Editorial for the Special Issue on Computational Linguistics Processing in Low-Resource Indigenous Languages

Many of the indigenous languages today are struggling to survive, and they are in danger of disappearing. Closely followed by Africa, Asia has the most indigenous languages. Though indigenous languages have many sources in existence from where we can obtain acceptable knowledge, mythology, history, and perception of their communities, their diversity is decreasing at an alarming rate due to social pressure, external forces, and demographic changes. The systematic disappearance of indigenous languages threatens the lives of millions of families, children, and indigenous communities as well as the survival of their languages worldwide.

Most indigenous languages have no written form; which makes them difficult to process and analyze using computational models. However, these languages need to be preserved as they are rich in oral traditions, and they remain remarkably consistent and reliable over time.

Presently, many researchers and scientists are actively finding more evidence on indigenous languages to create language processing models using a variety of techniques. Exploring more in terms of grammar, words, and unique rules of sound help us understand the language intuitively and create more efficient linguistic models. However, this process generally tends to be more complex as these languages have very few resources and are often spoken in remote areas by fewer people. Computational linguistics applies computer science techniques for the analysis and synthesis of written and spoken languages. The practical goal of using computational linguistics for indigenous languages is comprehensive. It helps formulate semantic and grammatical frameworks for distinguishing languages through the computationally manageable implementation of semantic and syntactic analysis. Hence, the discovery of more advanced computational linguistics processing algorithms and learning principles that can effectively use the structural and distributional properties of indigenous languages is crucial. It helps develop cognitively and neuroscientifically reasonable computational models that work in the same way that indigenous language processing and learning might occur in the brain.

This special issue was dedicated to explain how computational linguistics and natural language processing algorithms make inferences and gain insights into existing data of low-resource indigenous languages. The content mainly focuses on innovative research that formalizes human communication and spoken indigenous languages into the computational system. We welcomed researchers and practitioners from industry and academia to present their contributions against this background.

This special issue saw a total of 21 submissions, from which five papers were published. It was intentional to adhere to a strict acceptance rate and ensure that only the best papers in the scope of the special issue were accepted. The following few paragraphs summarize the contributions that our special issue collection presents.

In “Context Aware Emotion Detection from Low Resource Urdu Language using Deep Neural Network”, the authors investigate emotion detection in Urdu, a regional language of India and Pakistan. The detection of emotions is crucial in determining an individual's interest in a particular field. Humans typically convey their emotions through gestures, facial expressions, voice pitch, and choice of words. Although significant progress has been made in detecting emotions from textual data in high-resource languages like English, French, and Chinese, emotion classification in low-resource languages such as Urdu has not been extensively studied due to the lack of labelled datasets. In this study, the authors present the Urdu Nastalique Emotions Dataset (UNED), which contains annotated sentences and paragraphs with six different emotions and propose a deep learning-based approach for emotion classification. The authors conducted experiments to evaluate the quality of the UNED corpus and classify it using machine learning and deep learning techniques. The results show that the developed deep learning model outperforms generic machine learning approaches, achieving an F1 score of 85% for the UNED sentence-based corpus and 50% for the UNED paragraph-based corpus. The UNED dataset is publicly available, and the authors hope that it will facilitate further research on emotion detection in low-resource languages.

In “QEST: Quantized and Efficient Scene Text Detector using Deep Learning”, the authors focus on scene text detection using deep learning techniques to make the process efficient and accurate. Due to several environmental constraints, including illuminations, lighting conditions, small and curved texts, and many more, scene text detection techniques are complex. Along with improving the accuracy, the authors have worked on making the model lightweight. Here, two models, namely MobileNetV2 and ResNet50, have been used for model development, whereas post-training quantization has been applied by changing the precision from float32 to float16 and int8. The proposed method outperforms the state-of-the-art techniques by around 30-100 times in terms of inference time and Floating-Point Operations Per Second (FLOPS). Here, well-known datasets, i.e., ICDAR2015 and ICDAR2019, have been utilized for training and validating the performance of the proposed model.

In “Hybrid Deep Learning Model for Sarcasm Detection in Indian Indigenous Language using Word-Emoji Embeddings”, the authors focus their efforts on detecting sarcasm in Indian languages. Detecting sarcasm, irony, and emotions in real-time user-generated text is imperative for fine-grained sentiment analysis. The culturally diverse, country-specific trending topics, hashtags on social media, and the accessibility of aboriginal language keyboards for such applications add to the variety and volume of content in numerous languages and dialects. Automated sarcasm detection is deemed a complex natural language processing task, and extending it to a morphologically rich and free-order dominant indigenous Indian language, Hindi, is another challenge in itself. Furthermore, the lack of benchmark datasets and various analysis tools like POS tagger and dependency parser have restricted the scope of research in sentiment analysis for such low-resource languages. The research demonstrated through this paper, puts forward a hybrid deep learning model trained using two embeddings, namely word and emoji embeddings, to detect sarcasm. The authors evaluate the performance of a Hindi Twitter dataset, Sarc-H, and the results validate that automated feature engineering facilitates efficient and repeatable predictive models for detecting sarcasm in indigenous, low-resource languages.

In “A Framework for online Hate Speech Detection on Code Mixed Hindi-English Text and Hindi Text in Devanagari”, a framework is developed that aims to detect hate speech and other harmful content in code-mixed data and regional scripts and languages, thus identifying individuals who engage in this type of activity on the internet. This can help to make social media platforms safer, particularly in the Indian subcontinent. The proposed model's architecture is optimized through an accuracy-based grid search on hyper-parameters, making it a study of comparative analysis of various algorithms for code-mixed hate speech detection. The model outperforms the baseline models and state-of-the-art models in the sphere of hate speech detection in code-mixed data, with a test accuracy of 90%. The results of the model will be especially impactful in the Indian subcontinent, and the framework can be used by security operators to identify malicious elements on the internet, who can then be stopped from continuing with their harmful activity, thus preventing real-life violence. Overall, this proposed framework is an effective tool that will help in hate speech and threat detection in code-mixed, regional scripts and languages, hence identifying perpetrators of online hate speech and harmful content on the internet more effectively.

In “Joint Intent Detection Model for Task-oriented Human-computer Dialogue System using Asynchronous Training”, the authors first research single modelling of intent detection in spoken language understanding. To effectively capture the deep semantic information input by users, a multi-dimensional feature fusion decoder based on attention and CNN is proposed. Then the authors researched joint modelling of intent detection and slot filling. A joint model of intent detection based on asynchronous training is proposed. It is tested on the Chinese and English datasets and shows that it can effectively improve the accuracy of intent detection.

The editors would like to thank Arriane Bustillo for her countless hours spent on this Special Issue as well as Imed Zitouni, current Editor-In-Chief of ACM Transactions on Asian and Low-Resource Language Information Processing, for responding to all queries and concerns raised by us as editors to ensure the Special Issue was an overwhelming success.