english essay grader

english essay grader

Improving English Essay Grading System

1. Introduction

In academia, it has been widely recognized that writing skill has been becoming the key to success in doing academic research. However, it is a time-consuming task in grading essays manually. Therefore, automated essay scoring (AES) systems have aroused much attention to help reduce time of updating students with feedback. Although any attention on writing cannot contribute to improvements in writing ability, researchers have put efforts on the models for improvements of AES systems. The mentioned problem in past few years can be roughly divided into two categories. Models based on linguistic features are popular among commercial systems. Despite efficiency, the scoring accuracy is not satisfactory. To improve grading, researchers have attempted the usage of more sophisticated modeling with linguistic features from writings.

The problem of automated essay scoring (AES) has attracted research in both academia and industry for its significant economic impacts. Although it has been proven to work effectively and efficiently in grading, the scoring accuracy is still not satisfactory. AES systems now mainly focus on linguistic features of essays regardless of the ability to grasp the written topics. In this paper, we introduce a new approach by combining both linguistic features and topical modeling for essay scoring. We propose to use Auxiliary Document Plus keyPhrase model for the topic modeling to understand documents and obtain more useful distributions. Experimental results of slot-filling essays show that our model outperforms the previous methods.

2. Current Challenges

Third, inconsistency among the raters is the reason for doubts in Educational Measurement. This reflects singular stealing of raters, in addition to the fact that scoring irregularities and scoring methods commonly used for the free result tests are simpler to be controlled than human scoring. Fourth, there have been cases found in which more than one teacher has discrepancies with the grades of the essay although it is perceived from the essays that there is agreement between two teachers. Lastly, teachers might experience the burden of grading whether a particular paper is to deserve a 2 or a 3, and eventually one rater might fairly give a medium score of 2.5 taking into viewpoints of the others, leading to a long discussion. Yesudhas states that in most essay examinations, the grades should be led by more than one rater to enhance the reliability of the grading result. In the actual grading situation, different raters have their own knowledge, based on their personal experience that they use in assessing the essays, since there is often disagreement occurred between them.

There are several challenges that teachers usually face in grading English essays, namely subjectivity, time-consumption, inconsistency, and disagreement. First, grading essays is less than accurate due to the subjective nature of reading and interpretation. Teachers have their own reading process and, at times, have their own criteria of what they look for among the students. Guo et al further adds that reading strategies applied by teachers have a great influence on teachers’ decisions and they also elicit prior expectations in themselves. Second, the process of reading essays is very time-consuming. Gómez, Fernández, Huertas, and Trenas point out that the process of grading examinations is one of the least exciting and time-consuming tasks that any teacher faces in his/her professional life. According to most teachers, reading and marking essays can take a considerable amount of time to accomplish.

3. Proposed Solutions

In terms of model architecture, many variations have been made to the supervised structure of the models discussed up to this point, which have already been discussed in Section 3. With some modifications in the scoring function, it can be used for unsupervised representation learning as well. Variational Encoders give a continuous latent variable representation of the input, and such representations can be more naturally compared using attention, and can thus be well used in metrics like BLEU. Apart from this enhancement, the results have also shown that qualities which have been inferable from the continuous representation like sentence fluency and coherence have been able to be captured. Finally, the corrected relative position of words in a sentence can also be exploited to make these models work better. These observed improvements are also positive to the fact that the models that exploit the similar information structure in the AE system can also perform better than the ones which do not.

Several proposed solutions and improvements have been suggested to the current problem in automated essay grading, which are discussed below. One of the most common techniques that have been proposed is to use external features such as the number of words or number of unique words in the essay. Apart from these features, features that come from the linguistic structure of the essay, in the form of n-grams or word vectors, etc., can also be tested. Given the rapid and significant success of neural language models in a variety of NLP tasks, these models are a candidate for the features that can be used in the AE system. A novel feature extraction model is presented that can take advantage of the sentence structured information in the inputs. The inputs to the system are broken down into sentences and passed on to the encoder, which is trained to predict the sentence-level vector representations. The final vector representations are then used to generate a score. It has been proposed that such models of intermediate latent variable predictions have the right structure for understanding the linguistic information in the sentence level, which can be used in other tasks.

4. Benefits of Upgraded System

Second, manual grading takes time and energy. The upgraded grading system is more efficient; it measures students’ work more quickly and effortlessly. There are non-negligible savings in non-teaching personnel needed for grading employed either full-time or part-time. Precisely handled saved funds are those of workplace maintenance (salaries and benefits); insurance benefits for the grading personnel; and potential bilaterally rising coverage by health and retirement plans, once the days spent grading essays mount. Finally, the results transmitted to postgraduate institutions or potential employers show no grades, but essays and per segment performance measures. They will be able to make the judgments on the lure (talentedness, rigor, soundness, organizing powers, cognitive skills), not the easiness of college education. Also, the quality essays will be noted on the list and valued, given the need to select their future employees. Therefore, students not only have got excitement of demonstration essays of their hidden talents, but they have also attracted attention from the wider audience and closer to more successful continuation of their educational path after college.

The very first and most important benefit of an upgraded system is its objectivity. The upgraded system does not take sides, and it does not have its own imperative. As a consequence, based on this grading system, the outcome is strictly merit-based, and only a good essay earns a good grade, not only good impression. This results in a more fair evaluation of particular segments of students’ work – analytic, academic language, vocabulary, content and organization of the essay. It profiles students in all those areas, providing them a measure to which they can be destined after having mastered the language habits and the form of an essay.

5. Conclusion

To improve the grading system for essays, especially those that examine students’ English language competence, researchers need to pay attention to the areas tested in the ELPT, including content and language use, especially the organization of essays. Essays created are marked by graders who have been trained to mark essays according to the official ELPT marking scheme. There will be many students taking the ELPT in the coming years. This will be a burden for teachers or assessors to mark and grade the essays. If there is a contradiction between assessments, essays must be sent to a head assessor or team of head assessors for final scoring. If students cannot take ELPT, especially ELPT tests during the COVID-19 pandemic, there must be an alternative to overcoming these obstacles, which include online assessments. Currently, work is being done on an AI model for the ELPT written test. AI uses machine learning to interpret and understand essays, and makes changes to the weight of the main ideas and supporting ideas and the use of cohesive devices or connectors.

Improving English essay grading system. Writing is one of the most important skills in every language. This is because writing can communicate the knowledge and information of the writer to others. As in the world, English is the most widely used international language, so everyone must master English as the only foreign language by taking and learning it in schools or universities. However, some students are less enthusiastic about studying English, especially on writing assessments. Reading, understanding, brainstorming, creating an essay structure, writing an essay, and rewriting the essay for error correction, students must repeat these stages every time they want to write an essay and are graded on various aspects. If there is no motivation or good ways and processes, students will be forced to do this assignment.