Jing Cai

Vision

Language is the fundamental engine of human communication, allowing us to share complex thoughts, build cohesive societies, and connect with one another. For patients who lose the ability to speak due to severe neurological conditions like aphasia or ALS, restoring a reliable way to communicate is essential to breaking their isolation and returning their voice. While state-of-the-art English speech BCIs starts to allow individuals with severe paralysis to communicate via digital screens and text-to-speech synthesizers at conversational speeds, a comparable system for Mandarin Chinese does not yet exist. Meanwhile, we have yet to fully decipher the precise neural support that drives natural, day-to-day language production and comprehension.

Our research directly addresses these challenges through two main aspects: developing a Chinese speech BCI and deciphering the neural mechanisms of human language. By bridging cognitive neuroscience with clinical engineering, we aim to transform basic research into restorative medical devices.

Aims

We combine these highly precise neural recordings (such as single-neuron recordings, sEEG, and high-density microelectrode arrays) with advanced Large Language Models (LLMs) to engineer a high-throughput Mandarin decoding pipeline. In addition, by directly linking real-time neural dynamics with computational modeling, we reveal the neural mechanisms that support natural human conversation.

1. Developing a Mandarin Speech BCI

Our primary engineering goal is to pioneer high-precision speech neuroprostheses optimized specifically for Mandarin Chinese. Decoding Mandarin presents unique challenges that do not exist in non-tonal languages like English: Mandarin relies on precise tonal variations, features a dense concentration of homophones (words that sound identical but carry vastly different meanings), and depends heavily on contextual cues. To overcome these algorithmic hurdles, our lab leverages highly unique single-neuron recordings, advanced AI decoding models, and LLM-powered error correction. By unifying these technologies, we aim to transition Mandarin speech BCIs out of laboratory tasks and into robust, real-world clinical utility.

To achieve this, our research is focused on the following milestones:

·Engineering Real-Time, Full-Syllable Decoding: We are building a speech-synthesis pipeline capable of capturing the entire prounciation of Mandarin phonemes and tones, translating raw brain activity into fluid audio with minimal delay.

·Accelerating Training Procedure with Low-Sample Calibration: BCI systems traditionally require exhausting, days-long training sessions before they can accurately recognize a participant’s unique neural signals. We are developing rapid calibration algorithms that leverage pre-trained networks to drastically compress this setup time, allowing patients to communicate almost immediately.

·Designing Natural Conversation Decoding: We are shifting away from rigid laboratory tasks where patients read isolated words from a screen. Instead, we are creating platforms for spontaneous, fluid conversation by pairing real-time neural decoding with the contextual, predictive power of Large Language Models.

·Driving Clinical Translation via Wireless Technology: In partnership with local resources, we are deploying advanced wireless electrode arrays for speech decoding, bridging the gap between engineering R&D and clinical care.

2. Uncovering the Neural Mechanisms of Language

Optimizing advanced BCIs relies heavily on understanding the precise neural patterns during human speech. Our lab explores how the brain computes and organizes language in real-world settings. Our multi-scale approach allows us to track how isolated neurons and distributed networks process language across its major dimensions: sounds (phonetics), meanings (semantics), structures (syntax), and interactive dialogue (conversational states). Beyond answering a profound scientific question about human cognition, these insights provide the foundational framework required to build smarter, more intuitive clinical decoders.

Significance

By anchoring our work at the convergence of BCI engineering, artificial intelligence, and systems neuroscience, our laboratory is uniquely positioned to address the core technology of Mandarin speech decoding. Our ultimate mission is clinical: to restore natural, fluent, and real-time communication to patients locked in by severe speech impairments, giving them back their own independence. Simultaneously, by exploring how the human brain computes language—one of our most intricate cognitive faculties—our work drives major breakthroughs in speech neuroprostheses for complex, tonal languages.

2024     Mussallem Transformative Scholars in ALS Research Award

2021     American Fellowship, American Association of University Women

2018     Herbert Callen Memorial Prize

2016     Chase Risk Management Teamwork Award

2014     Hector Tyndale Fellow

2013     Hector Tyndale Fellow

2010     Werner Teutsch Prize

1. J. Cai, Y. Kfir, Mohsen Jamali, Hesen Huang, Young Joon Kim, S. Cash, Z. Williams, Mapping the neuronal building blocks of human language with language models, 2026 (Accepted by Nature)

2. J. Cai*, A. E. Hadjinicolaou, A. Paulk, D. Soper, T. Xia, A. Wang, J. Rolston, R. Richardson, Z. M. Williams* & S. S. Cash*, Natural language processing models reveal real time neural dynamics of human conversation. 2025 Nature Communications, 16(1), p.3376  

3. B. Grannan, M. Jamali, J. Cai, A. Khanna, W. Mu￱oz, W., I. Caprara, A. Paulk, S. Cash, E. Fedorenko, Z. Williams, Semantic encoding during language comprehension at single cell resolution. 2024, Nature, 631 (8021), pp.610-616

4. A. R. Khanna, W. Mu￱oz, Y. J. Kim, Y. Kfir, A. C. Paulk, M. Jamali, J. Cai, M. Mustroph, I. Caprara, R. Hardstone, M. Mejdell, D. Meszena, A. Zuckerman, J. Schweitzer, S. Cash, Z. M. Williams, Single-neuronal prefrontal components of speech production in humans, 2024, Nature, 626, 603-610

5. J. Cai, A. Sweeney, The proof is in the pidan: generalizing proteins as patchy particles, ACS Cent.Sci., 2018, 4 (7), pp. 840-853

6. J. Cai, J. Townsend, T. Dodson, P. Heiney, A. Sweeney, Eye patches: Protein assembly of index-gradient squid lenses, Science, 2017, 357 (6351), pp. 564-569