Automatic Generation of Fill-in-the-Blank Questions From History Books for School-Level Evaluation

Abstract

Fill-in-the-blank questions (FIBs) play an important role in educational assessment. FIBs are effective to assess the understanding of well-defined concepts, and these are often used in school level. But manual preparation of FIBs is time-consuming and requires sufficient expertise on the content. This paper presents the proposed system for automatic generation of FIB questions that accepts school textbook as input. First, we identify the informative sentences that can act as the basis of FIBs. A parse structure-based module works on the sentences to identify the concept or knowledge embedded in the sentence. The knowledge is extracted in form of subject–predicate–object triplet or expanded triplet. Then, a hybrid algorithm chooses the most appropriate word/phrase that can be marked as a gap. Proposed system is tested using class VII-level history textbook as input. The quality of the system generated questions is then evaluated manually using three defined metrics. Experimental result shows that the proposed technique is quite promising.