Enhancing Language Identification and Named Entity Recognition for Code-Mixed Languages

People from multilingual communities often use code-mixed languages to express their view or opinion in the form of text or speech on platforms like Facebook, Twitter, and Instagram. Codemixing or code switching happens when user tend to combine the words of different language in a sentence. In multilingual settings Named Recognition Entity (NER) becomes more complex as it involves identifying entities across languages and often requires tools that support language identification and adaptation to different grammatical structure. The informal and unstructured nature of code-mixed social media text presents challenges for language identification (LID). This work focuses on Named Entity recognition (NER) and wordlevel language identification of Hindi-English code -mixed data. This paper emphasizes on identifying the NER and language of code-mixed multiple BERT-based models. A multilingual BERT classifier distinguishes tokens into EN, HI, and ENHI classes, achieving 98.75% F1-score for EN-HI classification and 72.66% when including the ENHI class. For NER, separate models for English and Hindi achieve high performance with F1-scores of 97.96% and 97.88%, respectively. A multi-head architecture combining LID and NER yields competitive LID results (93.22% F1) but shows reduced NER performance (62.15% F1). The proposed approach serves as a baseline for future research on multilingual and code-mixed NLP tasks.