Devanagari Research Articles

MMHFND: Fusing Modalities for Multimodal Multiclass Hindi Fake News Detection via Contrastive Learning

Multimodal content contains more deception than unimodal information, causing significant social and economic impacts. Current techniques often focus on single modality, neglecting knowledge fusion. While most studies have concentrated on English fake news detection, this study explores multi-modality for low-resource languages like Hindi. This work introduces the MMHFND model, based on M-CLIP, which uses late fusion for coarse (Fake vs Real) and fine-grained (World vs India vs Politics vs News vs Fact-Check) configurations. We extract deep representations from image and text using image transformer ResNet-50, a BERT-based L3cube-HindRoberta text transformer handling headlines, content, OCR text, and image captions, paired M-CLIP transformers, and an ELA (Error Level Analysis) image forensic method incorporating EfficientNet B0 to analyse multimodal news in Hindi language based on Devanagari script. M-CLIP integrates crossmodal similarity mapping of images and texts with retrieved multimodal features. The extracted features undergo redundancy reduction before being channelled into the final classifier. The MAM (Modality Attention Mechanism) is introduced, which generates weights for each modality individually. The MMHFND model uses a computed modality divergence score to identify dissonance between modalities and a modified contrastive loss on the score. We thoroughly analyse HinFakeNews dataset in a multimodal context, achieving accuracy in coarse and fine-grained configurations. We also undertake an ablation study to evaluate outcomes and explore alternative fusion processes on three different setups. The results show that the MMHFND model effectively detects fake news in Hindi with an accuracy of 0.986, outperforming other existing multimodal approaches.

A Recognition System for Devanagari Handwritten Digits Using CNN

A Recognition System for Devanagari Handwritten Digits using CNN, a novel approach to recognizing transcribed digits in the Devanagari script using Convolutional Neural Networks (CNN). This framework represents a significant contribution to the field of pattern recognition and language processing objective of the research project is to perform a literature review, identify an algorithm for a digits recognition system implement the Devanagari digits recognition system for educational activities. In the first phase, a dataset of 150 transcribed digit images is curated, allocating 75% for training (113 images) and 25% for validation (37 images). A Convolutional Neural Network (CNN) is designed with five convolutional layers, each utilizing 3 × 3 filters with 16, 32, 64, 128, and 128 feature maps, respectively. The experiments conducted involve varying the number of epochs, with results captured at 5, 10, 20, and 100 epochs. This comprehensive evaluation aims to understand the model's convergence and performance over different training durations. The outcomes of this phase contribute to the fine-tuning and optimization of the model for subsequent phases. In the second phase, the dataset is expanded to 100*10 (1000) images, each resized to 28 × 28 pixels through cropping. The CNN architecture remains consistent, with the previously determined layer configuration. Similar experiments are conducted, assessing the model's performance over 5, 10, 20, and 100 epochs. This model with a data size of 1000 demonstrates superior accuracy (100% on mini-batches) compared to the 150 model, with consistently high validation accuracy, while both models exhibit decreasing trends in mini-batch and validation losses, favoring the larger dataset, and maintaining a constant learning rate at 0.0100, albeit with a slightly longer time elapsed for each epoch due to the increased data size. 98.37398 accuracy in the phase 2 experiment in 100 epochs. Similar research and contributions and Devanagari’s character and word recognition system.

Deep learning for ancient scripts recognition: A CapsNet-LSTM based approach

Efficient character recognition in ancient handwritten Devanagari documents is crucial for societal advancements. Challenges such as overlapping characters, missing headlines, and over-inked stains further complicate the recognition process. In response, we propose a Capsule Network (CapsNet) with LSTM to address hierarchical temporal dependencies in Devanagari scripts, following initial implementation of a simple CNN. We also explored a combined CNN+LSTM architecture for character recognition, leveraging CNN’s feature extraction capabilities with LSTM’s sequential modeling to handle temporal dependencies in Devanagari scripts. Our experimentation involved a dataset of 10,825 characters from ancient Devanagari manuscripts, encompassing basic characters, modifiers, and conjuncts, classified into 399 classes. Testing various training–testing ratios (9:1, 8:2, and 7:3), we visually and statistically evaluated the experimental data, demonstrating the superiority of CapsNet and LSTM in handling these challenges. We calculated recognition accuracy, precision, and recall values, with CapsNet achieving a maximum accuracy of 95.98% after 30 epochs. This research underscores the effectiveness of CapsNet and LSTM in advancing character recognition for ancient Devanagari manuscripts.

Profanity Filter and Hate Speech Detection for Hindi Language

In recent years, the expansion of online communication channels has highlighted the critical need for efficient content moderation technologies to address the growing flood of hate speech and vulgarity. This difficulty also applies to the Hindi language, which is spoken by millions of people throughout the world. This abstract provides an overview of cutting-edge methodologies and improvements in profanity filtering and hate speech identification, with a focus on the Hindi language. Profanity and hate speech in digital communication may have negative consequences for individuals and groups, propagating discrimination, harassment, and antagonism. Addressing this issue in Hindi, a language with significant cultural and linguistic variety, presents particular obstacles. The intricacy of Hindi script, dialect variances, and the context-dependent nature of hate speech complicate the work impossible

Transliterated multilingualism/globalisation: English disguised in non‐Latin linguistic landscapes as new type of world Englishes?

AbstractEnglish has, for historical reasons, risen to global prominence as the unchallenged lingua franca internationally. World Englishes (WE) has, as a result, established itself as a visible line of research, exploring localised/indigenised varieties of English from around the world (e.g. India, Sri Lanka, Pakistan, Hong Kong, Singapore, Nigeria). However, most of the discussions so far concern English written in the Latin script as people would normally expect. Against a backdrop of globalisation and the juggernaut of English, this article points to an increasingly salient phenomenon that English especially in superdiverse and/or (post)colonial societies (e.g. India and Pakistan) may disguise in seemingly inscrutable and ‘mysterious’ local scripts (e.g. Perso‐Arabic script and Devanagari script) and even ‘pass off’ as local languages in these countries’ linguistic landscapes through phonetic transliteration. This emerging trend begs the question whether these should be understood as new varieties of local languages or new kinds of world Englishes disguised in non‐Roman scripts. This phenomenon is theorised in this paper conceptually. To illustrate our point, examples of authentic signs taken from the linguistic landscapes relating to South Asia and South Asian communities are discussed. As English is increasingly glocalised and becomes part of other less dominant languages, this article calls on researchers in World Englishes (WE) and (socio)linguistics in general to look beyond English written in the Latin script in a conventional/traditional sense and to expand the scope and remit of WE research to explore how English, as a dominant code, becomes indigenised using local scripts and morphs into and even ‘passes off’ as ‘local’ surreptitiously. This fundamentally calls for the crucial need for researchers from diverse and multilingual backgrounds to work together to better understand English and other non‐dominant languages’ role in the 21st century.

Ensemble Classifier for Hindi Hostile Content Detection

Detection of hostile content from social media posts (Facebook, Twitter, etc.) is a demanding task in the field of Natural Language Processing. The increase of hostile content in different electronic media has opened up new challenges in language understanding. It becomes more difficult in regional languages. AI-based solutions are required to identify hostile content on a large scale. Although a satisfactory amount of research has been carried out in the English language, finding hostile content in regional languages is still under development due to the unavailability of suitable datasets and tools. In terms of the number of speakers, Hindi ranks third in the world and first on the Indian subcontinent. The objective of this article is to design a hostile content detection system in Hindi using coarse-grained (binary) classification and fine-grained (multi-class, multi-label) classification. We note that different baseline learning methods with different pre-trained language models perform differently. Using the Constraint 2021 Hindi Dataset, this research proposes a Bidirectional Encoder Representations from Transformers–(BERT) based contextual embedding technique with a concatenation of emoji2vec embeddings to classify social media posts in Hindi Devanagari script as hostile or non-hostile. Additionally, for the fine-grained tasks where hostile posts are sub-categorized as defamation, fake, hate, and offensive, we develop an ensemble classifier varying different learning methods and embedding models. With an F1-Score of 0.9721, it is found that our proposed Indic-BERT+emoji model outperforms the baseline model and other existing models for the coarse-grained task. We have also observed that our proposed ensemble method provides better results than the existing models and the baseline model for the fine-grained tasks with F1-Scores of 0.43, 0.82, 0.58, and 0.62 for the defamation, fake, hate, and offensive classes, respectively. The code and the data are available at https://github.com/skarifahmed/hostile .

A new approach to implement stream cipher using memristor-based Chua’s circuit for secure communication

The advancement in cryptographic algorithms is the primary need for modern secure communication systems. In secure communication system, the stream cipher serves the basis for cryptographic algorithms. The properties of cryptographic algorithms are often analyzed on the basis of randomness involved in implementing cipher texts while complexity of the hardware is also taken care of. Therefore, the main objective of this paper is to propose a new stream cipher scheme having higher complexity (high degree of randomness) while keeping the hardware as simple as possible. The proposed stream cipher algorithm gives the new direction to develop the highly secure and simple models in communication field. The proposed scheme offers the benefit of adjusting variable-length key(s) according to the variable length of plaintext using modular arithmetic and memristor—based chaotic behavioral Chua’s circuit. To explain the encryption–decryption process at transmitter and receiver ends, the Devanagari script has been taken. Although, any other script or data type is supported by the proposed algorithm. In addition, the synchronization is also achieved without having any requirement of additional process.

OCR for devanagari script

The introduction of Optical Character Recognition (OCR) technology revolutionized text digitization, allowing physical documents to be converted into editable and searchable digital representations. This paper goes into the unique challenges and advancements in OCR designed exclusively for the Devanagari script. It helps to preserve cultural heritage by digitizing historic manuscripts and religious writings and making them more accessible. Furthermore, Devanagari OCR has practical uses in administrative activities, data entry, and educational content digitization.

Empirical review on handwritten Devanagari script recognition techniques using AI approaches

Empirical review on handwritten Devanagari script recognition techniques using AI approaches

Empirical review on handwritten Devanagari script recognition techniques using AI approaches

Empirical review on handwritten Devanagari script recognition techniques using AI approaches

Exploring Gender-Specific Features from Handwriting for Devanagari Script

Handwriting is a very important mode used for communication. It has been in use for a long time for various purposes. Handwriting is found in many criminal cases and its analysis is quite tedious work to do and even time-consuming. But it would be better if the gender of the writer is known; it would be helpful in narrowing the search of the suspects. In this work, we have tried to explore certain features which could be used for the identification of the gender of the writer. Various researches have been done to determine gender from handwriting in various languages. No such work has been done on any of the Indian languages yet. We have selected the Marathi language, the script used for writing is the Devanagari script, which is the script for some other languages used in India. The database was created and the features were extracted manually including line quality, pen pressure, formation of dots, and many more. However, few features are seen in males while some others are in females. The study is still not conclusive and looking for further work on it.

Contextual information based segmentation and recognition of upper modifiers from Devanagari script

Contextual information based segmentation and recognition of upper modifiers from Devanagari script

The Question of Script for Sindhi in India: Reflections on Postcolonial Grammatology

When Sindhi Hindus came to India after the 1947 Partition, they had little to help them survive as a community. Given the linguistic organization of states in independent India, the community has been striving to forge an identity comparable to other communities that have a state/territory they can flourish in. First, Sindhis struggled to gain recognition for their language as an official language listed in the Eighth Schedule of the Constitution of India. Second, they demanded that entertainment content be broadcast in Sindhi in official national media spaces. The case of Sindhi stands as a fascinating case study at the intersection of ideas such as nationalism, citizenship, and minority identity. The case of Sindhi is also a narrative of self-transformation, one of which is its struggle for survival that has also led to the revival of the question of its script. In the 1960s, a faction among the Sindhi intelligentsia proposed that in order to stay relevant and alive in India, it must adopt the Devanagari script and give up its Perso-Arabic script associated with the language since the nineteenth century. In this essay, I revisit this debate to uncover postcolonial grammatology as an approach to deal with South Asian sites of language and writing.

KESAMAAN BUNYI BAHASA TULIS AKSARA LAMPUNG DAN AKSARA BATAK TOBA SUMATERA UTARA

This study aims to describe the similarities in the sounds of written language in the Lampung script and the Toba Batak script of North Sumatra. The method used is descriptive qualitative method. The results of the study show that there are several similarities in the sounds of written language in the Lampung script and the Toba Batak script. All the main letters of the Lampung script and Toba Batak script end in [a]. The entire sound of the Toba Batak script is exactly the same as the sound of the Lampung script, only the sound [ka] in the Lampung script is not found in the Toba Batak script. The sound of the children of this letter in the Batak Toba script [ng] and [h] has the same sound as the children of the Lampung script, namely Tekehole [ng] and Kelengiah [ah]. In conclusion, the similarities in the sound of written language in the Lampung script and the Toba Batak script can be seen from a historical background that both originate from India. The Lampung script comes from the Devanagari script and the Toba Batak script from Brahmi writing (India). Therefore, the Lampung script and the Toba Batak script have many similarities in the sound of their languages.

Cross-language information retrieval for poetry form of literature-based on machine transliteration using CNN

Transliteration is phonetically translating a language’s words into an international or non-native screenplay. The machine translation process now plays an essential role in scholarly research. The most crucial complement criterion of the English translation system is preserving the phonetic qualities of the language specification after English translation in the chosen language. However, a suitable bilingual text corpus is necessary for statistical models to attain improved transliteration accuracy. Marathi-to-English direct machine translation is done through a cross-language information retrieval system using the CNN classifier model in this proposed research. The proposed method considers a sequence labelling issue brought on by the split transliteration units used in the process. All half-consonant clusters in the Devanagari script are effectively mapped as half-consonant “a” s and labelled using the Modified Intermediate Phonetic Code (MIPC). After generating the phonetic units for each feature in the base and aim languages, the weight is assigned to a phonetic unit in both languages, and individual phonetic unit probabilities are computed. If the probability is zero, then segments are established and recalculated for each segment based on the target phonetic unit location in the word. Therefore, the proposed approach classifies the required phonetic unit with a high accuracy rate.

Assamese and BODO next Word Detection using LSTM

The official language of the Indian state of Assam is Assamese, an Eastern Indo-Aryan language. Assamese, the sole native Indo-Aryan language in the Assam Valley, has been heavily impacted by the nearby Tibeto-Burman languages in terms of lexicon, phonetics, and grammar. Its grammar is renowned for its highly inflected forms, and both honorific and non-honorific formulations can include a variety of pronouns and plural nouns. Additionally closely linked to Bengali, Assamese lacks grammatical gender distinctions like Oriya and Bengali. On the other hand, The Bodo language is a variety of dialects of the Tibeto-Burman branch of the Sino-Tibetan languages. Assam, Meghalaya, and Bangladesh are all home to speakers of the Bodo language, which is spoken in northeastern India. It shares linguistic kinship with the Dimasa, Tripura, and Lalunga languages and is written in Bengali, Latin, and Devanagari scripts. Another name for Next Word Prediction is Language Modeling. Predicting what word will be spoken right away requires commitment. It is one of the primary functions of NLP and has a wide range of uses. Our objective is to create this model as quickly and efficiently as possible. RNNs can interpret prior material and forecast words since they have a lengthy short-term memory, which might be useful for users when building sentences. This method creates words by using letter-by-letter prediction, or letter-by-letter prediction. Users can benefit from next word prediction, which makes typing faster and more accurate. The Assamese and Bodo languages rely on next word prediction since multiple characters can be created by pressing the same consonants combined with different vowels, vowel combinations, and special keys. As a result, we present a Long Short Term Memory (LSTM) network model for Assamese and Bodo next word prediction. With 63,300 point sentences, we test the suggested network model, and it achieves 96 per cent accuracy. In addition, we contrasted the suggested model with cutting-edge models like the LSTM. The proposed network model offers a promising outcome, according to experimental findings.

OCR Of Devanagari Script Using CNN

Devanagari script is widely used across India. It forms many languages like Hindi, Marathi, Nepali and Sanskrit languages. As the Devanagari characters are similar to the hindi character the national language of India. It is important to recognize the characters to understand the message that particular tries to tell. The automatic character recognition system is thus developing for the Devanagari script. The character recognition process converts an image of a character into machine-readable format also its English corresponds. In this paper, we are using Convolutional Neural Network for developing the character recognition system. Convolutional neural network learns directly from data. It is a type of Deep learning neural network architecture. CNN is useful as it does not require any human intervention and performs the identification of important features on its own. The proposed paper uses a CNN algorithm applied to a dataset of 49 characters of Devanagari script. The dataset contains of total 4018 Images. The algorithm of the Convolutional Neural Network is applied to train the dataset. The input image to be predicted is first preprocessed and then the model predicts the output result. The system is designed in Jupyter Lab using Python. The Convolutional Neural Network model's overall accuracy is 98%.

Word spotting and character recognition of handwritten Hindi scripts by Integral Histogram of Oriented Displacement (IHOD) descriptor

Word spotting and character recognition of handwritten Hindi scripts by Integral Histogram of Oriented Displacement (IHOD) descriptor

Compact temporal dilated convolution with Channel-wise attention and cost sensitive learning for Single trial P300 detection

The P300 speller is a challenging task due to various factors, such as morphological and temporal variabilities, noisy channels, imbalanced P300/non-P300 data, numerous trainable parameters, and lengthy spelling cycles. To overcome these issues, we propose a novel solution that incorporates temporal dilated convolution with channel-wise attention into a lightweight base classifier and employs cost-sensitive learning to handle data imbalance. We evaluate our system through extensive subject-dependent and cross-subject experiments on two standard datasets: Amyotrophic Lateral Sclerosis (ALS) and Devanagari Script (DS).Our results demonstrate a significant improvement in P300 classification with a 2–3 times reduction in trainable parameters compared to single-trial experiments. Our proposed model outperforms state-of-the-art compact models and offers better trade-off between computational complexity and classification performance.Our approach offers new insights into the challenges of P300 classification and provides inspiration for future research in this field. We are confident that our proposed model will contribute to the advancement of P300 classification and benefit the scientific community.

Devanagari Script based Fancy Vehicle Number Plate Recognition using Machine Learning Approach: A Survey

Devanagari Script based Fancy Vehicle Number Plate Recognition using Machine Learning Approach: A Survey