Semantic Similarity Based Approach for Automatic Evaluation of Free Text Answers Using Link Grammar

Most approaches towards automatic evaluation of free text answers are keyword centric. Though keywords essentially reflect and represent the primary concept coverage of an answer, they are incomplete without the associated texts. The words occurring before and after the keywords bring out the true meaning. The work presented in this paper proposes a semantic similarity based approach for evaluation of free text answers where both keywords and the associated text contribute to the score. A link grammar based approach is used to extract the keywords from the answer and through the process of identifying and extracting the relational expressions for a keyword the model automatically evaluates a learners' free text response SA to a given question Q with respect to a model answer MA. The score of the automated system have a high co-relation with human evaluator score as can be found from the reported results.