Assessing Readability of Learning Materials on Artificial Intelligence in English for Second Language Learners

Abstract

Many scientific publications and materials on artificial intelligence (AI) have been written in English; however, for many AI learners, English is their second language. Therefore, the difficulty (readability) of online self-teaching texts on AI for English-as-a-second-language (ESL) learners is essential for determining the language support ESL learners need to learn AI. However, only a few studies have addressed this issue. Therefore, we identified the difficulty level of English self-teaching texts for ESL AI learners. Because large-scale testing for ESL learners is impractical owing to the financial costs and time involved, we built two distinctive automatic readability assessors: one using sophisticated deep-learning-based natural language processing (NLP) technology, and another using classic NLP based on word frequency and applied linguistics. We conducted our evaluation using AI research papers and university-level online course texts. Interestingly, the distinctive automatic assessors, which were trained on different datasets, showed similar results. Intermediate-level ESL learners could read approximately 10% of online course texts. We also showed that they are significantly easier to read than AI research papers for ESL learners, demonstrating their usefulness in AI learning.