Adaptive Feedback Message Generation for Second Language Learners of Arabic

Proceedings of Recent Advances in Natural Language Processing, pages 752–757, Hissar, Bulgaria, 12-14 September 2011. Adaptive Feedback Message Generation for Second Language Learners of Arabic Khaled Shaalan The British University in Dubai, PO Box 345015 Dubai, UAE khaled.shaalan@buid.ac.ae Marwa Magdy Faculty of Computers & Information, Cairo University, 12613 Egypt m.magdy@fci-cu.edu.eg Abstract This paper addresses issues related to generat- ing feedback messages to errors related to Ar- abic verbs made by second language learners (SLLs). The proposed approach allows for in- dividualization. When a SLL of Arabic writes a wrong verb, it performs analysis of the input and distinguishes between different lexical er- ror types. The proposed system issues the in- telligent feedback that conforms to the learn- er’s proficiency level for each class of error. The proposed system has been effectively evaluated using real test data and achieved sat- isfactory results. 1 Introduction Second language acquisition is a difficult task. There are various methods to acquire a new lan- guage and all of them require some form of feedback, a reaction to what has been said or written. The recent trend is to automate the feed- back through Intelligent Language Tutoring Sys- tem (ILTS). The current trend concentrates on NLP tools and techniques geared towards the diagnosis of errors produced by SLLs and identifying the cause of their errors rather than providing the correct version directly. This paper is about the generation of feedback message based on individual proficiency levels. The proficiency level measure is based on the progression in the learner answers. In particular, when a SLL of Arabic writes a wrong verb, it distinguishes between this set of lexical error types: lexical category selection, pattern selec- tion, tense selection, mood selection, subject- verb agreement, verb conjugation, connected pronouns and/or consonant, and vowel letters. Nevertheless, it provides the intelligent feedback that conforms to the learner’s expertise for each class of error. There are three learning levels for each concept covered: beginner, intermediate and advanced. A learner who generally has mas- tered an Arabic concept might receive a hint just indicating the class of error. Whereas, the learn- er who generally knows the concept but still needs practice in its application the feedback is the type of the error. For the beginner learning level, the feedback is as specific as possible, the exact source of the error is provided. The edit distance technique is employed to an- alyze the erroneous Arabic verb. The deep analy- sis of the learner input helps in accurately detect- ing the lexical errors and issuing the appropriate feedback to the learner. To the best of our knowledge, very few re- searches has considered true diagnosis and issu- ing feedback of Arabic lexical errors. For exam- ple, Shaalan (2005a; 2005b) has developed an ILTS system for Arabic learners which just em- bed specific morphological analysis rules to pro- vide feedback. In addition, there exist some sys- tems that are designed for SLLs of other lan- guages than Arabic which still keep the behavior of spell checkers …

Content truncated to 500 words