Reading Level Assessment Using Support Vector Machines and Statistical Language Models

Proceedings of the 43rd Annual Meeting of the ACL, pages 523–530, Ann Arbor, June 2005. c⃝2005 Association for Computational Linguistics Reading Level Assessment Using Support Vector Machines and Statistical Language Models Sarah E. Schwarm Dept. of Computer Science and Engineering University of Washington Seattle, WA 98195-2350 sarahs@cs.washington.edu Mari Ostendorf Dept. of Electrical Engineering University of Washington Seattle, WA 98195-2500 mo@ee.washington.edu Abstract Reading proficiency is a fundamen- tal component of language competency. However, finding topical texts at an appro- priate reading level for foreign and sec- ond language learners is a challenge for teachers. This task can be addressed with natural language processing technology to assess reading level. Existing measures of reading level are not well suited to this task, but previous work and our own pilot experiments have shown the bene- fit of using statistical language models. In this paper, we also use support vector machines to combine features from tradi- tional reading level measures, statistical language models, and other language pro- cessing tools to produce a better method of assessing reading level. 1 Introduction The U.S. educational system is faced with the chal- lenging task of educating growing numbers of stu- dents for whom English is a second language (U.S. Dept. of Education, 2003). In the 2001-2002 school year, Washington state had 72,215 students (7.2% of all students) in state programs for Limited English Proficient (LEP) students (Bylsma et al., 2003). In the same year, one quarter of all public school stu- dents in California and one in seven students in Texas were classified as LEP (U.S. Dept. of Edu- cation, 2004). Reading is a critical part of language and educational development, but finding appropri- ate reading material for LEP students is often diffi- cult. To meet the needs of their students, bilingual education instructors seek out “high interest level” texts at low reading levels, e.g. texts at a first or sec- ond grade reading level that support the fifth grade science curriculum. Teachers need to find material at a variety of levels, since students need different texts to read independently and with help from the teacher. Finding reading materials that fulfill these requirements is difficult and time-consuming, and teachers are often forced to rewrite texts themselves to suit the varied needs of their students. Natural language processing (NLP) technology is an ideal resource for automating the task of selecting appropriate reading material for bilingual students. Information retrieval systems successfully find top- ical materials and even answer complex queries in text databases and on the World Wide Web. How- ever, an effective automated way to assess the read- ing level of the retrieved text is still needed. In this work, we develop a method of reading level as- sessment that uses support vector machines (SVMs) to combine features from statistical language mod- els (LMs), parse trees, and other traditional features used in reading level assessment. The results presented here on reading level as- sessment are part of a larger project to develop teacher-support tools for bilingual education instruc- tors. …

Content truncated to 500 words