performed by a human expert or a machine-generated output and for choosing targets to simplify in a difficult text. We also experimentally show the effect of text complexity on readers’ performance outcomes and cognitive processing through an eye-tracking experiment. Turning from analyzing text complexity and identifying sentential simplifications to generating simplified text, one can view automatic text simplification as a process of translation from English to simple English. In this thesis, we propose a statistical machine translation based approach for text simplification, exploring the role of focused training data and language models in the process. Exploring the linguistic complexity analysis further, we show that our text complexity features can be useful in assessing the language proficiency of English learners. Finally, we analyze German school textbooks in terms of their linguistic complexity, across various grade levels, school types and among different publishers by applying a pre-existing set of text complexity features developed for German