Natural language processing, pragmatics, and verbal behavior.

Gillberg (1992) asked a bold question: can Skinner’s Verbal Behavior make AI talk like a person? The paper maps each verbal operant to a computer module. The goal is a chatbot that cares about consequences, not just word strings.

No kids, no trials—just a blueprint. The author shows how mands, tacts, and autoclitics could guide a program to pick words that matter to the listener.

The article argues that pragmatics—why we say things—already lives inside Skinner’s boxes. Feed that structure into code and you get context-smart replies without hidden mental rules.

A sample architecture is sketched: sensors fire tacts, user desire fires mands, autoclitic frames glue the sentence together. The bot learns by consequences, same as a child.

Nangle et al. (1993) extends the idea. They prove intraverbals stay separate from tacts in real preschoolers, giving the AI model a clean operant to copy.

Hursh et al. (1974) also extends. Toddlers produced brand-new sentences after simple imitation plus reinforcement, showing Skinnerian rules can create generative language without grammar modules—exactly what the AI paper hopes to code.

Dosen (2005) is a successor paper. It warns that Skinner later dropped his old S-R skin. If you build the bot with rigid stimulus-response links, you would repeat a mistake Skinner already fixed.

You probably won’t code a chatbot this afternoon, but you can borrow the lens. When you write programs or teach staff, think in operants plus consequences instead of ‘meanings inside words.’ Check that your intraverbal drills really are intraverbals—see Nangle et al. (1993). And remember reinforcement can create novel language, so give kids room to combine words you never directly taught, as shown in Hursh et al. (1974).