A text-writing system for Easy-to-Read German evaluated with low-literate users with cognitive impairment

Ina Steinmetz, Karin HarbuschACL 2022

TL;DR

This paper introduces EasyTalk, an innovative text-writing assistant designed to support low-literate individuals with cognitive impairments in producing clear and coherent German text. By combining interactive sentence generation with features that promote understandability and audience design, EasyTalk aims to bridge the communication gap for users with intellectual and developmental disabilities.

What Problem Does This Paper Address?

Millions of adults, particularly those with intellectual and developmental disabilities (IDD) and complex communication needs (CCN), struggle with low literacy skills. This significantly hinders their participation in social spheres. While "Leichte Sprache" (Easy-to-Read German) offers a simplified communication standard, creating such texts often requires proficient German speakers, potentially leading to a disconnect with the target audience's actual needs and communication styles. Existing writing support systems, while helpful, often lack the specific adaptations needed for this user group, sometimes being too complex or not providing sufficient guidance for text structure and coherence. This paper addresses the need for a writing tool that is not only functional but also accessible, intuitive, and empowering for users with low literacy and cognitive impairments.

Methods

EasyTalk employs a multi-layered interface designed for ease of use. Its core functionality revolves around an interactive sentence generation process.

  • Natural Language Paraphrase Generator: At the sentence level, a generator suggests correctly inflected word forms and ensures correct word ordering, based on an extension of Leichte Sprache (ELS). This system is built on a lexicalized, unification-based Performance Grammar, originally developed for L2 German learners.
  • Bottom-Up Approach: Users build sentences by selecting constituents, with the system providing prompts and suggestions. This interactive process includes revisions, akin to scaffolded writing.
  • Audience Design Support: Beyond sentence construction, EasyTalk prompts users to add elements that enhance understandability and text coherence, such as temporal or locational attributes.
  • Connector Panel: To improve text flow, a dedicated panel offers suggestions for discourse connectors, drawing from Rhetorical Structure Theory (RST) to convey conceptual messages more effectively than simple, choppy sentences.
  • User-Centric Interface: The interface is structured into three layers: a Text Panel for viewing the entire text, a Sentence and Connector Panel for building sentences and linking them, and a Next-Word Panel for word suggestions. Read-aloud functionality is also integrated.

Datasets

The paper does not explicitly mention the use of large datasets for training the core NLP models. However, it references:

  • A corpus of over 29,000 sentences from various Leichte Sprache texts from the internet, used to identify issues with text coherence and sentence structure.
  • CELEX lexical database as the basis for the COMPASS system's lexicon, which EasyTalk utilizes.
  • ARASAAC symbol set as the default symbol set for EasyTalk.

No new benchmarks were introduced in this paper.

Key Insights and Ideas

  • Bridging the Gap with ELS: Extending Leichte Sprache (LS) to Extended Leichte Sprache (ELS) allows for more natural German constructions that are still accessible, reflecting how the target group might articulate their thoughts.
  • Interactive Sentence Building is Key: A bottom-up, interactive approach where users select constituents and receive real-time feedback is more effective than presenting an empty page.
  • Beyond Sentence Level: True writing support requires addressing text coherence and audience design, not just individual sentence correctness. EasyTalk's connector panel is a crucial step in this direction.
  • Intuitive Design for Cognitive Impairments: The success of EasyTalk hinges on its ability to present complex linguistic tasks through simple, intuitive dialogues and visual cues, minimizing cognitive load.
  • User Feedback is Paramount: Iterative evaluation with the target user group is essential for identifying usability flaws and ensuring system acceptance, especially for individuals with IDD and CCN.

Key Results

The study involved a qualitative case study with eight German-speaking participants aged 18-25, who had various conditions, writing, and computer skills.

  • High Usability: All participants successfully typed at least three sentences, with an average of four words per sentence.
  • Intuitive Interface: The interface was found to be easy to learn and use, with participants quickly grasping sentence construction and the use of prompts. Four participants spontaneously skipped the example rehearsal and typed their own sentences.
  • Positive User Feedback: Participants generally responded positively, with comments like "The headers help with concentration" and "The connectors between sentences are important."
  • Eye-Tracking Data: Eye-tracking confirmed that users focused on the intended dialogue elements and did not exhibit signs of searching for items, indicating efficient interaction.
  • Limitations in Connector Use: While the connector panel was present, users did not systematically use connectors. The study suggests this is due to the layout and the need for active training.

Limitations and Future Work

  • Connector Panel Improvement: The current layout of the Connector Panel was found to be less effective, with users often not exploring the second row of options. Future work includes a redesign and the introduction of an active training mode to teach connector usage.
  • Limited Sample Size: The study used a small sample of eight participants for a qualitative case study, which limits the generalizability of the findings.
  • Future Work: The authors plan to incorporate automatic storytelling concepts into the active training mode and evaluate this in longitudinal studies. They also aim to develop a native smartphone version of EasyTalk and personalize features for specific user groups.

Summary updated

BibTeX
@inproceedings{steinmetz2022text,
  author = {Ina Steinmetz and Karin Harbusch},
  title = {{A text-writing system for Easy-to-Read German evaluated with low-literate users with cognitive impairment}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2022},
}
Figure