Neural Semantic Embeddings for Assessing L2 Writing Competency in Education

During the past decades, writing tests in secondary schools were difficult, subjective, and produced different results because they are subjective. This is the reason that the study proposes to use an NLP-based Semantic Embedding Model to automatically grade student writing. The model employs transformer-based systems such as BERT to obtain deep semantic representations of student essays. These representations demonstrate the way of how meaning, coherence, grammar and content are united in a context. These embeddings are then inputted into a regression-based scoring system that has been trained on a set of essays with notes of various genres. The model proposed is more precise and consistent as compared to conventional natural language processing models, as demonstrated by the fact that it is correlated (r > 0.87) with the scores provided by experts. In addition, the model offers useful comments on the architecture and the quality of the language. Lastly, the system offers an equitable and adaptable method of composing school evaluations.

Ҳали таржима қилинмаган