As I pointed out in a 2023 paper,1 the principal limitation of Skinner’s analysis of verbal behavior is that it fails to address grammar. To put it somewhat differently, it fails to explain why some novel sequences of words are considered “correct”—in the sense that listeners do not notice anything odd about them—while equally intelligible novel sequences sound “odd,” or “wrong,” or “ungrammatical.” Why don’t all novel sequences sound odd?
Let’s consider one of the examples of the noted linguist, Stephen Pinker:2 The verbs give and donate are closely related and in some contexts synonymous, but they have separate “rules of usage,” in the sense that there are conditions under which we would not substitute one for the other. For example, one might say, My uncle gave the museum an Icelandic manuscript. It is almost certainly novel to you, the reader, but you have no trouble understanding it, and nothing about it sounds odd, or troubling. But consider a second example: My uncle donated the museum an Icelandic manuscript. Although it is equally intelligible, most native English speaker, and probably many non-native English speakers, would find this way of putting it “odd;” the usual way to say it would be: My uncle donated an Icelandic manuscript to the museum.
To put it in grammatical terms, give is a dative verb following which the direct object and indirect object can occur in either order; donate is a roughly synonymous dative verb following which the direct object and indirect object must occur only in that order. Novel utterances that conform to the rule sound “normal” while those that violate it sound “odd.” But I’ll bet nobody ever taught you that rule. Not because your teachers and parents were in league to keep you ignorant, but because nobody but a few linguists ever noticed the rule. Nevertheless, most fluent speakers of English speakers have somehow learned it by example, but the examples we learned it from presumably did not include my uncle, a manuscript, and a museum. What does the science of learning have to say about it?
Very little.
Very little, but not nothing: In Verbal Behavior, Skinner introduced the concept of autoclitic frames, (aka intraverbal frames), a pattern of fixed terms into which an indefinite number of variable terms are “inserted” according to circumstances. Prepositional phrases are clear examples: the frame under the X comes to strength in countless circumstances of superposition, but X will vary from one case to another and is supplied by the context. Three examples of autoclitic frames that are relevant to the present case would be X donated Y to Z; X gave Y to Z; and X gave Z Y., where X, Y, and Z are, in grammatical terms, subject, direct object, and indirect object, respectively. Learning a tact can occur in isolation (That’s a rutabaga), but learning the frames in which a term occurs cannot. That is, the fixed parts of an autoclitic frame are never emitted alone. They are the matrix in which an indefinite number of variable terms take their place, but those variable terms are constrained. Mo is pushing Larry into Curly differs from Larry is pushing Curly into Mo, as it differs from the four other permutations of the cast of characters, and the puzzling question for behavior analysts is how—moment to moment in time—stimulus control over verbal behavior shifts as speech is emitted to produce novel strings. I say “moment-to-moment in time” because our behavior differs in novel ways moment to moment. We might utter an intraverbal chain—a memorized utterance—under control of a single, unchanging antecedent, but novel functional sequences must have fluctuating controlling variables.
The puzzling question would have an easy answer if we consulted an instruction manual while speaking: “When about to embark on a statement in which the listener is interested in who was pushed into whom, by whom, a possible option is to say, X is pushing Y into Z, where X is the one doing the pushing, Y is the person being pushed, and Z is the person into whom Y is pushed.” So by implication, our experience as English speakers might provide us with the functional equivalent of such a manual: X is not merely a tact under the control of X, it is multiply controlled by X and by its role as the agent of the action (i.e., as the subject); Y is multiply controlled by Y and by its role as object of the action, and Z is multiply controlled by Z and by its role as recipient of the object.
For many years, I dismissed this explanation, for I could not see how “roles,” like “agency” or “object of an action” could have any common stimulus properties that would allow speakers to generalize across their multifarious manifestations. Almost anything can be an agent or an object.
But as I explained in my paper, I recently learned that Old English, the Anglo-Saxon precursor to modern English, was an inflected language, like Latin and Ancient Greek. The roles of agent, object, indirect object, and a few others were identified by case inflections, usually word endings, vowel changes, or distinctive articles. That is, the word for the object of an action would consist of its “meaning” and its case inflection as a direct object. The fact that Alfred the Great or the Venerable Bede could effortlessly tact both a stimulus and its grammatical role showed that I was wrong. Despite their multifarious manifestations, roles obviously can be tacted with a common verbal form.
In a presentation of the paper to the ABAI Theory & Philosophy conference in Chicago three months ago, I speculated that case inflections serve the same function as autoclitic frames, and for that reason, Old English could dispense with autoclitic frames, and modern English could dispense with case inflections. I was generalizing from my teaspoonful of knowledge of Latin: I knew that in Latin, word order can be scrambled for rhetorical effect, but as I am not a native speaker of the language, I cannot say how free word order actually is.
As an American, I am only dimly aware that there are languages in the world other than English. Perhaps I should have consulted one of them. Fortunately, one of the people attending the Chicago conference was Anna Pettursdottir, behavior analyst of the first water and a native speaker of Icelandic. Moreover, unlike most people attending a conference presentation, she does not check her e-mail and compose shopping lists on her cell-phone during the talk, but rather listens and takes notes. I had illustrated an autoclitic frames with a drawing of a mouse pulling an elephant on a dolly:

[Picture credit to So-Young Yoon, formerly of the F. S. Keller School, Yonkers, NY]
It can be described with either the active voice autoclitic frame (“The mouse is pulling the elephant”) or the passive voice autoclitic frame (“The elephant is being pulled by the mouse”). If it were the elephant doing the pulling, the terms would be reversed. By hypothesis, in Old English, a single sequence of terms, appropriately inflected, might have been sufficient.
A few minutes after my talk ended, I received an e-mail from Anna with the following attachment, featuring six alternative ways of saying each phrase in Icelandic:

It nicely supports the hypothesis that case inflections serve the same function as autoclitic frames. All six orderings of terms are acceptable and synonymous in Icelandic but not in modern English. Icelandic is an especially apt comparison, since it emerged from the same Germanic roots as Old English and is actually closer to Old English than is modern English. (The latter evolved rapidly following the Norman Conquest of Britain in 1066.)
I was grateful to Anna for evidence confirming that case inflections and autoclitic frames are at least to some extent redundant, for it greatly helps explain the shifts in stimulus control as we speak. Case inflections are explicit tacts of “roles” played by the topics of our conversations and are sufficient evidence that such roles have discriminable properties.
In closing, I will take the opportunity offered by an informal blog—free from the scrutiny of meddlesome editors who would politely but firmly ask for evidence—to speculate that the controlling variable for our tacts of “roles” is, at least in part, under control of our responses to the world, not solely under control of the world itself. When we watch a movie, we incipiently act with the actors; when we hear that someone has donated a manuscript to a museum, we incipiently feel the excitement of the archivist; when we see a dog chase a fox we incipiently run. That is, to observe the world is more than merely to see the world. To see is to sense, to observe is to behave. The difference is subtle, but surely you, the reader, know what I mean, and I am suggesting that that flimsy and nebulous tissue of evanescent behavior is one of the controlling variables for our orderly but novel verbal behavior.
Nice!!!