In this paper, we propose a text-to-scene model as one of educational games for Korean language writing education as a foreign language. In order to immediately provide feedback on user input sentences, the proposed text-to-scene model automatically generates the scene corresponding to the user input sentence by analyzing its semantic structure. For the purpose of improving the user s writing ability by correcting errors in the user input sentence, the proposed text-to-scene model provides both the scenes of the correct sentence and the scenes of the user input sentence; so that the user can easily recognize the errors in the user input sentence and correct them. Previous text-to-scene approaches have a weak point that it is difficult to construct the resources and to generate the elaborate scene. Considering that the construction cost and the processing time, the proposed text-to-scene model generates the simple scene that foreigner can easily understand. Experimental results show that the proposed text-to-scene model takes 14.94ms per sentence and produces every scene for 1,048,576 sentences, which is made by arbitrarily combining 15 words. On the paper writing test, a foreigner writes 4.7 incorrect sentences among 10 sentences; while the foreigner writes 0.87 incorrect sentences with the proposed text-to-scene model on the writing test.
1. Introduction
2. The Proposed Text-to-Scene Model
3. Results and Evaluations
4. Conclusion