Research Program
Investigating the neurocognitive architecture of human speech, language, and communication through multimodal Artificial Intelligence, neuroimaging, and computational linguistics—translating foundational brain science into scalable clinical biomarkers.
Frontier Research Vision: Decoding the Predictive Brain via Language Biomarkers
The Scientific Challenge: Neurodegenerative and neurodevelopmental brain conditions alter the neural networks supporting speech and language years before overt clinical symptoms emerge. Yet gold-standard biological markers (CSF amyloid/tau, PET neuroimaging) remain invasive, costly, and inaccessible for population-scale screening, while traditional behavioral assessments are subjective, time-consuming, and language-specific.
Our High-Gain Approach: Human language production requires millisecond-level coordination across distributed cortical and subcortical networks. Our research program pioneers explainable, neurobiologically grounded, and cross-linguistically generalizable AI models that extract acoustic, prosodic, morphosyntactic, semantic, and information-theoretic footprints from natural connected speech and writing. By coupling computational linguistics and foundation models with structural/functional neuroimaging (fMRI, pCASL cerebral perfusion, DTI) and non-invasive brain stimulation (tDCS), we build transdiagnostic digital biomarkers that generalize across languages, dialects, and clinical populations.
Four Core Research Pillars
Our research program integrates four synergistic areas spanning computational neurolinguistics, clinical AI, neuroimaging, and experimental phonetics.
Automated Diagnosis, Prognosis & Therapy Efficacy via AI
Differential diagnosis and variant subtyping of Primary Progressive Aphasia (nfvPPA, svPPA, lvPPA), amnestic vs. non-amnestic Mild Cognitive Impairment (MCI), Alzheimer’s disease, and pediatric neurodevelopmental conditions (Autism Spectrum Disorder, DLD) require multidisciplinary expertise. We design deep neural networks, sequential architectures, and large language model embeddings that automate screening and variant classification from short spoken or written tasks.
Beyond diagnostics, we evaluate treatment efficacy in clinical trials combining behavioral language therapy with transcranial direct current stimulation (tDCS), quantifying subtle acoustic and phonological improvements in patients with apraxia of speech and aphasia.
Key Representative Publications
- Language biomarker screening using AI: a transdiagnostic approach to the brain (Scientific Reports, 2026)
- Machine Learning Classification of Patients with Amnestic and Non-Amnestic MCI from Written Picture Description Tasks (Brain Sciences, 2024)
- Autism Detection in Children: Integrating Machine Learning and NLP in Narrative Analysis (Behavioral Sciences, 2024)
- Automatic subtyping of individuals with Primary Progressive Aphasia (Journal of Alzheimer’s Disease, 2021)
- Effects of tDCS on Sound Duration in Patients with Apraxia of Speech in PPA (Brain Sciences, 2021)
- Identification of Mild Cognitive Impairment from Speech in Swedish using Deep Sequential Neural Networks (Frontiers in Neurology, 2018)
Speech, Language & Neuroimaging Biomarkers
To ensure clinical AI models are biologically valid rather than opaque "black boxes," we map how structural atrophy, cerebral hypoperfusion, and disrupted functional connectivity alter human speech and language. Using Pseudo-Continuous Arterial Spin Labeling (pCASL) MRI, structural MRI, and Activation Likelihood Estimation (ALE) meta-analyses, we investigate how working memory, semantic, and prosodic brain networks support verbal learning and discourse.
In parallel, we identify novel behavioral biomarkers—including voice quality (cepstral peak prominence, harmonics-to-noise ratio), speech fluency, elaborative morphosyntax (function-to-content ratios, adjectives/adverbs), and information-theoretic surprisal/entropy metrics.
Key Representative Publications
- Normative clinical language data and task specific effects (Brain and Language, 2026)
- Linguistic and Emotional Prosody: A Systematic Review and ALE Meta-Analysis (Neuroscience & Biobehavioral Reviews, 2025)
- Written picture descriptions distinguish variants of primary progressive aphasia (Journal of Alzheimer’s Disease, 2025)
- The Contribution of Working Memory Areas to Verbal Learning and Recall in PPA (Frontiers in Neurology, 2022)
- Voice quality and speech fluency distinguish individuals with MCI from Healthy Controls (PLOS ONE, 2020)
- Part of Speech Production in Patients With PPA: An Analysis Based on NLP (AJSLP, 2020)
Open Brain AI & Automated Clinical Tools
Manual transcription and clinical scoring of spoken discourse, spelling, and phonology are prohibitively labor-intensive and prone to inter-rater variability. We architect open-access computational platforms and validated distance-metric algorithms that automate clinical language assessment in seconds while matching gold-standard human raters ($r_s = .99$).
Our flagship platform, Open Brain AI, integrates multilingual speech-to-text, IPA phonemic conversion, morphosyntactic tagging, semantic/lexical profiling, and specialized clinical apps for spelling and phonological error analysis across 15+ languages.
Key Representative Publications
- A whole-word computational algorithm for phonological error scoring: Differentiating children with DLD, hearing impairment, and typical development (Clinical Linguistics & Phonetics, 2026)
- Integrating AI and NLP to investigate lexical errors in autistic preschoolers (Research in Autism, 2026)
- Open Brain AI and language assessment (Frontiers in Human Neuroscience, 2024)
- Best practice guidelines for reporting spoken discourse in aphasia and neurogenic communication disorders (Aphasiology, 2023)
- Automatic metrics of spelling errors on words and non-words (JSLHR, 2020)
Speech Acoustics, Prosody & Sociolinguistic Variation
Speech signals simultaneously encode linguistic structure (phonemic category, stress, information structure, tonal alignment) and indexical/sociolinguistic identity (dialect, sociolect, L1/L2 background). Our experimental phonetics and sociophonetics research demonstrates that even brief sub-phonemic segments—such as stop consonant bursts, fricative spectra, and sonorant coarticulation—carry systematic dialectal signatures.
Understanding and modeling this acoustic gradience is vital for clinical speech-language pathology in bilectal, diglossic, and multilingual societies (such as Norway, Sweden, Greece, and Cyprus), preventing the misdiagnosis of dialectal or bilingual variation as speech pathology.
Key Representative Publications
- Sonorant spectra and coarticulation distinguish speakers with different dialects (Speech Communication, 2022)
- Vowel learning in diglossic settings: Evidence from Arabic-Greek learners (International Journal of Bilingualism, 2020)
- Dialect classification from a single sonorant sound using Deep Neural Networks (Frontiers in Communication, 2019)
- The Nature of Phonetic Gradience across a Dialect Continuum: Evidence from Modern Greek Vowels (Phonetica, 2017)
- Dialect classification using vowel acoustic parameters (Speech Communication, 2017)
- The bursts of stops can convey dialectal information (JASA Express Letters, 2016)
Funded Research Grants & Fellowships
Track record of competitive research funding across European, Nordic, and US federal funding agencies.
European & International Research Collaborations
We actively collaborate with clinical neurology departments, memory clinics, computational linguistics labs, and special needs education centers across Europe and North America, and welcome inquiries regarding ERC, Horizon Europe, FRIPRO, MSCA Postdoctoral Fellowships, and joint clinical cohorts.