A Supervised Deep Learning-based Approach for Bilingual Arabic and Persian Spell Correction

This paper proposes novel approaches for checking and correcting spelling errors in Sinhala essays written by candidates of grade five scholarship examination. They don't have a proper mechanism to identify their spelling mistakes in essays by themselves. Spelling errors by such students may occur due to the violation of spelling rules, missing or adding of letters, missing modifiers, inaccurate spelling in a similar structure, and similar sound letters]. To mitigate such challenges, the Sinhala corpus file has been developed to identify the accurate and inaccurate spellings of the written words. The role of this application is to identify the correct and incorrect words which are entered by the user and generate the most correct words as suggestions for the incorrect words. This paper introduces three new novel approaches to detect the correctly spelled words in Sinhala essays namely object word checker method, suffixes checker method and similar word checker method. With addition to that this paper discusses three approaches to generate accurate suggestions including one novel approach. When evaluating the accuracy of the spelling error detection and correction module the overall results for precision, recall, and the f -measure were recorded as 83.05%, 85.57%, and 86.62% respectively.

Chinese Spelling Check (CSC) aims to detect and correct spelling errors in Chinese. Most CSC models rely on human-defined confusion sets to narrow the search space, failing to resolve errors outside the confusion set. However, most spelling errors in current benchmark datasets are character pairs in similar pronunciations. Errors in similar shapes and errors which are visually and phonologically irrelevant are not considered. Furthermore, widely-used automatically generated training data in CSC tasks leads to label leakage and unfair comparison between different methods. In this work, we propose a feature (visual and phonological) enhanced siamese BERT to (1) correct spelling errors without using confusion sets; (2) integrate phonological and visual features for CSC by a glyph graph; (3) improve performance for unseen spelling errors. To evaluate CSC methods fairly and comprehensively, we build a large-scale CSC dataset in which the number of samples in different error types is the same. The experimental results show that the proposed approach achieves better performance compared with previous state-of-the-art methods on three benchmark datasets and the new error-type balanced dataset.

Increasingly people use computers for word processing. This helps reduce word processing time and fatigue of hands, but may increase the possibility of occurrence of spelling errors. Although spelling errors are generally easy to find and correct, it is hard to make a document totally free of spelling errors partly due to lack of knowledge of users or presence of spelling errors which are difficult to notice. Since there is no set of online word processing rules and manners in place and problems of spelling errors are not often raised, spelling errors in important documents may lead to decrease in reliability. Even experts cannot correct spelling errors perfectly, so there is a need for research to come up with spelling correction methods for the general public. This study aims to correct spelling errors using Korean alphabet similarity algorithm. To this end, words most similar to misspelled words found in a corpus containing spelling errors collected by previous research were identified to correct spelling errors by measuring frequency of simultaneous appearance with adjacent words.

Scientific writing requires precision and clarity to uphold credibility and effective communication. Errors such as spelling mistakes and typos can compromise the quality and reliability of scientific texts. This study proposes a Long Short-Term Memory (LSTM)-based approach to detect and correct spelling errors, enhancing text accuracy and readability. The dataset comprises 45,698 standard words, supplemented with typo variations to improve model performance. Data is sourced from the Indonesian Dictionary (KBBI) and undergoes normalization and preprocessing to capture diverse error patterns. The model’s performance is evaluated using a confusion matrix, achieving 93% accuracy and high precision, recall, and F1-score metrics. These results demonstrate that the proposed NLP-based LSTM model offers an effective and reliable solution for identifying and correcting spelling errors. This approach significantly enhances the quality of scientific writing, ensuring more transparent and credible communication.

Second language (L2) writing assessments seldom allow a spellchecker and often have a time limit. Naturally, test takers often submit responses with spelling errors. However, little is known about whether and how spelling errors in test taker responses affect trained raters’ scoring decisions. In this study, we investigated the impact of spelling errors on trained raters’ holistic evaluation of response quality. We selected 148 responses to four L2 writing tasks, and created error-free versions of the responses by correcting spelling errors in the original responses. Both the original and corrected responses were randomly assigned to trained raters, who scored the responses according to the same holistic scoring rubrics. We compared the resulting scores of original and corrected responses to gauge the impact of spelling errors. We also examined whether the impact of spelling errors varied across different task types and spelling error characteristics. The results showed that the spelling error correction led to an average of more than a half score point increase. The score gains from the error correction varied across task types and the quantity of corrected errors. These findings have multiple implications, including suggestions for rater training and assessment development.

Rapid increase of the scale of text carries huge costs for manual proofreading. In comparison, automatic proofreading shows great advantages on time and human resource, drawing more researchers into it. In this paper, we propose two attention based deep neural network models combined with confusion sets to detect and correct possible Chinese spelling errors in character-level. Our proposed approaches first model the context of Chinese character embedding using Long Short-Term Memory (LSTM) networks, then score the probabilities of candidates from its confusion set through attention mechanism, choosing the highest one as the prediction answer. Also, we define a new methodology for obtaining (preceding text, following text, candidates, target) quads and provides a supervised dataset for training and testing (Our data has been released to the public in https://github.com/ccit-proofread.). Performance evaluation indicates that our models achieve the state-of-the-art performance and outperform a set of baselines.

Chinese named entity recognition (NER) has been an important problem in natural language processing (NLP) field. Most existing methods mainly use traditional deep learning models which cannot fully leverage contextual dependencies that are very important for capturing the relations between words or characters for modeling. To address this problem, various language representation methods such as BERT have been proposed to learn the global context information. Although these methods can achieve good results, the large number of parameters limited the efficiency and application in real-world scenarios. To improve both of the performance and efficiency, this paper proposes an ALBERT-based Chinese NER method which uses ALBERT, a Lite version of BERT, as the pre-trained model to reduce model parameters and to improve the performance through sharing cross-layer parameters. Besides, it uses conditional random field (CRF) to capture the sentence-level correlation information between words or characters to alleviate the tagging inconsistency problems. Experimental results demonstrate that our method outperforms the comparison methods over 4.23–11.17% in terms of relative F1-measure with only 4% of BERT’s parameters.

In this paper we present the Arib system for Arabic spelling error detection and correction as part of the second Shared Task on Automatic Arabic Error Correction. Our system contains many components that address various types of spelling error and applies a combination of approaches including rule based, statistical based, and lexicon based in a cascade fashion. We also employed two core models, namely a probabilistic-based model and a distance-based model. Our results on the development and test set indicate that using the correction components in cascaded way yields the best results. The overall recall of our system is 0.51, with a precision of 0.67 and an F1 score of 0.58.

Automatic spelling correction for Bangla language is a very important thing as the modern world is almost completely dependent on digital devices where electronic keyboards are used. Bangla is a largely used language (3.05% of world population), worldwide securing the 7th most spoken language. But concerning the natural language processing it is still in its initial stage. Automatic spelling correction of Bangla language is just crossing its primitive stage. A few works have been done on automated Bangla spelling correction but those numbers are not still adequate. This research work is based on correct word prediction of typed Bangla sentences by using edit distance method and fractional accuracy. The results are promising. So, there is a high chance that the proposed model will help and have a great impact on automated Bangla spelling correction.

This research, which considered context-sensitive spelling error correction as the classification problem of words along with the context as the solution of semantic ambiguity, has a limitation that correction can be corrected in a particular words pair. To overcome this, this research suggested a technique for detecting and correcting context-sensitive spelling error probabilistically, by selecting the whole eojeol1 as the target words and generating corresponding available candidate. The context-sensitive spelling error correction model suggested by this research is a model based on the noisy channel. It utilized eojeol n-gram model as a language model, and the context-sensitive spelling error rate, the channel probability, was utilized as an equipment for user adaptive context-sensitive spelling error detection and correction. This research composed an evaluation data by transforming eojeol per sentence from 2,000 correct sentences randomly extracted from Sejong corpus to error eojeol. The test result showed function, i.e. precision over 95%, and recall over 70%. This showed the detection and correction method suggested by this research could be applied in the extremely unpredictable error generating language environments.

Automatic Parts of Speech(POS) tagging is one of the most fundamental tasks for a language in Natural Language Processing(NLP), which acts as a feature for solving advanced NLP tasks. Named Entity Recognition(NER) is another essential task of NLP for information retrieval. Researchers could not find up to the mark solution yet on these two tasks for Bangla language compared to other languages, for instance, English, Ger-man. Moreover, many solutions heavily depend on handcrafted features that require strong linguistic expertise. As these two sequence labeling tasks are similar, In this work, two different datasets of POS tagging and NER were prepared, and different deep neural network approaches studied for solving these two tasks separately. All of the approaches were end to end and did not need any handcrafted feature like word suffixes or affixes, gazetteers, dictionary. This study came up with an end to end solution using deep neural network-based model consisting of Bi-directional Long short-term memory(BLSTM), Convolutional Neural Network(CNN) and Conditional Random Field(CRF). The proposed model trained on respected datasets achieved an accuracy of 93.86% on POS tagging and a strict f1 score of 0.6285 on NER on prepared datasets, respectively.

Sequence modeling is a fundamental problem in Natural Language Processing (NLP) due to the sequential nature of language. With advances in Recurrent Neural Networks, this family of models has become the state-of-the-art for many sequence mapping tasks. In this thesis, we examine the effectiveness of this family of models for several classes of problems in NLP.

A sequence-to-sequence learning with neural networks has empirically proven to be an effective framework for Chinese Spelling Correction (CSC), which takes a sentence with some spelling errors as input and outputs the corrected one. However, CSC models may fail to correct spelling errors covered by the confusion sets, and also will encounter unseen ones. We propose a method, which continually identifies the weak spots of a model to generate more valuable training instances, and apply a task-specific pre-training strategy to enhance the model. The generated adversarial examples are gradually added to the training set. Experimental results show that such an adversarial training method combined with the pre-training strategy can improve both the generalization and robustness of multiple CSC models across three different datasets, achieving state-of-the-art performance for CSC task.

Mongolian is an alphabetic writing, its unique way of spelling often produces some words seemingly written correctly but misspelled in reality. This paper analyzes the main factors of causing Mongolian word spelling errors, and presents a method based on combination of dictionaries and statistical language model to achieve the spelling error correction of Mongolian words. The corrected Mongolian words are stored with Unicode.

Large language models excel in text generation and generalization, however they face challenges in text editing tasks, especially in correcting spelling errors and mistyping.In this paper, we present a methodology for generative spelling correction (SC), tested on English and Russian languages and potentially can be extended to any language with minor changes.Our research mainly focuses on exploring natural spelling errors and mistyping in texts and studying how those errors can be emulated in correct sentences to enrich generative models' pre-train procedure effectively.We investigate the effects of emulations in various text domains and examine two spelling corruption techniques: 1) first one mimics human behavior when making a mistake through leveraging statistics of errors from a particular dataset, and 2) second adds the most common spelling errors, keyboard miss clicks, and some heuristics within the texts.We conducted experiments employing various corruption strategies, models' architectures, and sizes in the pre-training and fine-tuning stages and evaluated the models using single-domain and multi-domain test sets.As a practical outcome of our work, we introduce SAGE 1 (Spell checking via Augmentation and Generative distribution Emulation).