Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neurotechnological Innovations for Multilingual Communication: Challenges and Future Directions for Low-Resourced Languages
Poster Session A - Sandbox Series, Thursday, October 24, 10:00 - 11:30 am, Great Hall 3 and 4
This poster is part of the Sandbox Series.
Abigail Oppong1; 1Independent Researcher
Recent breakthroughs in Brain-Computer Interface (BCI) technology have enabled individuals with various neurological conditions to communicate using neural signals decoded by implanted devices. These advancements are critical for improving the quality of life for individuals with severe motor impairments, speech disorders, and other conditions affecting communication abilities. Currently, these breakthroughs have worked in two languages: English and Spanish. This project aims to expand BCI communication capabilities to include low-resourced languages, addressing a significant gap in neurotechnological research. Ensuring multilingual inclusivity is vital for developing equitable neurotechnological solutions, as it allows for the evaluation of cross-linguistic generality and the exploration of neural mechanisms specific to diverse languages (Malik-Moraleda, Ayyash et al., 2022). This work seeks to leverage the successful implementation of brain implants in individuals with neurological conditions to enable multilingual communication through neural signal decoding as a case study to advocate for language inclusivity in BCI communication capabilities. The implant, consisting of electrodes placed on the sensorimotor cortex, captured brain activity as participants thought of specific words and phrases. This captured neural data was processed using artificial intelligence (AI) algorithms to translate thoughts into coherent speech across multiple languages. To ensure the robustness and inclusivity of the BCI system, this study highlights the need to incorporate several innovative methodologies, including biological and environmental diversity, by including languages of populations from diverse demographic backgrounds, accounting for age, gender, and environmental influences on brain activity. Using the state of the art techniques, neural signals processed using state-of-the-art AI techniques, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), and transformers, will be finetuned to handle the complexity and variability in neural signals associated with different languages. In enhancing multilingual Models, the AI algorithms will be fine-tuned using transfer learning from pre-trained models on high-resource languages and augmented with data from low-resourced languages. This ensures accurate decoding of neural signals into multiple languages, including those with limited linguistic resources. Comprehensive ethical protocols will be in place to address privacy and data security concerns, ensuring informed consent and mitigating biases in AI models. This study also seeks to expand on the need to collect neural data from speakers of various low-resourced languages and Data Augmentation to enhance data diversity through synthetic data generation and related language data. Leveraging pre-trained models on high-resource languages, fine-tuning with low-resourced language data, and Incorporating robust language models (e.g., GPT, BERT) for understanding linguistic structures.
Topic Areas: Multilingualism, Computational Approaches