From Encoding to Remembering: Pragmatic Inferences Reveal Distinct Routes of Word Learning in Autistic Children

Katherine Marie Trice^*Zhenghan Qi

• Northeastern University, Boston, Massachusetts, United States

Mentalizing skills—the capacity to attribute mental states—play critical roles in word learning during typical language development. In autism, mentalizing difficulties may constrain word-learning pathways, limiting language-acquisition opportunities. We ask how autistic children encode and retrieve novel words and what drives individual differences. We test whether autistic children’s word learning benefits from pragmatic inferences, as in non-autistic (Trice et al., 2025). Forty-nine 6-to-9-year-old verbal autistic children participated. During learning, four novel words in the direct-mapping condition (DM) could be uniquely mapped to one novel object and four in the pragmatic-inference condition (PI) required children to assume speaker intent. Immediate recall and retention (15-minute delay) were tested via four-alternative-forced-choice-task. Autistic children showed above-chance PI mapping, no immediate recall differences, and PI retention advantage. However, individual difference analyses suggest a bimodal PI-retention pattern: 55% showed above-chance PI word recognition (PI-Retained) and 45% at-or-below-chance (PI-Limited). Retention profiles don’t reflect general memory—most PI-Limited children remembered DM words well. Instead, profile was associated directly with learning success. For PI-Limited specifically, learning performance was at-chance. Eye-movement during learning showed converging evidence: only PI-Retained consistently diverged between looks-to-target and competitor. Only nonverbal IQ in conjunction with initial mapping reliably differentiated groups, not mentalizing or language measures. This suggests distinct pathways of word-meaning acquisition in autistic children with otherwise similar profiles. While PI resolution may facilitate word-meaning acquisition for some, DM better serves others. This underscores the importance of characterizing learning processes as a pathway to understanding the heterogeneity of language in autism.