Text Classification for Italian Proficiency Evaluation

NLP technologies and components have an increasing diffusion in mass analysis of text based dialogues, such as classifiers for sentiment polarity, trends clustering of online messages and hate speech detection. In this work we present the design and the implementation an automatic classification tool for the evaluation of the complexity of Italian texts as understood by a speaker of Italian as a second language. The classification is done within the Common European Framework of Reference for Languages (CEFR) which aims at classifying speakers language proficiency. Results of preliminary experiments on a data set of real texts, annotated by experts and used in actual CEFR exam sessions, show a strong ability of the proposed system to label texts with the correct language proficiency class and a great potential for its integration in learning tools, such systems supporting examiners in tests design and automatic evaluation of writing abilities.