PLoS ONE (Jan 2022)

Toward mapping pragmatic impairment of autism spectrum disorder individuals through the development of a corpus of spoken Japanese.

  • Sumi Kato,
  • Kazuaki Hanawa,
  • Vo Phuong Linh,
  • Manabu Saito,
  • Ryuichi Iimura,
  • Kentaro Inui,
  • Kazuhiko Nakamura

DOI
https://doi.org/10.1371/journal.pone.0264204
Journal volume & issue
Vol. 17, no. 2
p. e0264204

Abstract

Read online

The central symptom of autism spectrum disorder (ASD) is deficiency in social communication, which is generally viewed as being caused by pragmatic impairment (PI). PI is difficulty in using language appropriately in social situations. Studies have confirmed that PI is the result of neurological, cognitive, linguistic, and sensorimotor dysfunctions involving intricately intertwined factors. To elucidate the whole picture of this impairment, an approach from a multifaceted perspective fusing those factors is necessary. To this end, comprehensive PI mapping is a must, since no comprehensive mapping has yet been developed. The aim of this research is to present a model of annotation scheme development and corpus construction to efficiently visualize and quantify for statistical investigation occurrences of PI, which enables comprehensive mapping of PI in the spoken language of Japanese ASD individuals. We constructed system networks (lexicogrammatical option systems speakers make choices from) in the theoretical framework of Systemic Functional Linguistics, from which we developed an annotation scheme to comprehensively cover PI. Since system network covers all possible lexicogrammatical choices in linguistic interaction, it enables a comprehensive view of where and in what lexicogrammar PI occurs. Based on this annotation scheme, we successfully developed the Corpus of ASD + Typically Developed Spoken Language consisting of texts from 1,187 audiotaped tasks performed by 186 ASD and 106 typically developed subjects, accommodating approximately 1.07 million morphemes. Moreover, we were successful in the automatization of the annotation process by machine learning, accomplishing a 90 percent precision rate. We exemplified the mapping procedure with a focus on the spoken use of negotiating particles. Our model corpus is applicable to any language by incorporating our method of constructing the annotation scheme, and would give impetus to defining PI from a cross-linguistic point of view, which is needed because PI of ASD reflects cross-linguistic differences.