Developing an AI-based Tool for Real-Time Dialect Identification in Multilingual Settings

Thesis Abstract

Effective communication in multilingual environments often faces significant challenges due to the subtle variations in dialects that influence comprehension, social interaction, and information dissemination. Traditional dialect identification methods are limited by their reliance on manual, time-consuming processes and expert knowledge, making them inadequate in contexts requiring rapid, scalable solutions. This study aims to develop an AI-based tool capable of real-time dialect identification across multiple languages, thereby enhancing the effectiveness of communication platforms in diverse linguistic settings. The specific objectives include (1) designing a robust speech corpus encompassing major dialectal varieties within selected multilingual regions, (2) developing machine learning models trained for dialect classification, (3) evaluating the tool's accuracy and efficiency in real-world scenarios, and (4) exploring users' acceptance and usability of the system. The research adopts a mixed-methods design, integrating quantitative and qualitative approaches to optimize both the technical performance and user experience aspects. The population of the study comprises speech samples from 1,500 native speakers representing five dialectal groups within a multilingual country. A stratified random sampling technique is employed to select participants from local linguistic communities. Data collection involves recording natural speech using high-fidelity microphones in controlled environments and supplementing these with ethnographic interviews to capture contextual nuances influencing dialectal variation. The primary data analysis involves training supervised machine learning algorithms, such as convolutional neural networks (CNN) and support vector machines (SVM), utilizing Mel-Frequency Cepstral Coefficients (MFCC) and spectrogram features extracted from speech recordings. Model performance will be evaluated through metrics including accuracy, precision, recall, and F1-score, with cross-validation procedures ensuring the models' robustness. Furthermore, techniques such as confusion matrices will help identify misclassification patterns, guiding iterative model refinement. A thematic analysis based on user feedback collected via semi-structured interviews will provide insights into system usability and acceptance, analyzed using NVivo software. Expected findings indicate that the AI-driven tool will achieve an accuracy rate exceeding 85% in identifying dialects across the tested languages, with particular proficiency in dialects with distinctive phonetic markers. The model’s performance is anticipated to outperform existing dialect identification approaches in terms of speed and scalability. The usability analysis is expected to reveal critical factors affecting system acceptance, such as ease of interaction, perceived reliability, and cultural sensitivity. These findings will demonstrate the potential of integrating advanced AI techniques within speech recognition and natural language processing domains to address dialectal diversity challenges effectively. This research contributes significantly to theoretical and applied linguistic knowledge by extending the application of deep learning models in dialect identification, grounded within socio-psycholinguistic frameworks like the Dynamic Systems Theory, which emphasizes language variation as an evolving, context-dependent phenomenon. Additionally, it offers practical implications for governments, educational institutions, and communication service providers seeking to improve multilingual interaction through automated, real-time dialect recognition systems. In conclusion, the study will provide a validated prototype of an AI-based dialect identification tool with potential for deployment in real-world multilingual settings. Recommendations will include strategies for system enhancement, integration with existing speech recognition technologies, and considerations for ethical and privacy concerns in deploying such tools. Future research directions may involve expanding the system to include more languages, dialects, and linguistic features, as well as exploring its adaptability for voice-based authentication and semantic analysis applications. This research underscores the transformative potential of artificial intelligence to bridge linguistic divides, fostering more inclusive and effective communication in diverse multilingual communities.