Generative Language Learning in Severe Autism: Experimental Evaluation of a Mobile Application

Background: Severe autism includes a “delay in, or total lack of, the development of spoken language”. Many learn to communicate through alternative means such as tablets and mobile technology. However, utterances on these devices are often very limited; minimally-verbal individuals with autism do not surpass single-word responses limited to requesting and labeling purposes, and vocabulary repertoires are small. Matrix training is a language intervention to systematically build up vocabulary and teach longer word combinations to produce more complex utterances. In this generative approach to instruction, words are arranged in a matrix format so that some multiword phrases are taught and others develop without direct instruction. Specifically, linguistic elements (e.g., nouns, verbs, etc.) are presented in systematic combination matrices, which are arranged to induce generalized rule-like behavior, a particular difficulty in autism.

Objectives: A mobile application, SPEAKmore!, was developed to carry out matrix training on a tablet device (see Figure 1). This study aimed to answer

1. Does language training with SPEAKmore! facilitate production of action-object combinations on a tablet device? This was accomplished by measuring the percentage of correct target forms in intervention probes.
2. Do newly learned skills generalize to untrained action-object combinations. This was achieved by taking generalization probes during the intervention phase assessing performance on combinations that were never taught before.

Methods: An experimental single subject design, that is a multiple probe design (Horner & Baer, 1978), was used across sets of action-object combinations with generalization probes of untrained combinations. This design is currently implemented with five participants, who are between 8-12 years old, have an official diagnosis of severe autism according to CARS-2 and ADOS-2 scores, qualify as minimally verbal by having no more than 10 spoken words, and communicate primarily on a tablet. These students were taught action-object combinations on a 6x6 matrix with SPEAKmore!. From the total pool of 36 possible symbol combinations, the researcher created four different sets of three symbol combinations each that were actively taught. The remaining 24 combinations were tested for generalization effects. 

Results: Preliminary results are available for two of the five participants. Figure 2 shows participants’ performance measured as the percentage of correct symbol combinations. Both participants demonstrate a similar pattern of successful acquisition of word combinations during the intervention condition and subsequent generalization to untrained stimuli. Within three intervention sessions, both participants reached over 80% correct and their performance remained at this level. Performance on generalization increased steadily for both over the course of intervention. Effect sizes as measured by the Non-overlap of all Pairs Index indicate a medium-strong effect for participant 1 and a strong effect for participant 2.

Conclusions: Results suggest that matrix training through a mobile application may be a promising approach to teach new vocabulary and enhance the complexity of utterances for tablet communicators with severe autism. To further investigate the robustness of this technology intervention, findings need to be replicated using (a) different language targets (e.g., agent-action, adjective-object combinations), and (b) expansions from two-term to three-term semantic relations (e.g., agent-action-object).