Search Results (4)

The posting of offensive content in regional languages has increased as a result of the accessibility of low-cost internet and the widespread use of online social media. Despite the large number of comments available online, only a small percentage of them are offensive, resulting in an unequal distribution of offensive and non-offensive comments. Due to this class imbalance, classifiers may be biased toward the class with the most samples, i.e., the non-offensive class. To address class imbalance, a Multilingual Translation-based Data augmentation technique for Offensive content identification in Tamil text data (MTDOT) is proposed in this work. The proposed MTDOT method is applied to HASOC’21, which is the Tamil offensive content dataset. To obtain a balanced dataset, each offensive comment is augmented using multi-level back translation with English and Malayalam as intermediate languages. Another balanced dataset is generated by employing single-level back translation with Malayalam, Kannada, and Telugu as intermediate languages. While both approaches are equally effective, the proposed multi-level back-translation data augmentation approach produces more diverse data, which is evident from the BLEU score. The MTDOT technique proposed in this work achieved a promising improvement in $F_1$-score over the widely used SMOTE class balancing method by 65%. Full article

In response to international agreements, recent Indian legislation has raised expectations that all children, regardless of need or ability, should gain access to formal education that is inclusive and addresses their social and learning needs. Initiatives designed to support the implementation of this legislation have been undertaken in several parts of India. Reports related to such initiatives have largely focused upon developments in large urban connotations, with studies in rural areas being less in evidence. This paper reports a small-scale study conducted in Telangana a state in the south-central part of India. Through the application of semi-structured interviews data were obtained to enable a comparison to be made of the experiences of two purposive samples of families of children with disabilities and special educational needs, and the professionals who support them. The first sample was located in Hyderabad, a large metropolitan city, the capital of Telangana State. The second was situated in villages in Sangareddy, a single rural district of the same state. Interviews were conducted either in English or in Telugu, the state language with all interviews transcribed and subjected to thematic analysis. The findings, which will be used to support further development in the area, reveal a willingness on the part of professionals to support the education and social welfare needs of children with special educational needs and their families and an awareness of current national legislation aimed at achieving this objective. A disparity exists between the availability of professional support services available to families and children, with those living in the rural district experiencing greater difficulty in accessing appropriate support than their counterparts in the metropolitan city. The lack of opportunities for training and professional development is perceived to be a major obstacle to the progress of inclusive education as required by national legislation in both locations. Recommendations are made for further research that is closely allied to changes in practice, for the development of professional development of teachers and other professionals, and for the development of centralised provision in rural areas to address the needs of families. Full article

Named entity recognition (NER) is a fundamental step for many natural language processing tasks and hence enhancing the performance of NER models is always appreciated. With limited resources being available, NER for South-East Asian languages like Telugu is quite a challenging problem. This paper attempts to improve the NER performance for Telugu using gazetteer-related features, which are automatically generated using Wikipedia pages. We make use of these gazetteer features along with other well-known features like contextual, word-level, and corpus features to build NER models. NER models are developed using three well-known classifiers—conditional random field (CRF), support vector machine (SVM), and margin infused relaxed algorithms (MIRA). The gazetteer features are shown to improve the performance, and theMIRA-based NER model fared better than its counterparts SVM and CRF. Full article

Multilingualism has become an integral part of our present lifestyle. India has twenty two registered official languages with English and Hindi being most widely used for all official activities across the nation. As both these languages are introduced later in life, it was hypothesised that comprehensive reading will be better and faster if the native medium was used. Therefore present study aimed to evaluate the differences in performance while using one of the four Indian Dravidian vernaculars (Tamil, Telugu Kannada and Malayalam) and two nonvernacular (English and Hindi) languages for onscreen reading task. A multidimensional approach including physiological (Eye movement recording), subjective (Language Experience And Proficiency Questionnaire, LEAP-Q, Legibility rating) and Objective (Reading time and Word processing rate) measurements were used to quantify the effects. Forty-four Indian infantry soldiers from each of the Dravidian language groups participated in the study. Volunteers read aloud two simple story passages onscreen in their respective vernacular and non-vernacular languages using both time bound and self-paced reading mode. Reading time was lower and word processing rate was higher respectively in case of vernacular than non-vernacular. Consideration of fixation count in both the modes of reading indicated better performance with vernaculars. Legibility score was better in Dravidian languages than others. Results indicated that reading text was faster in vernacular media followed by English and Hindi. Use of vernaculars in onscreen text display of high density workstation may therefore be recommended for easier and faster comprehension. Full article