English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

To obtain extensive annotated data for under-resourced languages ischallenging, so in this research, we have investigated whether it is beneficialto train models using multi-task learning. Sentiment analysis and offensivelanguage identification share similar discourse properties. The selection ofthese tasks is motivated by the lack of large labelled data for user-generatedcode-mixed datasets. This paper works on code-mixed YouTube comments for Tamil,Malayalam, and Kannada languages. Our framework is applicable to other sequenceclassification problems irrespective of the size of the datasets. Experimentsshow that our multi-task learning model can achieve high results compared withsingle-task learning while reducing the time and space constraints required totrain the models on individual tasks. Analysis of fine-tuned models indicatesthe preference of multi-task learning over single-task learning resulting in ahigher weighted F1-score on all three languages. We apply two multi-tasklearning approaches to three Dravidian languages: Kannada, Malayalam, andTamil. Maximum scores on Kannada and Malayalam were achieved by mBERT subjectedto cross-entropy loss and with an approach of hard parameter sharing. Bestscores on Tamil was achieved by DistilBERT subjected to cross-entropy loss withsoft parameter sharing as the architecture type. For the tasks of sentimentanalysis and offensive language identification, the best-performing modelscored a weighted F1-score of (66.8\% and 90.5\%), (59\% and 70\%), and (62.1\%and 75.3\%) for Kannada, Malayalam, and Tamil on sentiment analysis andoffensive language identification, respectively. The data and approachesdiscussed in this paper are published inGithub\footnote{\href{https://github.com/SiddhanthHegde/Dravidian-MTL-Benchmarking}{Dravidian-MTL-Benchmarking}}.

Benchmarking Multi-Task Learning for Sentiment Analysis and Offensive Language Identification in Under-Resourced Dravidian Languages