Checking essays written by students is a time-consuming task. In addition to detecting spelling and grammar errors, educators also need to report predictive segmentation. This study focused on one such aspect of predictive segmentation of essays by analysis of errors in text. The study used machine learning techniques to predict segmentation in essays. The data set consisted of essays written by students in L2 (second language). The study explored and implemented various embedding techniques proposed in recent years. The results of the study showed that it was possible to predict distinctive student errors separately with zero human intervention using currently available tools and techniques. framework will be based on Model of Error Analysis by Corder (1974) as citied in Sarfraz, S. (2011). 1387 essays were collected on 20 various prompts/topics from two private sector universities and three public sector colleges undergraduate students. Off-topic and On-topic detection was developed by convolutional neural networks techniques TFIDF and Latent semantic analysis. Syntactical error analysis model was integrated by LanguageTool open-source library for error detection. This automated error analysis system could be very useful for checking and analyzing text that is mainly submitted in soft form.

Submitted in fulfillment of the requirements for the Master degree of MS (Linguistics) in the Department of Sciences and Humanities.

TABLE OF CONTENTS ABSTRACT TABLE OF CONTENTS 1. INTRODUCTION 1.1 Background 1.2 Research Problem 1.3 Research Questions 1.4 Research Objectives 1.5 Significance of the Study 1.6 Scope of the Research 1.7 Conceptual Framework 2. LITERATURE REVIEW 2.1 Previous Researches 2.2 Writing 2.3 Descriptive Essay 2.4 Error and Mistake 2.5 Error Analysis 2.6 Corpus Based Error Analysis 2.7 Automated Essay Scoring 2.8 Convolution Neural Network 2.9 Word Vectorization 3. METHODOLOGY 3.1 Research Design 3.2 Model 3.3 Data Collection and Preprocessing 3.4 Labelling 3.5 Preprocessing Data 3.5.1 Tokenization 3.5.2 Case Normalization 3.5.3 Punctuation Removal 3.5.4 Stop Word Removal 3.5.5 Lemmatization 3.6 Corpus Selection for Model Training 3.7 Corpus Selection for Model Testing 3.8 TF-IDF for Off-Topic and On-Topic Detection 3.9 Latent Semantic Analysis (LSA) for Off-Topic and On-Topic Detection 3.10 Combining TF-IDF and LSA 50 3.11 Evaluation Metrics 3.12 Data Collection and Corpus Preparation 3.13 Integration and Customization of LanguageTool 3.14 Evaluation Metrics and Framework 3.15 Experimental Design 4 FINDINGS 4.1 Dataset Description 4.2 Testing Setup 4.3 Evaluation Results and Analysis 4.4 Comparison with Baseline Methods 4.5 Accuracy and Precision of LanguageTool 4.6 Error Detection Performance Casing Errors 4.7 Typos 4.8 Grammar Errors 4.9 Punctuation Errors 4.10 Miscellaneous Errors 4.11 Comparison with Other Tools 4.12 Limitations and Challenges 4.13 Demonstration of Prototype 5 DISCUSSION 5.1 Interpretation of Results 5.2 Comparison with Baseline Methods 5.3 Representation and Semantic Meaning 5.4 Dataset Quality and Size 5.5 Choice of Classifiers 5.6 Advanced NLP Techniques 5.7 External Knowledge Sources 5.8 Ensemble Methods and Hybrid Approaches 5.9 Data Considerations 5.10 User Studies and Human Feedback 5.12 Integration into model 5.13 Potential Benefits and Implications 5.14 Future Directions 5.15 Advanced Error Detection Techniques 5.16 Customization and Personalization Features 5.17 Longitudinal Studies on Writing Improvement 6 CONCLUSION REFERENCES. .