{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,18]],"date-time":"2026-09-18T06:09:06Z","timestamp":1789711746870,"version":"4.0.1"},"reference-count":34,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,18]],"date-time":"2026-09-18T00:00:00Z","timestamp":1789689600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:p>Text normalization (TN) is a crucial preprocessing task in Natural Language Processing (NLP) that enables the conversion of noisy, informal texts into standardized forms suitable for downstream applications, such as machine translation, speech recognition, text-to-speech systems, sentiment analysis, and information retrieval. User-Generated Content (UGC) from social media and messaging platforms often contains irregularities such as abbreviations, phonetic spellings, code-switching, and orthographic inconsistencies. These irregularities have significant implications, especially for low-resourced languages like Hausa, which have limited annotated corpora and computational resources. Existing rule-based and statistical methods struggle in their performance with Hausa text because of spelling variations, code-mixing, abbreviations, and contextual ambiguities. In contrast, neural networks like transformers require a large dataset for efficient training; however, the lack of specialized resources like a parallel corpus further worsens the inefficiency of current models in handling Hausa UGC. To address these gaps, two corpora were developed: a Hausa UGC corpus collected from Twitter and WhatsApp, and a standard Hausa corpus consisting of 27,847 sentences each. The UGC corpus was analyzed to identify Out-of-Vocabulary (OOV) words, their features, and usage patterns. A Dual-BERT Adversarial Network model was then designed and trained to map non-standard Hausa words into their canonical forms accurately. The developed model integrates dual-BERT within an adversarial learning framework, enabling it to capture both the semantic and contextual relationships between informal and standard text. One BERT processes noisy, user-generated Hausa input, while the other learns from standard Hausa sentences, allowing the system to model bidirectional correspondences between the two linguistic forms. Their representations are fused and refined through a generator-discriminator architecture, where the generator produces normalized outputs and the discriminator enforces contextual correctness by distinguishing genuine standard text from generated ones. This adversarial dual-BERT design enhances robustness, contextual awareness, and generalization, making the model highly effective for transforming inconsistent Hausa UGC into standardized text suitable for downstream NLP applications. Experimental results demonstrate that the proposed model achieved an Exact Match Accuracy of 83.99%, BLEU score of 0.7534, WER of 12.02%, and CER of 7.03%, outperforming the evaluated baseline models across all reported metrics. Qualitative analysis further shows that the model effectively normalizes abbreviations, phonetic spellings, character elongations, and many Hausa-English code-mixed expressions, while insertion and deletion errors remain comparatively more challenging. The primary contributions of this work are (i) the construction of a manually annotated parallel corpus for Hausa user-generated text normalization, (ii) the development of a Dual-BERT adversarial framework for context-aware normalization in a low-resource setting, and (iii) a comprehensive evaluation demonstrating the effectiveness of the proposed approach for improving the quality of Hausa user-generated text and supporting downstream NLP applications.<\/jats:p>","DOI":"10.3389\/frai.2026.1905680","type":"journal-article","created":{"date-parts":[[2026,9,18]],"date-time":"2026-09-18T05:39:24Z","timestamp":1789709964000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Dual-BERT adversarial learning for Hausa user-generated text normalization"],"prefix":"10.3389","volume":"9","author":[{"given":"Sulaimon Adebayo","family":"Bashir","sequence":"first","affiliation":[{"name":"Department of Computer Science, School of Information & Communication Technology, Federal University of Technology","place":["Minna, Nigeria"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Abubakar Ahmad","family":"Aliero","sequence":"additional","affiliation":[{"name":"Department of Computer Science, Faculty of Computing, Abdullahi Fodio University of Science and Technology","place":["Aliero, Nigeria"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Amina Gogo","family":"Tafida","sequence":"additional","affiliation":[{"name":"Department of General Studies, Federal University of Technology","place":["Minna, Nigeria"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mukhtar Ibrahim","family":"Hussaini","sequence":"additional","affiliation":[{"name":"Department of Computer Science, Rayhaan University","place":["Birnin Kebbi, Nigeria"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Nur Syadhila Che","family":"Lah","sequence":"additional","affiliation":[{"name":"Department of Computing, Universiti Teknologi PETRONAS, Seri Iskandar","place":["Malaysia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Maged","family":"Nasser","sequence":"additional","affiliation":[{"name":"Department of Computing, Universiti Teknologi PETRONAS, Seri Iskandar","place":["Malaysia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yunusa Adamu","family":"Bena","sequence":"additional","affiliation":[{"name":"Faculty of Engineering, Kebbi State University of Science and Technology","place":["Aliero, Nigeria"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,18]]},"reference":[{"key":"ref1","doi-asserted-by":"publisher","first-page":"101","DOI":"10.4314\/swj.v19i1.13","article-title":"Twitter sentiment analysis for Hausa abbreviations and acronyms","volume":"19","author":"Abdullahi","year":"2024","journal-title":"Sci. World J."},{"key":"ref2","first-page":"1","article-title":"Sentiment analysis of users\u2019 reviews on COVID-19 contact tracing apps with a benchmark dataset","author":"Ahmad","year":"2021"},{"key":"ref3","article-title":"Improving text normalization by optimizing nearest neighbor matching","author":"Ansari","year":"2017","journal-title":"Comput. Lang."},{"key":"ref4","doi-asserted-by":"publisher","first-page":"156","DOI":"10.14569\/IJACSA.2020.0111021","article-title":"Rule-based text normalization for Malay social media texts","volume":"11","author":"Ariffin","year":"2020","journal-title":"Int. J. Adv. Comput. Sci. Appl."},{"key":"ref5","first-page":"420","volume-title":"NAACL HLT 2015\u20132015 Conf. North Am. Chapter Assoc. Comput. Linguist. Hum. Lang. Technol. Proc. Conf","author":"Baldwin","year":"2015"},{"key":"ref6","first-page":"1877","article-title":"Language models are few-shot learners","volume":"2020","author":"Brown","year":"2020","journal-title":"Adv. Neural Inf. Process. Syst"},{"key":"ref7","doi-asserted-by":"publisher","first-page":"2","DOI":"10.1016\/j.sbspro.2011.10.577","article-title":"Text normalization in social media: progress, problems and applications for a pre-processing system of casual English","volume":"27","author":"Clark","year":"2011","journal-title":"Procedia - Soc. Behav. Sci."},{"key":"ref8","first-page":"7122","volume-title":"ICASSP, IEEE Int. Conf. Acoust. Speech Signal Process. - Proc., 2022","author":"Dai","year":"2022"},{"key":"ref9","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1145\/2414425.2414430","article-title":"Lexical normalization for social media text","volume":"4","author":"Han","year":"2013","journal-title":"ACM Trans. Intell. Syst. Technol."},{"key":"ref10","first-page":"67","volume-title":"W-NUT 2021 - 7th Work. Noisy User-Generated Text, Proc. Conf","author":"Higashiyama","year":"2021"},{"key":"ref11","doi-asserted-by":"publisher","first-page":"36202","DOI":"10.1109\/ACCESS.2020.2974674","article-title":"A text normalization method for speech synthesis based on local attention mechanism","volume":"8","author":"Huang","year":"2020","journal-title":"IEEE Access"},{"key":"ref12","doi-asserted-by":"publisher","first-page":"101807","DOI":"10.1016\/j.jksuci.2023.101807","article-title":"Bangla text normalization for text-to-speech synthesizer using machine learning algorithms","volume":"36","author":"Islam","year":"2024","journal-title":"J. King Saud Univ. - Comput. Inf. Sci."},{"key":"ref13","first-page":"1","article-title":"Bridging gaps in natural language processing for Yor\u00f9b\u00e1: a systematic review of a decade of progress and prospects","author":"Jimoh","year":"2025","journal-title":"Comput. Lang."},{"key":"ref14","first-page":"2126","volume-title":"COLING 2020 - 28th Int. Conf. Comput. Linguist. Proc. Conf., no. 2012","author":"Kawamura","year":"2020"},{"key":"ref15","first-page":"335","volume-title":"Proc. 13th Int. Conf. Web Soc. Media, ICWSM 2019","author":"Lourentzou","year":"2019"},{"key":"ref16","first-page":"129","volume-title":"Proc. - 2021 4th Int. Conf. Comput. Informatics Eng. IT-Based Digit. Ind. Innov. Welf. Soc. IC2IE 2021","author":"Maghfur","year":"2021"},{"key":"ref17","volume-title":"2014 5th Int. Conf. Inf. Commun. Technol. Muslim World, ICT4M 2014, no. 2020","author":"Maitama","year":"2014"},{"key":"ref18","first-page":"7284","volume-title":"Proc. Annu. Meet. Assoc. Comput. Linguist","author":"Makarov","year":"2020"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1371\/journal.pone.0299652","article-title":"A normalization model for repeated letters in social media hate speech text based on rules and spelling correction","volume":"19","author":"Mansur","year":"2024","journal-title":"PLoS One"},{"key":"ref20","doi-asserted-by":"publisher","first-page":"p. 102368","DOI":"10.1016\/j.ipm.2020.102368","article-title":"An unsupervised lexical normalization for Roman Hindi and Urdu sentiment analysis","volume":"57","author":"Mehmood","year":"2020","journal-title":"Inf. Process. Manag."},{"key":"ref21","first-page":"590","volume-title":"2022 Lang. Resour. Eval. Conf. Lr. 2022","author":"Muhammad","year":"2022"},{"key":"ref22","author":"Muhammad","year":"2025"},{"key":"ref23","doi-asserted-by":"publisher","first-page":"256","DOI":"10.1016\/j.socnet.2021.08.006","article-title":"Connecting mobile social media with psychosocial well-being: understanding relationship between WeChat involvement, network characteristics, online capital and life satisfaction","volume":"68","author":"Pang","year":"2022","journal-title":"Soc. Networks"},{"key":"ref24","first-page":"1","volume-title":"2018 Int. Jt. Symp. Artif. Intell. Nat. Lang. Process. iSAI-NLP 2018 - Proc., no. Iv","author":"Poolsukkho","year":"2018"},{"key":"ref25","doi-asserted-by":"publisher","first-page":"115","DOI":"10.35940\/ijeat.E1029.0785S319","article-title":"Text normalization and its role in speech synthesis","volume":"8","author":"Rahate","year":"2019","journal-title":"Int. J. Eng. Adv. Technol."},{"key":"ref26","article-title":"Multilevel text normalization with sequence-to-sequence networks and multisource learning","author":"Ruzsics","year":"2019","journal-title":"Comput. Lang."},{"key":"ref27","first-page":"14","article-title":"Design of a spell corrector for Hausa language","volume":"5","author":"Salifou","year":"2014","journal-title":"Int. J. Comput. Linguist."},{"key":"ref28","doi-asserted-by":"publisher","first-page":"7","DOI":"10.1007\/978-981-16-9488-2_39","article-title":"Sentiment analysis of Hausa language tweet using machine learning approach","volume":"8","author":"Sani","year":"2022","journal-title":"J. Res. Appl. Math.."},{"key":"ref29","article-title":"RNN approaches to text normalization: a challenge","author":"Sproat","year":"2016","journal-title":"Comput. Lang."},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1016\/j.eswa.2020.114500","article-title":"Is neural always better? SMT versus NMT for Dutch text normalization","volume":"170","author":"Veliz","year":"2021","journal-title":"Expert Syst. Appl."},{"key":"ref31","article-title":"Automatic correction of writing anomalies in Hausa texts","author":"Wali","year":"2025","journal-title":"Autom. Correct. Writ. Anomalies Hausa Texts. arXiv Prepr. arXiv2506.03820."},{"key":"ref32","first-page":"1","volume-title":"AftricaNLP Work. ICLR2023","author":"Zandam","year":"2023"},{"key":"ref33","doi-asserted-by":"publisher","first-page":"293","DOI":"10.1162\/COLIa00349","article-title":"Neural models of text normalization for speech applications","volume":"45","author":"Zhang","year":"2019","journal-title":"Comput. Linguist."},{"key":"ref34","article-title":"Text normalization for low-resource languages of Africa","author":"Zupon","year":"2021","journal-title":"Comput. Lang."}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1905680\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,18]],"date-time":"2026-09-18T05:39:28Z","timestamp":1789709968000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1905680\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,18]]},"references-count":34,"alternative-id":["10.3389\/frai.2026.1905680"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1905680","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,18]]},"article-number":"1905680"}}