Natural Language Processing (NLP) encompasses a variety of tasks, each with distinct methodologies and applications, including Named Entity Recognition (NER), sentiment analysis, classification, machine translation, summarisation, and information extraction. These tasks underpin numerous real-world applications, from virtual assistants to automated content analysis.
Named Entity Recognition involves identifying and classifying named entities—such as people, organisations, locations, and dates—within unstructured text. NER is foundational to many NLP applications, as it provides structured information from raw text. Early work by Nadeau and Sekine (2007) outlined rule-based, statistical, and hybrid approaches to NER, with machine learning models like Conditional Random Fields (CRFs) gaining prominence due to their ability to model contextual dependencies. Recent advancements leverage deep learning, particularly transformer-based models like BERT (Devlin et al. 2019), which have been superseded by more recent architectures, including DeBERTa-v3 and GLiNER models. Current state-of-the-art approaches employ generalist NER models like GLiNER (Zaratiana et al. 2023) that can identify any entity type using bidirectional transformer encoders, and improved transformer variants like DeBERTa-v3 (He et al. 2021), which outperform BERT through disentangled attention mechanisms.
Sentiment analysis, or opinion mining, determines the emotional tone expressed in a text, typically classifying it as positive, negative, or neutral. Pang and Lee (2008) provided a seminal review of sentiment analysis, highlighting lexicon-based and machine learning approaches. Lexicon-based methods rely on predefined sentiment dictionaries, while machine learning models, such as Support Vector Machines (SVMs) or neural networks, learn from annotated corpora. Contemporary research has established transformer-based models as the leading approach for sentiment analysis. Comprehensive surveys demonstrate that models such as BERT, RoBERTa, and DistilBERT consistently outperform traditional approaches, including CNNs and RNNs through their self-attention mechanisms and bidirectional context understanding (El Azzouzy et al. 2025). Recent comparative analyses show RoBERTa achieving superior performance across diverse domains due to its robust optimisation through larger pre-training datasets and dynamic masking techniques (Zekaoui et al. 2023). Hybrid ensemble approaches combining multiple transformer models have emerged as particularly effective, leveraging complementary strengths to improve robustness and accuracy (Albladi et al. 2025).
Text classification assigns predefined categories to text, encompassing tasks like spam detection, topic classification, and intent recognition. Early approaches, as discussed by Sebastiani (2002), relied on feature engineering and algorithms like Naïve Bayes or SVMs. The advent of deep learning has shifted focus to neural architectures, with transformer-based models achieving superior performance by learning hierarchical feature representations (Vaswani et al. 2017). Current best practices favour DeBERTa and RoBERTa variants over BERT for classification tasks, with DeBERTa-v3 showing consistent improvements of 0.9-3.6 percentage points over RoBERTa-large (Timoneda & Vallejo Vera 2025). While large language models like GPT-4 offer reasonable zero-shot performance, fine-tuned encoder models maintain superior accuracy, particularly when training data exceeds 1,000 samples (Wang et al. 2024).
Machine translation (MT) automates the translation of text between languages. Early systems evolved from rule-based to statistical approaches, which used bilingual corpora to model translation probabilities. The introduction of neural MT, particularly sequence-to-sequence models with attention mechanisms (Bahdanau et al. 2014), improved fluency and accuracy. Transformer-based models continue to dominate MT (Vaswani et al. 2017), with recent advances including multilingual models like mT5 and character-level approaches such as ByT5 showing particular effectiveness for low-resource languages and scenarios with limited fine-tuning data (Edman et al. 2024). Current state-of-the-art systems employ large language models including GPT-4, Claude-3.5, and specialised MT models, with WMT24 evaluations showing significant quality improvements through model scaling and improved training methodologies (Kocmi et al. 2024).
NLP tasks like NER, sentiment analysis, classification, machine translation, summarisation, and information extraction form the backbone of modern language technologies. Their applications span industries, from healthcare to finance, transforming how we process and utilise text. While challenges persist, ongoing advancements in deep learning and ethical AI promise to expand their capabilities. By bridging human language and machine understanding, these tasks continue to shape a data-driven world.
References:
1. Albladi, Aish & Uddin, Md Kaosar & Islam, Minarul & Seals, Cheryl. 2025. TWSSenti: A Novel Hybrid Framework for Topic-Wise Sentiment Analysis on Social Media Using Transformer Models. arXiv preprint arXiv:2504.09896. https://arxiv.org/abs/2504.09896. ^ Back
2. Bahdanau, D., Cho, K., & Bengio, Y. 2014. ‘Neural Machine Translation by Jointly Learning to Align and Translate’. arXiv preprint arXiv:1409.0473. ^ Back
3. Devlin, Jacob, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. ‘BERT: Pre-Training of Deep Bidirectional Transformers for Language Understanding’. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, 4171–86. ^ Back
4. Edman, Lukas & Sarti, Gabriele & Toral, Antonio & van Noord, Gertjan & Bisazza, Arianna. 2024. Are character-level translations worth the wait? Comparing ByT5 and mT5 for machine translation. Transactions of the Association for Computational Linguistics, 12: 392–410. ^ Back
5. El Azzouzy, Oussama & Chanyour, Tarik & Jai Andaloussi, Said. 2025. Transformer-based models for sentiment analysis of YouTube video comments. Scientific African, 29, e02836. https://doi.org/10.1016/j.sciaf.2025.e02836. ^ Back
6. Kocmi, Tom & Avramidis, Eleftherios & Bawden, Rachel & Bojar, Ondřej & Dvorkovich, Anton & Federmann, Christian & Fishel, Mark et al. 2024. Findings of the WMT24 general machine translation shared task: The LLM era is here but MT is not solved yet. In Proceedings of the Ninth Conference on Machine Translation, 1–46. ^ Back
7. Nadeau, David, and Satoshi Sekine. 2007. ‘A Survey of Named Entity Recognition and Classification’. Lingvisticae Investigationes 30(1): 3–26. ^ Back
8. Pang, Bo, and Lillian Lee. 2008. ‘Opinion Mining and Sentiment Analysis’. Foundations and Trends® in Information Retrieval 2(1–2): 1–135. ^ Back
9. Sebastiani, Fabrizio. 2002. ‘Machine Learning in Automated Text Categorization’. ACM Computing Surveys (CSUR) 34(1): 1–47. ^ Back
10. Timoneda, Joan C., and Sebastián Vallejo Vera. 2025. BERT, RoBERTa, or DeBERTa? Comparing Performance Across Transformers Models in Political Science Text. The Journal of Politics, 87(1): 347–364. ^ Back
11. Tumasjan, Andranik, Timm Sprenger, Philipp Sandner, and Isabell Welpe. 2010. ‘Predicting Elections with Twitter: What 140 Characters Reveal about Political Sentiment’. Proceedings of the International AAAI Conference on Web and Social Media 4(1): 178–185. ^ Back
12. Vaswani, Ashish, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. 2017. ‘Attention Is All You Need’. arXiv. doi:10.48550/ARXIV.1706.03762 – ^ Back
13. Wang, Yu & Qu, Wen & Ye, Xin. 2024. Selecting between BERT and GPT for text classification in political science research. arXiv preprint arXiv:2411.05050. ^ Back
14. Zaratiana, Urchade – Tomeh, Nadi – Holat, Pierre – Charnois, Thierry. 2023. Gliner: Generalist model for named entity recognition using bidirectional transformer. arXiv preprint arXiv:2311.08526. ^ Back
15. Zhang, Lei, Shuai Wang, and Bing Liu. 2018. ‘Deep Learning for Sentiment Analysis: A Survey’. Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery 8(4): e1253. ^ Back
