the project

Language, mediation, and the reading mind

Texts of all kinds — news, reports, speeches, institutional and strategic documents — circulate in a space marked by a plurality of languages, genres, and regimes of mediation. The same content is presented differently depending on the language of the text, its context of production, and the constraints of translation — and these transformations affect the informational structure of the message and the cognitive processes of its receiver: attention, cognitive load, comprehension, recall, and credibility assessment.

COGNILANG is an international research project directed by Prof. Mathieu Guidère, open to partner language teams worldwide: the pilot (EN–FR–AR, as an example) sets the protocol, and each new team extends the comparison to its own languages. Its analytical backbone is laid out in the Cognitive Approach and the Neurolinguistic Approach; its security and defence component is presented on the LLM page.

research questions

One central question, six sub-questions

RQ0 — To what extent do the linguistic, discursive, and translational differences that characterise texts across languages (for example English, French, and Arabic) modify the cognitive processes of comprehension, memorisation, and interpretation — and how can AI help objectify these transformations?

Code	Question
RQ1	How is the same content reformulated across the studied languages — lexical choices, syntax, framing, information hierarchy, modalisation?
RQ2	What cognitive effects do these reformulations produce in readers — comprehension, recall, perceived clarity, credibility?
RQ3	Which mental operations characterise the translation of texts — selection, prioritisation, pragmatic inference, reformulation, intercultural adaptation?
RQ4	Which textual markers are associated with cognitive load, memorisation, and comprehension?
RQ5	How can AI tools identify, at scale, regularities in interlinguistic transformations and in the representational frames of texts?
RQ6	What role does linguistic prediction play in the reception of texts and in translational decisions — and how do the predictability profiles of versions of the same event differ, before and after translation?

hypotheses

Six falsifiable hypotheses, each tied to observables

H	Hypothesis	Observable variables
H1	Texts differ across languages in framing, informational density, and modalisation.	Frame labels; propositional density; modal and evidential marker counts; evaluative-lexis indices.
H2	Translation introduces measurable transformations — omission, explicitation, modulation, reformulation.	Aligned pairs coded by shift type; explicitation ratio; omission rate; semantic similarity scores.
H3	These transformations affect comprehension, recall of key information, and credibility assessment.	Comprehension scores; immediate/delayed recall of idea units; credibility ratings.
H4	Readers do not process the same event identically in an original versus a translated version.	Between-condition differences in reading time, subjective effort, recall accuracy.
H5	AI tools can identify, at scale, regularities in interlinguistic transformation and framing.	Precision/recall of automatic detection against a hand-annotated gold subset.
H6	Predictability profiles differ across languages and are modified by translation; lower predictability raises load and lowers recall.	LM surprisal per token/sentence; cloze probability; correlation of surprisal with reading time, effort, recall.

Discipline rule. Effects of language system, of translational mediation, and of editorial context are coded and argued separately. A difference between two versions of an event is never automatically a translation effect.

theoretical framework

Four fields, one articulation

Discourse linguistics

Framing, information hierarchy, modalisation, evaluation, agentivity, nominalisation — the descriptive categories applied to the texts.

Cognitive science

Attention, working memory, inference, cognitive load, predictive processing and surprisal, levels of comprehension (textbase vs. situation model) — the constructs and measures applied to readers.

Cognitive translation studies

Translation as problem-solving and decision-making under constraint; explicitation, omission, modulation as cognitive traces; anticipation as the interpreter's survival skill.

NLP / AI

Multilingual embeddings, cross-lingual alignment, divergence detection, frame classification, surprisal computation — the instruments that make three-language comparison tractable.

The originality of COGNILANG lies in connecting the textual level (what changes between versions) with the receptive level (what those changes do to readers) — with AI as the bridge that objectifies the textual level, and linguistic prediction as the construct present at all three levels: a property of texts (measurable via surprisal), a competence of mediators (anticipation in translation and interpreting), and a determinant of reception (processing cost in readers).

scope

Six notions every output must engage

Notion	Operational meaning in this project
Texts	Comparable versions of the same content across sources in each studied language — news, reports, speeches, official and strategic documents.
Cognition	Attention, working memory, cognitive load, comprehension, recall, credibility judgement.
Translation	A situated cognitive activity of decision and selection — never mere lexical equivalence.
AI / NLP	Alignment, semantic divergence detection, frame comparison, semi-automatic annotation, surprisal computation.
Linguistic prediction	Anticipatory processing as a foundational mechanism of comprehension and of translational expertise — operationalised through surprisal and cloze measures.
The languages	Any set of two or more languages. The pilot used EN–FR–AR as an example — three writing systems, two directionalities, distinct discursive traditions; each team chooses its own set.