Measuring and Modelling Reader Comprehension of AI-Assisted Insurance Documents

DIALOGIC rests on a testable claim: documents written and delivered with its tools are easier to understand, especially for the readers who struggle most with complex contracts. This role tests that claim with real users, and helps define how what readers understand is captured in DPDF (Dynamic Pragmatic Document Format), the platform's novel document format.

Most of your time will go to the pilots with insurers and brokers, which run during the internship period. You will contribute to the setup of the pilots under regulatory sandbox conditions (Task 4.2), the analysis of comprehension outcomes and user feedback (Task 4.4), and the platform refinements that follow (Task 4.5). Your activities will include:

• Helping design the pilot protocols: comprehension measures, baselines, the sampling of user groups including elderly, low-literacy and migrant readers, and consent and data-handling procedures.
• Preparing study instruments such as comprehension tests, surveys and interview guides, and contributing to the pilot setup documentation (Deliverable 4.2).
• Analysing quantitative outcomes, such as comprehension against baseline, query volumes and time to resolution, together with qualitative feedback from interviews and focus groups, and contributing to the comprehension and feedback analysis (Deliverable 4.4).
• Translating the findings into concrete requirements for the editor, DPDF, JedAI and TRIC scoring during the iteration phase.

In parallel, you will contribute to the definition of the DPDF schema (Task 5.1). DPDF has two layers: the Doc-Genotype, which encodes a document's linguistic structure and its taxonomy of concepts, and the Doc-Phenotype, which records how real readers interact with and understand it. Defining this schema is ontology work: deciding which concepts, relations and reader states the format represents, and how they map to standards such as JSON-LD and EU interoperability frameworks. Your activities will include:

• Defining how the comprehension and interaction measures from the pilots are represented in the Doc-Phenotype, so that what you measure becomes part of the format.
• Contributing to the ontology of document concepts behind the Doc-Genotype and TRIC informativeness scoring, and checking it against what readers actually understand.
• Contributing to the alignment of the DPDF schema with JSON-LD and EU interoperability standards (Deliverable 5.4).

Measuring comprehension is a challenging task, and representing it formally raises its own research questions. We need someone who brings rigorous experimental and mixed-methods practice from psychology, psycholinguistics or human-computer interaction, can apply it in a real regulated setting rather than a lab, and wants to help turn those measurements into an ontology that a document format can use.

• Experience designing and running comprehension studies with real users outside the laboratory, in a regulated sector.
• A direct contribution to the project's pilot setup documentation, its comprehension and feedback analysis, and its DPDF schema alignment report.
• Practical experience applying research ethics and GDPR in applied studies, including with vulnerable groups.
• Hands-on experience turning comprehension and behavioural measures into a formal data model used by a product.
• An inside view of how evidence from users shapes an AI product and its regulatory positioning.

These activities feed directly into the project's expected outcomes: robust pilot evidence on comprehension, support needs and user satisfaction, a platform refined on that evidence, and a DPDF format that records how real readers understand a document.

Deep-Tech Talents (research-focused internship).

• A PhD, completed or in progress, in experimental psychology, linguistics, psycholinguistics, cognitive science, human-computer interaction or a related field.
• Experience running studies with human participants, including comprehension or reading measures.
• Solid quantitative skills, including experimental design and statistics in R or Python, and experience with qualitative methods such as interviews and thematic coding.
• Knowledge of research ethics and informed-consent procedures.
• C1-level proficiency in English.

• Able to communicate clearly with non-specialists.
• Organised and able to work to pilot timelines.
• Comfortable working in a multidisciplinary team.

• Duration: 3 to 6 months, to be agreed with you in line with the rules of your programme.
• Preferred start: January 2027, so that you can take part in the pilot setup.
• Interviews: we will start contacting candidates as soon as the vacancy is published and review applications on a rolling basis.
• Work modality: on-site at our Barcelona office.
• Supervision: our CEO, Alessandro Rufo, acts as tutor, with scientific supervision by our Chief Scientist, Prof. Cristina Becchio.