MT - Language as Control of Perception

Posted by Bruce Nevin:

In PCT linguistics, as in any research field for PCT, we first have to identify what perceptions are controlled. Every concept or proposal that we look to bring in from the established methods and theories of linguistics is subject to this scrutiny.

Phonetics concerns the range of sounds and articulatory postures and gestures that can be controlled in speaking. A phonetician can control these while not controlling the perceptions that constitute any given language. Which differences of sound are controlled to make different syllables and words in a given language? In English, foo and who contrast, but in Japanese it doesn’t matter whether you say the name of the big mountain southwest of Tokyo as Fuji or Huji.

Most of the time when we talk about language we look at written words. Writing substitutes differences between letters (or between word-shapes made of letters) in place of the sounds and articulations of speech. The sounds and articulations do not begin and end within letter-size segments; they overlap, and the production of each is a disturbance to the production of its neighbors (what is called co-articulation). We set all that aside when we consider written words. We don’t perceive and control syllables (unless we’re sounding out an unfamiliar word), we perceive and control words.

Linguistic theories get most complicated and disputatious when we get to relationships between words, perceived entities of some sort that are made of word relationships (talked about as e.g. syntax), and associations between those words and entities in language and our multitudinous non-language perceptions. (The latter associations are taken up by linguists in e.g. semantics and pragmatics.) For every proposed entity, relationship, process, rule, etc. we have to ask: Can this be perceived? Is it in fact perceived? Can we disturb that perception and test for control of it? Are we sure we’re not disturbing some other perception, which is in fact what is perceived and controlled? Is the proposed perception a part of a complex perception or set of perceptions which are controlled collectively by linguists, philosophers, logicians, literary critics, or some other collectivity of people when they are talking about language? Are they in fact among the perceptions that Joe and Alice control as they discuss the menu and place their order with the waiter in a restaurant?

An example is the trace t that some linguists perceive as occurring at a location whence they perceive that a word was moved in the course of uttering e.g. “He asked which salad dressing you prefer t ?” Since there are other ways of talking about language which do not propose that saying (or writing) that sentence requires us control a perception of moving a word from one place to another, we may suspect that speakers, writers, hearers, and readers of that sentence in actual living situations of using language, such as Joe and Alice in that restaurant, do no such thing. How do we test that? This is the sort of question that PCT brings to bear on all such proposals from the established methods and theories of linguistics.

Two contributions to the Festschrift for Bill powers are on language.

Joel Judd, Language: The Control of Perception

Bruce Nevin, What is controlled in speaking and in speech recognition?

Martin Taylor has written extensively on language in a partially articulated form of PCT called Layered Protocol Theory (LPT).

Layered protocols for computer-human dialogue. I: Principles

Dialogue analysis using layered protocols

Posted by Martin Taylor:

Thank you for mentioning Layered Protocol theory, but I’m not clear why you call it a “partially articulated form of PCT”. I don’t think of it as a form of PCT at all. Rather, I think of it as a specialized application of PCT to a situation in which two (or sometimes more – in my chapter in the upcoming Handbook of Perceptual Control I have an example to a three-person interaction – parties (human, animal, fish, or vegetable, or even non-living computer) cooperate in trying to reach agreement on something, or perhaps try to deceive one another into perceiving that agreement has been reached. It’s a problem in two parts. Part 1 is the control of the relationship by both parties between two perceptions of the “something”. Part 2 is how the parties interact in controlling the relationship between two perceptions that cannot be directly compared, since they are in two different control hierarchies.

I won’t go into the details of protocol functions here, as it takes a whole paper or book chapter to deal with the various possible complications, because there are nineteen perceptions to be controlled at any one level of a protocol hierarchy, ten in one party, nine in the other. Suffice it to say that an “initiator” I will call Ivan perceives, rightly or wrongly, that a “responder”, who I will call Rob, has a perception of something that differs from Ivan’s reference value for the perception Ivan wants Rob to have. Ivan can’t know the state of Rob’s perception is, or even if Rob has any perception of it at all, but as we know, Ivan has a way to find out – the Test for the Controlled Variable.

If Rob isn’t controlling the perception of interest to Ivan, Ivan’s attempts to disturb it won’t be followed by any consistent action to restore it to its initial value. If, for example, Ivan says “Fran went to the ball game yesterday”, and Rob says “No she didn’t. She was with me.” Ivan knows that Rob wants him to change his perception of Fran’s doings yesterday. Why Rob would do this is a question about some higher-level perception Rob is controlling. Maybe Rob is lying because he wants Ivan to feel jealous, maybe he wants Ivan to perceive the world as he, Rob, does. It doesn’t matter to the theoretical interpretation, though it may matter to Ivan, who may follow up with “You are kidding, aren’t your?”. Even if he asks this, Ivan knows that Rob is controlling a perception of Ivan’s perception of Fran’s doings yesterday.

The “Layered” aspect of LPT, and eighteen of the controlled perceptions supporting a layer, has to do with Part 2 of the problem. In controlling any perception, the output of the output function of an ECU produces reference values for lower-level perceptions in a hierarchical fashion. In a dialogue, all the control loops at every level pass through the control operations of both parties. Each such loop is potentially an entire protocol, but in any practical situation each party will perceive a lot of the perceptions controlled to be already at their reference value, and no overt change of action can be observed by an outside observer.

For example, Ivan might believe Rob to know from the recent history of their interaction that Ivan believes Rob want’s to get together for lunch. Ivan might quite probably say no more than “Where? When?”. Or a couple might stop at a particularly scenic viewpoint. He might wave an arm from one side of the view to another, saying “Wow”, to which she might respond “Oh, my”. In neither example was anything said more than necessary, not was anything omitted from what could have been said as the sentences “Isn’t that a fantastic view” “Yes, I really agree, it is a fantastic view”. If something extra is actually included in either party’s overt behaviour, is is likely to be perceived by the other as indicating some problem in the “over-expressing” party’s perception of the other’ perception. That way lies confusion.

The situation is different when the message is conveyed in written language (not including texting). Apart from texting, the reader has no opportunity to query what the writer intended. The redundancy necessary to ensure accuracy in getting the message across has to come from within the written text, in the form of “syntax” at different levels, syntax in word spelling,syntax in word relationships, syntax within paragraphs, and so on. Mostly in this written message, I am using speech as the medium into which the message is translated.

Th main point here is not to object to Bruce’s characterization of Layered Protocol Theory, nor to criticize Bruce’s argument about collective control of the forms of a specific language. Indeed, I use the same arguments about collective control of the forms of protocols used in different cultures and linguistic communities. If I would criticize the argument at all, it would be that the collective control of “standard” forms of language or protocol are not necessarily the forms used in any particular interchange. If either party uses a form whose function is not understood by the other party, when the user intends it to be understood, whatever the user intended to be controlled will not be moved closer to its reference value. That way lies learning, either by a child or by a foreigner.

The “collective control” aspect of this occurs when the child or the foreigner has the same kind of failure in getting more people to do what he hopes. That failure could be at any level, from the pronunciation of a vowel or consonant, the use of wrong prosody (pitch and amplitude contour), or expectation that the other understands the underlying nature of Fourier Transforms. Usually, however, the message in the higher-level protocol can be understood despite failures at lower levels, either because of redundancy in it expression or because of repeated interchanges in which each party tries to clarify the meaning of the message.