Professional platform · stateless & multilingual
Ethics charter
PSYNUM
FR Charter Professional access
Home/Methodology

A scientific approach
grounded in linguistic precision.

PSYNUM is built on an artificial-intelligence architecture designed to mirror the linguistic analysis of clinical discourse, never to replace it. The objective is not to automate diagnosis, but to strengthen the consistency, traceability and reliability of professional reasoning.

Archives — institutional library
01 · Corpus construction

Data sources and corpus construction.

The learning models used by PSYNUM have been trained on corpora drawn from clinical practice in mental health, university research and anonymised multilingual linguistic archives.

  • Patient verbatims from clinical interviews in psychiatry, clinical psychology and speech-language therapy.
  • Transcripts of diagnostic assessments (semi-structured interviews, clinical scales).
  • Longitudinal corpora of CBT and exposure-therapy sessions.
  • Case studies documented in the scientific literature in psycholinguistics and neurolinguistics.

Each verbatim has been pre-processed, anonymised and coded according to a grid of lexical, syntactic, semantic, pragmatic and emotional markers validated by specialists for each language covered.

02 · Combined architecture

Model architecture and learning processes.

PSYNUM combines several artificial-intelligence approaches to address the psycholinguistic and neurolinguistic dimensions of clinical discourse.

  • Supervised learning — classification of clinical markers and detection of correlations between linguistic variables.
  • Multilingual language models — fine analysis of discourse in the patient's mother tongue.
  • Vector-based semantic modelling — analysis of narrative coherence, lexical richness and discursive structure.
  • Pragmatic analysis — detection of communicative intentions, avoidance strategies and alteration cues.

Algorithms have been calibrated to favour transparency and interpretability, allowing the clinician to understand the relations between detected linguistic markers and formulated clinical hypotheses.

Computational architecture — circuits and data
Laboratory — validation and control
03 · Scientific validation

Validation and reliability control.

Each tool has been subjected to an internal evaluation phase based on the comparison between the model-generated reports and real, documented clinical cases.

  • Clinical coherence of the reports produced.
  • Stability of linguistic markers across languages and profiles.
  • Conformity with theoretical references in clinical psycho/neurolinguistics.
  • Independent expert reviews & supervision by a multidisciplinary scientific committee.

PSYNUM follows an evolving approach: each model update is subjected to a cross-validation procedure and to per-language linguistic quality control.

04 · Interpretive prudence

Reliability principle and interpretive prudence.

Artificial intelligence never replaces the interview, the observation or the clinical judgement. The reports obtained should be considered as decision-support documents, never as definitive diagnostic conclusions.

Every output of the model is designed to be understandable, justifiable and integrable into an evaluation process based on the practitioner's responsibility.

Reliability rests on the interaction between the machine and the human: AI structures the verbatim, the clinician gives it meaning.
Clinical gesture — medical close-up
A clear epistemological orientation: turning AI into a tool of observation, modelling and understanding of clinical discourse — without ever altering the human dimension of mental-health care. — A technology in the service of clinical thinking
Method & ethics

Technology as an extension of clinical listening.

In the service of rigour, prudence and multilingual accompaniment.