📄 Making Jobs Accessible through AI-supported Easy Language Translation

Summary

People with cognitive impairments struggle to enter the primary labor market, partly because workplace documents are rarely available in Easy Language (EL). Producing such texts by hand is slow, costly, and depends on scarce expert translators. This project paper introduces STARK-LS, a German initiative running from 2025 to 2029 and funded by the Federal Ministry of Labour and Social Affairs. It tests whether AI can help close this gap. SUMM AI's tool machine-translates workplace materials into EL, Mainz University's Tra&Co Center then revises the output, and the finished texts are used in real internships organized by ILIN at Karlsruhe University of Applied Sciences. The planned evaluation combines several methods: lab-based eye-tracking and questionnaires, interviews with interns and EL experts, a company survey, error annotation following DIN SPEC 33429, and keystroke logging to measure post-editing effort. No empirical results are reported yet because the internships are still under way. The work matters because related efforts leave gaps: iDEM does not cover German, and TOP.KI targets Plain Language instead of EL. The project also aims to produce practical guidance for employers and rehabilitation agencies.

Key points

• The project has four phases: organizing internships, AI translation plus post-editing of workplace materials, accompanying research, and the creation of guidelines, models, and market analyses.
• The partners are Johannes Gutenberg University Mainz (Tra&Co), Karlsruhe University of Applied Sciences (ILIN), and SUMM AI, together with the District Administration of Germersheim, employers, and sheltered workshops.
• Quality is checked in two ways. Error annotation based on DIN SPEC 33429 compares raw AI output with post-edited texts, and keystroke logging compares the effort of post-editing AI-generated EL with human translation.
• User studies with participants with cognitive impairments use eye-tracking and augmented reality, along with comprehension tasks, interviews, and questionnaires.
• The project builds on the earlier LeiSA findings, which showed that EL texts frequently ignore guidelines and are seldom used by companies. Results are still pending.