2nd Workshop on Tools and Resources for People with REAding DIfficulties (READI)
- submission deadline: April 10th, 2022 April 15th, 2022 12:00 (noon)
- notification of acceptance: May 2nd, 2022
- deadline for camera-ready versions: May 23rd, 2022
- workshop: June 24th, 2022
Carolina Scarton (University of Sheffield, UK)
Personalised Text Simplification: When, How and For Whom to Simplify
Text Adaptation (TA) tasks, such as Automatic Text Simplification and Summarisation, are important in the Natural Language Processing area, with multiple applications, including improving other NLP-related tasks (e.g. Machine Translation) or enabling access to information by specific groups (e.g. children, elderly, non-native speakers). In recent years, these tasks have become increasingly popular in the research community, motivated by the availability of data and the advent of deep learning models. Researchers proposed text-to-text transformation models, following encoder-decoder architectures, similar to research done in Machine Translation (e.g. the models are trained with parallel pairs of original-simplified text). Together with new models, datasets for benchmarking and evaluation metrics/methods tailored for text adaptation tasks were also proposed.
However, a very important aspect of TA is the target audience. For instance, it is expected that a simplified text targeting a non-native speaker of English will not be suitable for children. Therefore, since most research done to date mainly focuses on "one size fits all" approaches, important questions still remain unanswered: When should specific adaptations be performed in a text?; Which adaptation operations should be performed? and How to create adapted texts that are really useful for a given target audience?
This talk will discuss the task of Text Simplification (TS), presenting work that focuses on outputs tailored for specific audiences. The main challenges of personalised TS will be presented, highlighting the importance of well-defined datasets as well as reliable evaluation metrics or methods that can capture the specificities of each audience.
Arne Jönsson (Linköping University, Sweden)
Tools and Services for Text Adaptation
There are a variety of tools and resources for text adaptation, to make texts easier to read, including lexical and syntactic simplification and automatic text summarization. Related to these are a number of ways to measure, and visualise, a text's complexity, from simple measures trying to cover readability in one number to more refined measures trying to capture a text's complexity from various perspectives.
The users of such tools and resources can broadly be divided in two groups: text producers and text consumers. Text consumers are a heterogenous group with different needs for text adaptation depending on their reading difficulties. Text producers such as journalists, communicators, authors, and teachers are also a heterogenous group with different needs that need to be accounted for.
Bridging the gap between available tools and resources and the needs of the different user groups' is a less studied area. Given that few users can articulate their needs for text adaptation makes this even more challenging.
In this talk we will present the challenges encountered when trying to allow users to combine various text adaptation techniques and visualize a text’s complexity. Design of useful services for text adaptation and issues related to the needs of various user groups will also be presented.