Quantifying narrative ability in autism spectrum disorder: a computational linguistic analysis of narrative coherence.
A one-minute computer scan shows autistic learners keep up on picture stories yet lag on open recall.
01Research in Context
What this study did
Losh et al. (2014) fed two short stories into a computer program. The program counts how well each sentence links to the next.
They asked high-functioning autistic students and neurotypical peers to tell a picture-book story and later recall it from memory. The computer gave every story a coherence score.
What they found
On the picture task both groups earned similar coherence scores.
On the memory recall task the autistic students scored lower. The tool spotted the drop quickly.
How this fits with other research
Losh et al. (2003) already showed autistic children slip when stories need feelings or cause-effect links. The new study adds a fast computer check for the same problem.
Mulder et al. (2020) looked at witness stories and found equal coherence. Their task gave a clear script; the 2014 recall task gave none. The two papers seem to clash, but the difference is open recall versus structured prompts.
Saravanaperumal et al. (2025) worked with younger Tamil-speaking preschoolers. Even when language level matched, autistic kids told shorter, simpler tales. The trouble starts early and stays.
Why it matters
You can run the same computer check in under a minute. Use it to see which tasks trip up your learner. If coherence drops on open questions, add visual cues or sentence starters. Keep picture-based work for quick wins.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →After story time ask the learner to retell without pictures, run the free LSA tool, and note any coherence dip to guide next week’s supports.
02At a glance
03Original abstract
Autism is a neurodevelopmental disorder characterized by serious difficulties with the social use of language, along with impaired social functioning and ritualistic/repetitive behaviors (American Psychiatric Association in Diagnostic and statistical manual of mental disorders: DSM-5, 5th edn. American Psychiatric Association, Arlington, 2013). While substantial heterogeneity exists in symptom expression, impairments in language discourse skills, including narrative (or storytelling), are universally observed in autism (Tager-Flusberg et al. in Handbook on autism and pervasive developmental disorders, 3rd edn. Wiley, New York, pp 335-364, 2005). This study applied a computational linguistic tool, Latent Semantic Analysis (LSA), to objectively characterize narrative performance in high-functioning individuals with autism and typically-developing controls, across two different narrative contexts that differ in the interpersonal and cognitive demands placed on the narrator. Results indicated that high-functioning individuals with autism produced narratives comparable in semantic content to those produced by controls when narrating from a picture book, but produced narratives diminished in semantic quality in a more demanding narrative recall task. This pattern is similar to that detected from analyses of hand-coded picture book narratives in prior research, and extends findings to an additional narrative context that proves particularly challenging for individuals with autism. Results are discussed in terms of the utility of LSA as a quantitative, objective, and efficient measure of narrative ability.
Journal of autism and developmental disorders, 2014 · doi:10.1007/s10803-008-0618-y