Title: TextAD-LiU-GPU
DNr: LiU-gpu-2023-1
Project Type: LiU Compute
Principal Investigator: Daniel Holmer <daniel.holmer@liu.se>
Affiliation: Linköpings universitet
Duration: 2023-03-03 – 2024-04-01
Classification: 10208
Homepage: https://liu.se/forskning/textad


Everybody needs easy to read text, but some need it more than others (Lundberg & Reichenberg, 2008). Reading comprehension is often described in relation to two types of processes, decoding and language comprehension. Poor reading abilities are found in various populations, including individuals with intellectual disability (ID) and dyslexia. These groups have distinct cognitive and language profiles associated with the reading process. The different cognitive and language deficits selectively impair different aspects of reading (Elwér, Keenan, Olson, Byrne, & Samuelsson, 2013), such that reading comprehension is compromised but for many different reasons. Therefore, what is perceived as complicated text likely differs depending on the background of the readers. Textual features such as word and sentence length, vocabulary frequency, syntactic complexity and idea density affect reading comprehension in different ways depending on the prerequisites of the readers. Many teachers struggle in teaching heterogeneous groups and trying to find reading materials that fit every student is complicated and very time demanding. With readers who show specific difficulties such as individuals with ID and dyslexia, this is especially challenging, as these readers are commonly far from their grade level in reading achievement.  Perceived complexity of a text can be described on a general level, for instance by using the Swedish text complexity measurement LIX (Björnsson, 1968). Traditional readability measures, such as LIX, are, however, far too simplified to correctly assess a text’s complexity. Modern language technology techniques and machine learning allow for development of much more sophisticated measures that can assess all aspects of a text’s complexity, c.f.  Falkenjack et.al (2013) and Heimann Mühlenbock (2013) for Swedish initiatives. At the Department of Computer and Information Science in Linköping, several interactive tools have been developed that measure text complexity but also automatically adapt digital text from various perspectives including summarization, lexical and syntactic simplification (Falkenjack et al., 2019). The tools have been developed in collaboration with both writers of easy to read texts and people with reading and writing difficulties and can easily be modified to accommodate new user groups. The usefulness of the various text adaptation techniques and the measures of text complexity have, however, never been systematically evaluated for various groups of people with reading disability. Reading performance is often described at either the individual level (the reader has a set of abilities), or at the text level, (complexities integrated within the text). The aim of the present study is twofold; to learn more about reading performance and reading disability by examining the relationship between comprehension and text complexity in unique controlled ways, and to determine if reading comprehension can be improved by customized automatic adaptation of texts. We will use adaptation which includes lexical and syntactic simplification as well as automatic text summarization.