Festschrift for William T. Powers

Pronunciation: ‘fes(t)-“shrift

Etymology: German, from Fest celebration + Schrift writing
Date: 1898
: a volume of writings by different authors presented as a tribute or memorial, especially to a scholar


The William T. Powers Festschrift is a contribution by persons influenced by the work of William T. Powers in honor of the 30th anniversary of the publication of Behavior: The Control of Perception. The contributions range  from personal tributes to thematic essays. They have come in from  all over the world and were put together on various local  versions of word processing programs. I have refrained from  editorial intervention, so that the style and format used by the  authors could be retained. There are still some inadequacies of presentation in a few of the contributions. I apologize  for any inconvenience this may cause, but it was impossible for  me to alleviate all of these problems at the last moment.  Corrections  are welcome and will be incorporated into the Festschrift for eventual final presence on the web.

Since the contributions  are only a mouse click away, it is unnecessary to comment on them  any further. However, I would like to offer a partof the “web site  story” as my contribution. The emergence of this project  began in Saint Louis at the 2000 CSG conference. Realizing the forthcoming anniversary of the publication of B:CP in 2003 and being aware of festschrifts as ways of recognizing such landmark events, I suggested the notion of a web-based festschrift forWilliam T. Powers. The suggestion met with immediate approval and was supported as energies allowed throughout the project. The organizational work, approaching contributors,  getting emails out to them, doing the amateur webdesign and construction and doing all those  “little” things that pave the way for a project like this, was done by myself. Fred Good, Dag Forssell, Rick Marken and, of course, Mary Powers and her daughters, Barbara and Allie, helped. Thanks to all who helped and all who contributed and who will contribute to this project in recognition of William T. Powers.

After the presentation of this webbased Festschrift, contributors arestill invited to submit additions to this project in honor of WilliamT. Powers’ contribution to science.

Lloyd Klinedinst

Organization by Topics

  1. Papers Related to PCT
  2. Personal Reflections
  3. Pictures Related to William T. Powers and PCT History
  4. Miscellaneous
  5. Links
  6. List of Contributions, Alphabetically Ordered by Author’s Name

1. Papers Related to PCT

1.1 PCT—Yeah? So What?! By Timothy A. Carey

 Living things do not respond to stimuli. Nor do their brains command their muscles to behave in certain ways. Living things control their experiences. Controlling experiences seems a simple notion yet its implications are profound. My observation has been that when this idea is explained to most people heads nod in understanding and agreement. People seem to easily notice the phenomenon of control once they know what they are looking for. While control becomes apparent to most people who take the time to notice, what remain invisible are the implications of this phenomenon. Let me see if I can describe what some of these implications might be.

    Behavioral scientists who understood that living things controlled their perceptual experiences would not be interested it trying to find connections between stimuli and responses. They would become disinterested because they would know that there is no connection between a stimulus and a response. In fact, not only would they understand that the link between stimulus and response is chimeric, they would also wise up to the fact that, as far as living things are concerned, there is no such thing as either a stimulus or a response. Calling something a “stimulus” is to assert that this something is able to stimulate another something. A stimulus is only a stimulus to the extent that it produces a response in the thing it is stimulating. In the same way, calling something a response is to suggest that this something was brought about by a stimulus.

    Perhaps the terms stimulus and response would linger on in the vernacular. After all, we still talk about sunsets and sunrises rather than “planetary rotational illusions” or “visual orbital effects”. When working to learn about the nature of living things, however, the activity of stimulating a living thing and measuring its responses would vanish. Rather, people who wanted to know about living things would investigate both what and how perceptual experiences are controlled.

    The seemingly innocuous idea of studying the process of control rather than the stimulus/response connection would irrevocably change the activity of perhaps every behavioral scientist who clocks in for work today. Rather than scrutinizing the eye-blink response and the way it varies in relation to different puffs of air, a control scientist would be interested in what conditions of the eye are being maintained in particular states. What is the moisture level on the surface of the eye? How much variation in moisture level is tolerated before this variation is eliminated? Is a particular amount of pressure on the surface of the eye also kept constant?

    Answers to new questions about what and how an eye controls would begin to shed some light on the mystery of how an eye goes about the business of seeing. Some things about what we already know would probably still be useful. In attempting to answer new kinds of questions about what an eye does, however, we would have to become a lot less sure about what we think we already know. For a time we would be less certain about that small part of the world that we thought we had pinned down. The functioning of the eye is one thing we would be less certain about once we recognized the controlling nature of things that live. But for people who have difficulty dealing with uncertainty, the bad news is that our knowledge about the eye would not be the only thing that would be affected. Everything, in fact, that we currently believe about living things would need to be reconsidered from a control perspective and almost everything would either be revised or rejected. 

    Seeking to understand different aspects of the phenomenon of control would of course be a fatal blow for the IV-DV research approach that currently exists. IV-DV is just a sneaky way of writing stimulus-response. The idea here is that we vary IV’s and measure the effect on DV’s. Behavioral scientists who cared anything at all about doing a good job realized long ago that there actually is no reliable connection between IV’s and DV’s. So that they could continue to think they were doing a good job—they were, after all, controlling their experiences—they began to use statistics.

    Statistics are to a behavioral scientist what a top hat is to a magician. Illusory relationships are made to look real when statistics are employed. Behavioral scientists discovered that if you gathered together a whole bunch of living things and engaged them in an activity, you could assign numbers to different aspects of the activity. Statistical procedures could then be used to combine the numbers that were produced from the results of all their individual activities. Methods were invented to turn all these individual numbers into just one number. The new number that was created was meant to represent something. That special single number could then be tested to see if it was special enough.

    If the number wasn’t special enough or if behavioral scientists wanted to make it seem even more important they created a bigger bunch of living things. The more living things the more important the number. Unfortunately, as the bunch of living things grows and as the impressiveness of the combined number soars, the less we know about any particular individual in the bunch. This is no way to unearth the scientific laws of living.

    Other things can also be done to make numbers seem more important. Ultra-tricky statistical procedures have been created which add new bits to the numbers that are already there or even take bits away from the numbers that exist. This sometimes has the effect of making the number that represents the bunch become more important. These kinds of procedures will tell you a lot about numbers but they won’t tell you much at all about living things. Numbers don’t care what you do to them—living things do.

    Control scientists wouldn’t spend time combining large bunches of living things together to find out how they behave. Control scientists wouldn’t look for IV-DV relationships. Control scientists would want to know what living things control. To find that out they would study individual living things. They would identify the controlling characteristics of the individual and then they would look for similar characteristics in other individuals. Control scientists would be hunting for whatever it is that all living things of a particular kind control. After a successful hunt, control scientists would understand a little bit more about what it means to be that kind of living thing.

    Control scientists wouldn’t need to spend time learning to do tricky things with numbers. Instead they would spend their time learning to build models that work. When control scientists thought they had tripped over a good idea in their laboratory, their next job would be to build a model of the idea. If the model they built was able to control in the same way as whatever it was they were trying to understand then they would think their idea was pretty good. If the model didn’t control in the same way then they would need to do some rethinking. That’s how control science would work—playing with models, not numbers.

    Some behavioral scientists are so good at statistical procedures that they call what they do modelling. This modelling is very different from the modelling that a control scientist would do. Statistical modelling uses large bunches of numbers to create important connections between different sets of numbers. In other words, it is still trying to explain the IV-DV link. Sometimes in a sophisticated kind of way the IV is called a predictor variable and the DV is called a criterion variable. No matter. They are still describing the same kind of relationship. The toys in their playpen are still stimuli and responses.

    In many ways it would be unfair to leave the discussion of statistics at this point. The purpose of this foray into the world of statistical methods was not to malign the legitimacy of these procedures. It is not the techniques that are at fault but their application that is awry. When statistics are used to determine the relative importance of a particular number no reliable conclusions can be made about the behavior of any particular individual in the group from which the number came. Statistical methods will provide information about general characteristics of a group but they will yield absolutely zilch in terms of elucidating the principles and laws that govern the nature of living things. Statistics are not the problem. The stories created from their results are.

    Stimuli and responses can be disguised in other ways as well. One sly move that occurred a few years ago was to extract the stimulus from the environment and to pop it into the head of a living creature. Some optimistic people called this tricky maneuver a revolution—a cognitive revolution. When a stimulus is inside a living thing rather than outside it is called a cognition—or a belief, a thought, a brain command, an attitude—the labels are many but a stimulus by any other name …

    Because control scientists would use models to check their ideas they wouldn’t be very concerned with the name they attached to any particular bit. For a control scientist what would be important would be how the bit functioned in the model not the particular word sound that was used to name it. Control scientists then wouldn’t consider that they had improved their theory when they produced better sounding words. Nor would they think their model was better when the diagram looked nicer or when double headed arrows connected more boxes in a greater variety of ways. They would consider they had improved their theory when their model controlled like the creature they were modelling. We don’t need new word sounds; we need to figure out what’s going on.

    Obviously some things are going on and some things aren’t going on. If you’re trying to solve the puzzle of how something actually works, some answers will be correct and some will be incorrect. Behavioral scientists, however, are very reluctant to say that any idea is wrong. With a wave of the hand and a puff of smoke, statistics can be used to show that almost any idea is interesting and useful. Behavioral scientists came up with the convenient tool of “operationalising”. Operationalising just means saying that something you don’t understand very clearly can be what you say it is. Depression can be operationalized by counting up someone’s scores to answers on a test. The number they get on the test then represents how much depression there is. At least that’s the way the story goes.

    Another very handy strategy to use with operationalising is “forgetting”. Behavioral scientists tend to forget that the number they have obtained is, at best, a representation of depression. They say they are measuring depression, not scores on a test. They say that depression has decreased when the numbers get smaller and worsened if the numbers get bigger. They use statistics to do things to the numbers and then tell us what happened to the depression. They don’t seem to know that people are controlling even when they are putting circles around numbers on an official looking piece of paper.

    Control scientists would not be interested in operationalising. They would not be interested in what something was called. They would be interested in how something functioned. By being interested in the way something functioned they would have to accept that some ideas would be right and some would be wrong. Calling an idea “wrong” is not a nice thing to do when you’re mixing with behavioral scientists. With crafty methods like operationalising and statisticalising any idea can be made to seem even a little bit right. Behavioral scientists need long sleeves. The wizardry of statistics requires much sleight of hand.

    Control scientists would roll their sleeves up and spend time investigating the phenomenon of control. Control scientists would discard an idea if it was wrong. It would be OK to call an idea wrong with a control scientist. Control scientists wouldn’t spend time making wrong ideas seem a little bit right. They would want to know what it was that was wrong and how the model needed to be changed so that it would be right. “Right” to a control scientist would mean accurate. If the model accurately simulated what was being observed then it would be regarded as the closest thing to “right” at the moment. If the model wasn’t accurate then something about it would be regarded as being wrong and it would need to be changed.

    Wrongness and rightness seem to be uncomfortable notions for many psychologists and behavioral scientists. In the world of psychology anything goes. “Do what you wanna do, be what you wanna be, yeah …” is the tune that is hummed. The arena of psychotherapy is a startling demonstration. Someone who could be bothered to count them once estimated that there were over 400 different types of psychotherapy. This is inclusivity and acceptance gone berserk. As one method fails to help all people all the time it is tinkered with and adapted. Finally it is given a new name and an inspiring practitioner runs new workshops to teach new techniques that hold greater promise than previous methods. Some people think they understand what is happening and remain loyal dispensers of one method or another. Other people acknowledge that they don’t know what’s going on and call themselves “eclectic”.

    It is mind-numbingly obvious that not all of the different methods can be accurate in their explanations of the problem. If 20 different people were all treated for something called depression with 20 different types of psychotherapy and they all got better, then the 20 different methods used can undoubtedly not be the reason the people got better. Much of what is happening in psychotherapy cannot account for why people get better. A great number of the methods in psychotherapy are at best unnecessary and at worse detrimental.

    Control scientists would only want to do what was necessary. Control scientists who wanted to help people get better would not accept any explanatory story that came along. Control scientists would base their helping methods on the most accurate explanation of control that they could find. This of course would limit the kinds of things that control scientists could do. When methods are based on theoretical principles then some methods will be consistent with the principles and some won’t. The benefit of this limitation however is an accurate understanding of people’s problems. From this understanding comes an ability to assist people efficiently and meaningfully where assistance is required and also to determine when assistance is not necessary.

    If control was the only show in town much of the same type of activity would still occur. People would still have problems that they would need help with. Problems would need to be assessed, diagnosed, and treated. Where problems existed, however, they would be seen as problems with control. A something that is designed to control will only experience problems when its ability to control is interfered with in some way. When people with problems put themselves in front of a control scientist what would be assessed would be the people’s abilities to control different things. Perhaps the range and limits of their control abilities would be explored. Diagnoses would be statements about their abilities to control. Where treatment was required treatment would focus on helping people improve their abilities to control.

    Control scientists would understand that what people control are their perceptions. That is, what they see and hear and feel and taste and smell and touch—what they experience. People don’t control their behavior. If people are going to control their experiences they have to let their behavior change in whatever way is necessary so that their experiences continue to be the way they want. They have to constantly let the muscle tensions in their legs change so that they can remain upright on the deck of a rolling boat. They have to let their racquet go to the right place so that the ball will land in the middle and sail back over the net. They have to be prepared to turn the taps any which way until the temperature of the bath feels right. They need to squeeze the bottle differently so that the right amount of detergent oozes into the sink.

    So control scientists wouldn’t have to operationalize imaginary ideas like intelligence and personality. They wouldn’t assess people by asking them to produce numbers that supposedly represent different bits of intelligence or different types of personality. Nor would they use space-age machinery to make different bits of the brain light up in response to different stimuli. They would know that different colors on images of a brain would tell them nothing about difficulties in controlling.

    Control scientists wouldn’t diagnose people based on reports of behaviors that occur too much or too little and they wouldn’t treat people by trying to change their behavior. Social skills programs, assertiveness programs, and anger management programs are some of the programs that wouldn’t be necessary. The use of drugs to increase or decrease behavior would also not be necessary.

    We’re not sure just yet exactly what would be necessary. Many people seem to find a way to solve their problems when they are told to do different things with their behavior. But since people don’t control their behavior it can’t be anything about the particular behavioral program that helped them solve their problem. Nor could it be anything about a drug changing their behavior that helped them solve their problem. Somehow, when people are given ways of changing their behavior, some people manage to figure out how to control their experiences more effectively. Control scientists would prefer to use science rather than serendipity to help people solve their problems. Using serendipity is an approach that anyone is entitled to take. It is perhaps not unreasonable to expect, however, that when people practice serendipitously that they would say that’s what they are doing rather than pretending they are practicing scientifically. Serendipity and science are not the same and one should not be called the other.

    So let’s see if I can bring together all of what I am trying to say. The implications of control are profound. They reverberate through every area of investigation of living things. At a cellular level control scientists would be interested in how cells keep certain chemical concentrates at particular levels. Even before cells get off the ground we could benefit by thinking about genetic instructions as part of the process of control. In this way genes wouldn’t be seen to be commanding anything to occur other than particular states of the DNA strand. Everything else would be seen as a side effect of this control.

    But back to cells … By studying how cells control we might come to understand the normal functioning of a cell. Once we understand how a cell functions and what it actually is that a cell does when it’s doing normal celly things—we might also begin to comprehend what it is that goes wrong when problems occur. Perhaps for the first time we would begin to understand really scary diseases like cancer. Considering something like cancer from the perspective of cellular control would enable us to finally figure what’s going on when cancer occurs. Once we get to know cancer for what it is we’ll be in a mighty position to beat the rascal at it’s own game. Until that time we’ll continue to use treatments as battering rams and rely on the strategy of hope.

    At the level of cellular systems we’d strike it rich by considering problems from a control perspective. We might understand what happens to a pancreas when it isn’t able to control effectively anymore and someone develops diabetes. It’s also likely that we’d begin to understand what it is about control processes that were collapsing when someone develops a degenerative disorder like multiple sclerosis or Huntington’s disease. Once we understood what was going on here in terms of disruptions to control processes we’d be in a position to begin to design interventions that might arrest the degeneration and support the functioning of organs on strike.

    Moving up to individual creatures, I’ve already spent some time talking about how we might be able to understand the functioning of a creature better by considering their activity from the control angle. Problems would be understood as problems with control. Recovery from these problems would be recognized as the regaining of control in important areas. Tests and examinations would only assess a person’s control capabilities. Behavioral programs would vanish and people would be assisted to control more effectively not to increase or decrease certain behaviors—which they couldn’t do without controlling, even if they wanted to.

    Once we get away from individual creatures and start thinking about what happens when creatures gather together, the control perspective is still vital. If psychologists were control scientists they might be able to make themselves indispensable to governments and social planners by really understanding what is going on. As control scientists they would understand that individuals control their perceptions and they would know that it would be counterproductive to try to control an individual’s actions. They would know that, in the long run, it is useless to try to make people behave as we want them to without considering control as it seems from where they’re standing. Not bombs nor armies nor trade sanctions nor threats nor snubs at dinner parties will change another’s behavior unless, serendipitously, these particular strategies happen to affect what the other is controlling. If control was understood then methods of making others behave as we want would be seen as problematic and perhaps even dangerous. Important people might finally be able to behave like grown ups rather than children in a playground squabbling over a favorite toy. By considering the controlling nature of each other and themselves they could reach amazing decisions with glorious consequences for the humanity of now and later on.

    Control is important all over the place. In every possible way the earnest study of control will provide clarity regarding our most troublesome diseases, sicknesses, conflicts, and battles. That’s some “so what?”. But wait! There’s more … By assigning the phenomenon of control to it’s rightful place in the natural scheme of things, the life sciences will become a legitimate science for the first time. Currently division in the sciences of life is the order of the day. Division is commonplace as different strands of the one area battle for supremacy. In psychology it is important to know whether you are a clinical psychologist or a neuropsychologist or an educational psychologist or a research psychologist, and still there are more. The division is all the more quizzical considering that there is still no consensual body of knowledge to be divisive about. People in the field of psychology say that psychology is the study of behavior and yet they are unable to define what behavior actually is. This squabbling over positions on the club ladder is more reminiscent of rival craftspeople in a cottage industry than it is of a scientific endeavor.

    The study of control would provide a unifying thread from which a genuine scientific discipline could grow. The bar has just been raised. A standard now exists by which people can begin to authentically understand the functioning of living things. Once we understand how living things actually go about the business of living we will be well placed to eliminate forever some of the most pervasive problems that currently exist. In fact, some problems might not be problems at all once we learn to ask the kinds of questions that would be relevant from a control paradigm. That’s not bad for one little phenomenon and the theory that explains it.

    Thank you Mr. Powers.

1.2 What kind of science has Bill Powers wrought? by Dag Forssell 

For the Festschrift. . .

   How do I love thee. . . Let me count the ways! . .    

   Bill, you have brought us a new way of thinking about ourselves and life, a way that I find compelling based on reading your lucid writings and using what I have learned from you as I live, experience and reflect on life. Here is one feeble attempt to share with others the magnificence and significance of your achievement. I was still formulating this paper as we celebrated the fiftieth anniversary of the birth of PCT.

 Descriptive versus Generative Scientific Theories

    Love, Dag

1.3 The Spark! A Tribute to William T. Powers. By Perry Good

1.4 Language: The Control of Perception. By Joel B. Judd

Joel B. Judd
Adams State College
208 Edgemont Boulevard
Alamosa, CO  81102
for consideration in the
Festschrift for
William T. Powers

    It may only be a slight exaggeration to claim that the linguistic Holy Grail consists of explaining the connections between brain functioning and language.  In other words, a theory of human communication is an extension of our efforts to explain brain and behavior. If this is the case our beliefs about the brain should have direct implications for our characterization of language, and vice versa.  With apologies for co-opting the title of Powers’ seminal book, I suggest the most promising theory addressing the brain-language connection is Hierarchical Perceptual Control Theory (HPCT).

    A recent summary of sociolinguistics (Coulmas, 2001) also frames the real question for linguistics:

How is it that language can fulfill the function of communication despite variation?

    Powers’ insight was to emphasize the role of behavior in perceptual control.  Likewise, the key starting point for linguistics is to observe that language is just as purposeful as other behaviors.  One might argue it is the purposeful behavior, for it is quintessentially human.  Language forms the basis for most of our complex interactions with family, friends, colleagues and others.  It allows for the possibility of resolving disputes and conflicts without resorting to violence.  Although Coulmas’ statement refers primarily to the way languages change over time, it also defines the HPCT perspective.  From phonemes to syntax, all aspects of individual linguistic behavior—whether written, signed, or spoken—constantly vary, yet we manage in most instances to communicate sufficiently well with those around us to accomplish our day-to-day purposes.

    As with other behaviors, the fact that our linguistic behavior varies is not a new discovery.  Especially after 1900, the means to document individual linguistic variability increased along with measurement technology.  For example, Pillsbury and Meader (1928) conducted a small study among colleagues on the musculature involved in speech production and noticed that phonemes constantly vary.  Their conclusion: “It may be seriously questioned whether one ever makes the same group of speech movements twice in a lifetime.  If one does, the fact is to be attributed to chance rather than law” (p. 218).

    Again in parallel with behavior generally, linguistic variability at the phonemic level doesn’t usually impact our ability to communicate.  However, “higher” level variability in morphology and syntax may be crucial, even life-threatening (What did you call me?!).  Linguistics, though, has done little beyond describing (or in some cases ignoring) such variability.  In other words, most research has relied on attempts to tie observable—or measurable—language behavior to either environmental contexts or presumed cognitive faculties.  This has been the source of fundamental problems for much of linguistic research, both from a behaviorist and cognitivist standpoint.  As Stanley Sapon pointed out over 30 years ago, “…if we take as our only data the formal properties of an utterance, then the only predictions we can make are predictions of form, not of substance” (Sapon, 1971).

    Sapon is alluding to the research pitfall of relying on descriptions of observable behavior in order to predict future behavior (or explain it).  For example, attempts to predict exactly which member of the class of words called ‘noun’ will finish a sentence like “What I really like to eat in Summer is ________” have failed.  As a result, some researchers have labeled such prediction “trivial.”  However, Sapon finds it disingenuous for scientists—and in particular linguists—to

…describe [as trivial] the specific predictions which ordinary, unenlightened people are wont to consider crucial.  I, for one, look with compassion upon the craftsman who does not know how to produce estimable work and settles instead for esteeming the kind of work he can do. (p. 77)

    What HPCT offers is a recast of the important questions about individual linguistic behavior, how it might develop and function.  If there is a hierarchy to perception in general, then the same principles should apply to language as well.  If behavior is the control of perception, then linguistic behavior is the control of perception as well. 

A Perceptual Control Hierarchy for Language

The following hierarchy is after Powers (1973, 1989) and Judd (1992). 
Applicable linguistic concepts are assoc iated with the appropriate level.

 Level #Perceptual LevelType of Perceptual ControlLinguistic Equivalents
SUBJECTIVEREALITY11SYSTEMCoherent grouping of principles (‘citizen’; ‘religious’; ‘family’)One’s language “identity”; code-switching in bilinguals; language variation (Coulmas, 2001); socio-cultural aspects
10PRINCIPLE“Meta-awareness” (thoughts about thoughts); usefulness of programs; guiding ideas (‘honesty’; ‘politeness’)Pragmatics and usage; language proficiency or aptitude; “prototypes” (e.g., Competition Model)
9PROGRAM“If-then” decisions from lower levels (perceptual, not behavioral)Self correction using explicit rules: 3rd-person singular -s, ‘i’ before ‘e’; Krashen’s Monitor (Krashen, 1985)
8SEQUENCEPerceived order of events (‘beginning’ à ‘end’)Syntactic ordering of lexical items; description [ordering] of visual episodes (Tomlin, 1987)
7CATEGORYGrouping, naming, etc. of experiencesNaming (objects, concepts, parts of speech); semantics
6RELATIONSHIPConnections among events (causation)Ordering of “arguments” (e.g., “slots”, MacWhinney, 1987)
PHYSICALREALITY5EVENTWholeness of an activity; a “familiar pattern in time” (symphony)Lexemes (words); idiomatic phrases
4TRANSITIONChange over time (a 5th to major 7th)Intonation; stress-timed vs. syllable-timed languages
3CONFIGURATIONUnitary, recognizable arrangements (musical chords)Letters or characters; syllables; phonemes (though see Kaye, 1989)
2SENSATIONSound quality (pitch); alignment; edge, contrastN/A
1INTENSITYLoudness; brightness; pressure [1st order systems-only direct contact with outside world]N/A

    The preceding chart outlines some of the more self-evident connections between proposed hierarchical levels of perception and their linguistic equivalents.  If nothing else it should be clear that HPCT paints a picture of language that goes beyond the traditional phonology/morphology/syntax/pragmatics breakdown.  It also offers a structural basis for some kind of modularity (though probably not the innate modularism of Fodor), as Powers has argued:

. . . the division into levels and the subdivision of levels into independent control systems means that there is modularity at all levels:  each level consists of general-purpose control systems of a given type, available for an infinite variety of uses by whatever systems you want to add at the next level.  This makes sense in design terms, and in terms of development, evolutionary or within one lifetime.  If you want to build a self-organizing robot, make the system modular; that is, hierarchical. (Powers, CSGnet, November 8, 1991)

    The remainder of this discussion will offer a few examples that seem to evidence a good fit between HPCT and language.

    Our point of departure is the grouping of perceptual levels into “subjective” and “physical” reality.  Not only does this distinction make sense in general but particularly for language and language development.  There is growing evidence for similarities in the development of “lower-level” perceptions across languages.  The kinds of abilities referred to include sensitivity to the same kinds of temporal and spatial relationships defined by the configuration, transition, and event levels of perception (e.g., Petitto, 2000).

    As for the levels included in subjective reality, HPCT offers insights into language learning, linguistic variation, and other sociolinguistic and pragmatic questions such as: Why do men and women use language differently [gender differences]?  Does our language shape the way we think, or does the way we think shape our language [linguistic relativism]?  How does language serve to mark membership and identity [social class]?  Why might some people have a “knack” for learning language while others struggle [aptitude]?  What is the most effective way to learn another language [pedagogy]?

    In the area of second language acquisition and learning, HPCT may answer, or offer more satisfying explanations for, questionable distinctions and characterizations such as whether there is a difference between “learning” and “acquisition,” or between “instrumental” and “integrative” motivation; if there really are “compound” and “coordinate” bilinguals; or what “fossilization” really means.

    These questions are of course tied to the fact that as perception becomes more complex—as we move “up” the hierarchy—it also becomes more idiosyncratic and more subjective.  Many are the cortical stimulation studies which localize the same perceptions involved in somatic and propio-receptive behaviors across people (toe movement, vocalization, temperature sensation), few if any are the ones which find higher levels of perception (words, numerical calculation) in the same places across people (e.g., Ojemann and Whitaker, 1978 in the case of bilinguals).  As a result, HPCT implies a different research perspective for much of linguistics—one that takes a “specimen” approach (Runkel, 1990) rather than one which continues to rely on recording and tallying behavioral snapshots.

    All things considered, there doesn’t seem to be any good reason to propose a separate hierarchy for language.  While language is a kind of perceptual experience, perhaps the most important kind for humans as social beings, it is certainly not the only or even always the most effective type of experience.  At both extremes of the perceptual hierarchy there are what might be called “a-linguistic” perceptual levels.  At the initial levels, first-order systems for language perception are not any different than those for any other perception.  Intensities are intensities, whether they will be eventually be perceived as “language” makes no difference.  At the other extreme, the fact that we have difficulty putting into words concepts such as ‘love’ or defining terms such as ‘patriotism’ fits nicely with the idea that language is a means to an end.  With it we can act to control our perceptions, but ultimately it is not the only way we perceive ourselves or the world around us. 

Development and Foundation of Language

    Any discussion of language acquisition includes some discussion of philosophy as well.  This arises from the tendency of linguists (and applied linguists, and teachers) to speak of the ‘mind,’ while biologists, neurologists and others speak of the ‘brain.’  As a result, some have postulated linguistic constructs (e.g., ‘tense’ or “parameters”) that may have little connection to neurophysiological reality.

    There is in consequence a kind of dualism in language studies.  Jacobs (1988), in his discussion of language acquisition, points out,  “[D]ualistic thought on such matters is unjustified:  physiological states and mental states of the brain, incomprehensible as both may be, are one and the same . . . all learning involves anatomical/physiological alteration of neural substrate” (p. 306).  From a philosophical perspective, Bunge (1986) speaks of this same dualism when he says that “. . . the spinning of speculative hypotheses concerning immaterial ‘mental structures’ has taken the place of serious theorizing and experimenting on the linguistic abilities and disabilities of living brains” (p. 229), [emphasis added].

    In support of a unitary approach to language study (that is, one which avoids theorizing about mental structures that violate known neurological principles), some have begun to see principles of perceptual control in language.  Phonology again provides an example.  In HPCT the brain begins to distinguish printed or spoken language from other perceptions at the configuration level.  The quality of the perception doesn’t matter as much as recognizing that it is “language” and not some other kind of visual or aural perception.  As with, say, a musical chord, phonemes can be recognized whether a high, squeaky voice or a deep booming one produces them.  So while the traditional linguistic definition of phonemes as “basic units of speech” seems to be appropriate, there is still something missing.

    Perhaps this is why some have challenged traditional conceptualizations.  Kaye (1989) has said, “The phoneme is dead” (p. 154).  His claim stems from consideration of questions such as: how does one determine what a phoneme is?  Feature-based definitions become impractical when one considers how phonemes appear in spoken language (recall Pillsbury and Meader’s observations).  In short, many phoneticians are adopting approaches to phonology that deal with principles of phonological elements and their combinatorial possibilities and variations (Kaye, 1989, p. 164).  Such developments would support Powers’s description of configuration perceptions.

    Viewing phonology as part of a hierarchy of perceptual control can account for another problematic phonological entity, diphthongs.  While a diphthong may at first glance appear to require perception of transition (and perhaps developmentally children initially perceive them that way), it is categorized as phoneme, at least in English (Mackay, 1987).  This is because its sound contrasts meaningfully with other sounds.  English-speaking adults are generally completely unaware that the [ai] vowel sound in a word like ‘high’ is a diphthong since it is perceived as a single phone.

    Insights into other aspects of early language development also corroborate implications of HPCT.  Infants begin to perceive key phonemic distinctions within a matter of weeks after being born (Molfese, 1977; Molfese et al., 1983), and children in a monolingual environment settle on their language’s repertoire of sounds before a year has passed (Kuhl, Williams, Lacerda, Stevens & Lindblom, 1991).  Interestingly, though perhaps not surprisingly, Mack (1989) showed that children raised in a bilingual environment develop a “middle ground” for their production of similar phonemes in their two languages, thus simplifying the neurological load as well as satisfying reference levels—perhaps the PCT equivalent of killing two birds with one phone!

    Any discussion of development also raises the question of critical periods: is there one or more such periods for human language?  Strictly speaking, a biological critical period by definition requires that an organism have certain experiences within a particular temporal window or else key perceptions or behaviors never emerge (e.g., birdsong).  This is the strong form of the hypothesis.  Because such a clear-cut case for human language has never been documented (and it would be slightly unethical to artificially create one), weaker versions of this hypothesis—“sensitive” periods—are proposed for language (Snow, 1987).

    Part of the reason there is no all-or-nothing evidence no doubt stems from the fact that human language is complex, and becomes even more so if viewed in light of HPCT.  A true critical period for “language” would involve virtually the entire brain!  An alternative would be to propose multiple critical or sensitive periods: one for configuration-related perceptual development, another for transition-related perceptual development, and so on.  While again there is no definitive evidence for even this kind of perspective, there is some evidence of the need for early interaction with certain fundamental components of language.

    One of the most obvious results of someone learning a language after childhood (non-native or “L2” learning) is an accent.  Enough formal and anecdotal evidence has accumulated over the years to claim that the likelihood of having an accent in a L2 increases rapidly after about age five, and those individuals who learn a language after puberty almost never sound like “native” speakers.  Why is this the case?  One explanation could be that given the early timeframe for developing fundamental phonological perception and production, those who attempt to alter such perceptions later in life face the daunting task of modifying systems that have become automatized and subsumed into higher-level perceptions to the point where it is extremely difficult to access them again.  Furthermore, if the more mature (older) individual’s communicative needs (i.e., higher-level references) are met even with an accent, sounding like a native speaker may not be high priority.  Indeed, some interesting work has shown that certain accents are considered prestigious and may in fact be desirable!

Language and Subjective Perceptions

    What of higher-level, or subjective, linguistic perceptions?  HPCT offers just as provocative possibilities.  Some of Sapon’s concerns have born fruit as linguists and psychologists struggle to provide more than a descriptive account of language learning yet one that avoids the triviality he criticizes.

    Probably the main issue confronted by researchers is the logical problem of language acquisition.  This is the problem of overgeneralization, or the fact that children seem immune to [adult] attempts to correct their grammatical mistakes [1] .  Most well known linguists have promoted a search for some sort of innate constraint(s) on learning that allow a child to recover from overgeneralizing.  There is however at least one well-known alternative to the search for a “black box” and that is a model which fits nicely with PCT principles: the Competition Model (MacWhinney, 1987, 2001).  In contrast to the apparent either/or description of linguistic choices one finds in much research, close examination of language corpus reveal that children take a weighted approach to linguistic decisions.  That is, the acquisition of articulatory skills, vocabulary, semantic relationships, even grammatical roles are not neatly packaged, discrete choice mechanisms—one does not select his or her language components from environmental or genetic multiple choice databanks.  Instead, linguistic behaviors result from the interplay of various contextual and neurological factors, or in PCT terms, reference levels and disturbances to them.

    In fact, these abilities seem to fall more in line with “fuzzy logic” descriptions of decision-making.  A good example would be the development of linguistic “prototypes.”  Although the fact that we “learn” vocabulary and parts of speech gives the impression of finiteness to our linguistic knowledge, in reality that knowledge is fluid and variable.  The semantics for objects we drink out of, or for concepts such as ‘team,’ change as we gain relevant experiences.  In the same way, our [implicit] understanding of language grows and changes through our experiences.  These experiences help us develop prototypical examples and categories for grammatical relationships and other types of linguistic knowledge needed for effective communication.  They also form the basis for our decisions regarding such things as word choice (categories) or lexical ordering (sequence).

    Such a system would be more in line with our understanding of brain and neural functioning, and posits a view of language that reflects its functional nature (Bates & MacWhinney, 1989).  In fact, the implications of an HPCT perspective on language fits well with MacWhinney’s (1999) suggestion that the nature-nurture interaction be replaced with “emergentism”; that is, replacing models that stipulate “specific hard-wired neural circuitry” with “structures [that] emerge from the interaction of [biological and environmental] processes.

    Another but by no means final contribution of HPCT to understanding language involves the provocative nature of the “highest” levels of perception and their relationship with language and learning.  As already mentioned, a Perceptual Control Hierarchy for Language suggests that at both the lowest and highest levels our perceptions of the world diverge from mostly language-based perceptions.  It is important to realize that we can create a world of words that have little connection to personal (or “real world”) experience.  This danger was wonderfully described by Powers (1973):  

When that happens [failing to test symbol-manipulation against experience], word-manipulation carries them out over an empty abyss.  Words lead to other words, but all links with experience are left behind.  One can easily find himself chasing what may prove to be a ghost; what is the “real” meaning of intelligence or concept or vicarious meditation or quark?  (p. 166)

    Or in the case of language “aptitude” or “proficiency” and so on?  We need to be able to move beyond “word-manipulation” if we are to completely understand the connection between language and behavior.  HPCT offers perceptual control at a more overarching level than program or sequence.  These higher levels offer explanations for how influential our view of self is in our eventual attainment of language abilities.  The field of Second Language Acquisition (SLA) is a good source of tantalizing evidence.

    One popular notion in SLA is that of “fossilization”—the point(s) at which learners appear to cease progressing in their development of second language skills (Selinker, 1972).  As one might suspect, closer examination of individual subjects provides glimpses into higher-level perceptual control.  For example, Selinker (1972) cites a thesis by Kenneth Coulter that examined two Russian speakers learning English.  After investigating their failure to continue developing English language skills, Coulter concluded that some “strategy” lets learners know when they have “…enough of the L2 in order to communicate.  And they stop learning” (p. 217).  Another SLA researcher provided Selinker with personal commentary on learning strategies, suggesting that such strategies “…evolve whenever the learner realizes, either unconsciously or subconsciously, that he has no linguistic competence with regard to some aspect of the L2” (p. 219).

    Along these same lines, more recent work has focused on the strategies used by second language learners.  An interesting pair of students (Spanish-speakers learning English) observed by Abraham and Vann (1987) showed marked differences in the amount of time each employed key strategies (e.g., paraphrasing for understanding, repeating corrections) and how often they used such strategies.  The authors’ interpretation of their observations is revealing:

Gerardo appeared to take a broad view [of language]: his flexibility, variety of strategies, and concern with correctness suggest a belief that language learning requires attention to both function and form…In contrast, Pedro’s view of second language learning was relatively limited.  He appeared to think of language primarily as a set of words and seemed confident that if he could learn enough of them, he could somehow string them together to communicate.  The exact way they should be combined (and, indeed, the actual forms the words should take) was relatively unimportant.  Acting in accordance with this view, Pedro had adopted certain positive strategies that enabled him to be successful in unsophisticated oral communication…(p. 95, emphasis added)

    The italicized words suggest higher-level perceptual control that extends down through and influences many aspects of language ability.  It turned out that upon more detailed conversation with the two students Gerardo had aspirations of completing college and accepted the fact that effective, academic use of English was important to attaining this goal.  Pedro, on the other hand, had a narrow view of what he needed to know and was mostly interested in talking to girls on the beach.  Such revelations would make them excellent candidates for a version of “The Test,” though that is a topic outside the scope of this discussion.

    Finally, Tucker (1991) reported on adult immigrants to the U.S. and their progress in learning English.  One woman had emigrated to New York when she was in high school.  Her English writing was understandably poor—it lacked the expected elements of organization and grammar.  These faults persisted through school and two attempts at college.  However, in her third try at college suddenly this woman’s writing improved markedly.  In talking at length with her, Tucker found that since her last try at college she had divorced, gained insights into her own identity and abilities, and had developed professional goals for herself.  As with the two students above, it appears that with a marked change in her definition of concepts such as “independent” and “businesswoman,” this woman found that she could effect significant change in what had appeared to be static or “fossilized” language skills.

    The fact that such evidence fits well with HPCT encourages further investigation along these lines.  Who knows but what much of the tedious details involved in language learning would actually change if we simply understood better how learners ultimately see themselves?  And wouldn’t learning itself change when seen mainly as a question of whether learners perceive a discrepancy between where they “are” and where they “want to be”?

    This Big Picture view of language no doubt reflects a highly influential level of perceptual processing, one which deserves more attention.  As with other kinds of purposeful behavior, its effects have been noted for some time now.  Witness this observation, again from Pillsbury and Meader (1928):

When beginning to write, each stroke needs attention; when beginning to sew or crochet, each stitch; but with practice the vague intention to write a letter on a certain subject may be all that is necessary to take one to the typewriter in the other room and complete the writing. (p. 169, emphasis added)

    Later they offer an interesting summation of communicative purpose, “We are primarily interested in the effect we desire to produce upon the listener and so are usually not attentive to the processes by which we produce the effect” (p. 255).  Pillsbury and Meader are not alone; Dunkel (1948) concludes a chapter on speaking by suggesting:

… most of the time, “one thing leads to another” in speech and we end up by having said what we wanted to say.  We may have “purpose,” “volition,” “will,” “motive,” or what not.  Whatever this guiding force may be, we know little about it though it is the ultimate dictator of the whole process of speech.  (p. 60, emphasis added)

    The examples could go on, but the few given suggest a fruitful role for linguistic research from the perspective of HPCT.  The intent here was to point out some ways in which the principles of control theory, applied to language through a hierarchy of perceptual control, might lead to answers for key questions regarding language and language learning.  Certainly we are often skeptical when one theory looks to be an answer for everything.  But until something better comes along, HPCT promises a wealth of insights into the how and why of human language.

Works Cited

Abraham, R., & Vann, R.  (1987).  Strategies of two learners: A case study.  In A. Wendin and J. Rubin (Eds.) Learner strategies in language learning (pp. 85-102).  Englewood Cliffs, NJ: Prentice Hall.

Bates, E. & MacWhinney, B.  (1989).  Competition, variation and language learning.  In B. MacWhinney and E. Bates (Eds.) The crosslinguistic study of sentence processing. New York: Cambridge University Press.

Bunge, M.  (1986).  A philosopher looks at the current debate on language acquisition.  In I. Gopnick & M. Gopnick, (Eds.) From models to modules (pp. 229-239).  Norwood, NJ: Ablex Publishing Company.

Coulmas, F.  (2001).  Sociolinguistics.  In M. Arnoff & J. Ross-Miller (Eds.)  The Handbook of Linguistics (pp. 563-581). 

Dunkel, H.  (1948).  Second Language Learning.  Boston: Ginn.

Jacobs, B.  (1988).  Neurobiological differentiation of primary and secondary language acquisition.  Studies in Second Language Acquisition10, 303-337.

Judd, J.  (1992).  Second Language Acquisition as the Control of Non-primary Linguistic Perception: A Critique of Research and Theory.  Unpublished doctoral dissertation, Urbana, IL.

Kaye, J.  (1989).  Phonology.  Hillsdale, NJ: Lawrence Erlbaum.

Kuhl, P., Williams, K., Lacerda, F., Stevens, K., & Lindblom, B.  (1991).  Linguistic experience alters phonetic perception in infants by 6 months of age.  Science255, 606-608.

Mack, M.  (1989).  Consonant and vowel perception and production: Early English-French bilinguals and English monolinguals.  Perception and Psychophysics46(2), 187-200.

Mackay, I.  (1987).  Phonetics (2nd ed.).  Boston: Little, Brown and Co.

MacWhinney, B.  (1987).  The Competition Model.  In B. MacWhinney (Ed.).  Mechanisms of language acquisition (pp. 249-308).  Hillsdale, NJ: Erlbaum.

MacWhinney, B.  (1999).  The Emergence of Language (preface).  Mahwah, NJ: Lawrence Erlbaum.

MacWhinney, B.  (2001).  First Language Acquisition.  In M. Arnoff & J. Ross-Miller (Eds.).  The Handbook of Linguistics (pp. 466-487). 

Molfese, D.  (1977).  The ontogeny of cerebral asymmetry in man: Auditory evoked potentials to linguistic and non-linguistic stimuli.  In J. Desmedt (Ed.).  Progress in Clinical Neurophysiology, Vol. 3.  Basel: Karger.

Molfese, V., Molfese, D., & Parsons, C.  (1983).  Hemispheric processing of phonological information.  In N. Segalowitz (Ed.).  Language function and brain organization.  NY: Academic Press.

Ojemann, G. & Whitaker, H.  (1978).  The bilingual brain.  Archives of Neurology35, 409-412.

Pettito, L.  (2000).  On the biological foundations of human language.  In K. Emmorey and H. Lane (Eds.).  The signs of language revisited: An anthology in honor of Ursula Bellugi and Edward Klima.  Mahway, NJ: Lawrence Erlbaum.

Pillsbury, W. & Meader, C.  (1928).  The psychology of language.  NY: D. Appleton and Co.

Powers, W.T.  (1973).  Behavior: The control of perception.  Chicago: Aldine.

Powers, W.T.  (1989).  Living control systems.  Gravel Switch, KY: Control Systems Group.

Powers, W. T.  (1991).  CSGnet, November 8, 1991

Runkel, P.  (1990).  Casting nets and testing specimens.  NY: Praeger.

Sapon, S.  (1971).  On defining a response: A crucial problem in the analysis of verbal behavior.  In P. Pimsleur & T. Quinn (Eds.).  The psychology of second language learning  (pp. 75-86).  London: Cambridge University Press.

Selinker, L.  (1972).  Interlanguage.  IRAL10, 209-231.

Snow, C.  (1987).  Relevance of the notion of a critical period to language acquisition.  In M Bornstein (Ed.)  Sensitive periods in development: Interdisciplinary perspectives (pp. 183-209).  Hillsdale, NJ: Erlbaum.

Tucker, A.  (1991).  Decoding ESL.  NY: McGraw Hill.

1.5 Looking Back Over The Next Fifty Years of PCT. By Richard S. Marken

Looking Back Over The Next Fifty Years of PCT
Richard S. Marken
The RAND Corporation

The year 2003 is the 30th anniversary of the publication of William T. Powers’ Behavior: The control of perception (B: CP), the first book to describe the theory of behavior that has come to be known as Perceptual Control Theory or PCT.  It is also, as stated in the request for contributions to this volume, the 50th anniversary of Powers' "initial steps in the research that has led to PCT". I might add that it is also the 25th anniversary of my own involvement with PCT, which began in earnest in 1978. So now seems like a nice time to take stock of the state of PCT.  And we are doing this with this well deserved Festschrift in honor of William T. Powers. I would like to contribute to this Festschrift by looking forward rather than backward.  I have done my share of reminiscing about the past history of PCT, so far as I am familiar with it. I have lamented, in private and in print, the failure of PCT to attract the interest of behavioral scientists over the last 30 years, since the publication of B: CP made the PCT perspective readily accessible to the behavioral science community.  What I would rather do now is look back on the future of PCT by taking an imaginary look at what I think the next 50 years of PCT will have been like.

Looking back over the next fifty years I see that PCT has become the dominant perspective in the behavioral sciences, having replaced behaviorism, cognitive science and evolutionism.  I see this because to see anything else would be foolish. If PCT has not become dominant then this essay, and the Festschrift for which it was composed, will have been completely forgotten.  So what do the behavioral sciences look like now that they are based on PCT?  Perhaps what is most obvious to this visitor from 50 years in the past is the almost complete absence of statistical analysis in behavioral research. Research aimed at testing theories of individual behavior is now based on control models of individuals rather than statistical models of aggregates.  Researchers no longer report statistical significance but real significance, in terms of how well the behavior of the model matches the behavior they have observed. 

Modeling is now the basis of behavioral science research.  Modeling tools are available which make it easy for the researcher to quickly build a model of the behaving system that includes an accurate model of the physical environment in which the system’s behavior is produced.  These modeling tools take advantage of the ever-increasing power of digital technology to produce real-time digital simulations of dynamic interactions between system and environment.  Behavioral research, like physics and chemistry, is now a science based on modeling rather than a guessing game based on statistical significance testing.

Behavioral science is based on modeling because behavioral research methods are now based on testing for controlled variables (Marken, 1997). Behavioral scientists now understand that the apparent randomness of behavior was an illusion created by ignoring the variables that organisms control.  What behavioral scientists had called "responses" are now understood to be actions that protect controlled variables from disturbances. Disturbances correspond to what behavioral scientists had called "stimuli". When many disturbances affect the state of a controlled variable, actions will appear to be randomly related to any one of those disturbances (stimuli). PCT has moved the focus of behavioral science from the randomly-noticed stimulus-response relationships that were the subject of statistical studies of behavior to the consistently controlled perceptions that are now the centerpiece of models of behavior (Marken, 2001).

Research in all areas of behavioral science is now organized around testing for controlled variables. Behavioral scientists no longer ask, “What is the cause of the organism’s behavior?” They now ask, “What perceptual variable(s), if controlled by the organism, would lead me to see the organism behaving in this way?”  This emphasis on testing for controlled perceptual variables has led to a new style of research in which the subjects of behavior studies are allowed to have better control over variables in their environment.  The style of research which was aimed at measuring an organism’s “responses” to the presentation of discrete “stimuli” has been replaced by research aimed at measuring an organism’s ability to control perceptual variables that are being influenced by smooth variations in environmental variables that are disturbances to these variables.  

Ingenious new experimental techniques have been developed that allow researchers to observe the state of hypothetical controlled variables while the variables are being disturbed. These techniques are similar to those developed long ago in the study of the perceptions controlled by baseball outfielders when they catch a fly ball.  For example, McBeath, et al (1995) used a video camera attached to a fielder’s shoulder to observe the state of optical variables, such as the optical trajectory and acceleration of the ball, that the fielder might be controlling while catching fly balls.  These early efforts were often limited by the failure of the researchers to record disturbances, such as the actual trajectory of the ball, to these hypothetical controlled variables. But these studies were important precursors to current PCT-based research inasmuch as they focused the attention of researchers on the importance of monitoring the state of possible controlled variables.

Research aimed at the identification of the perceptual variables controlled by humans and other organisms has been going on for several decades and the catalog of controlled variables continues to grow.  Much of the research effort these days is aimed at classifying controlled variables and studying the relationship between systems controlling different types of perceptual variables. Much of this work supports the basic framework of a hierarchy of perceptual control systems that was originally proposed by Powers (1973, 1998).  In particular, the research results are consistent with Powers’ brilliant suggestion, based at the time only on subjective experience, that the hierarchy of control is organized around a limited number of different classes of perceptual variables.  Although these perceptual classes are not precisely the same as those suggested by Powers it is now clear that there are a limited number of different kinds of perceptual variable. The research is also consistent with Powers’ suggestion that lower level classes of perceptual variables are used as the means of controlling higher level classes of perceptual variables.  It is a testament to the scientific depth of Powers’ work that this hierarchical relationship between perceptual classes was suggested well before there was any significant objective data to support it.

Progress in research and modeling has gone hand in hand ever since scientists started looking at behavior through PCT glasses (Marken, 2002).  This is because research and modeling are inextricably interrelated in the PCT approach to behavior.  Progress in research depends on the development of models that explain the research results. Similarly, progress in the development of models of behavior depends on research aimed at testing the predictions of these models.  This tight interrelationship between research and modeling has resulted in the development of models that produce behavior that is remarkably realistic.  Some early models based on PCT (Powers, 1999; Marken 2001) hinted at the kind of realism that could be produced by models based on PCT.  Current models benefit from many years of research into the variables that organisms actually control while carrying out various behaviors.  They also benefit from the realism that can now be achieved in terms of simulation of the physical environment in which behavior actually occurs.

The science of PCT has not only increased our understanding of behavior, it has also contributed to developments in many areas of practical endeavor.  For example, PCT-based models of behavior have paved the way for the development of robots that can perform very complex and dangerous tasks in highly unpredictable, disturbance-prone environments.  PCT models of economic behavior have made it possible for policy experts to design economic policies that preserve the best results of capitalism, in terms of the production of wealth, while eliminating its worst wrongs, such as the maintenance of egregious wealth inequality. World population is stabilizing near zero population growth, poverty has now been largely eliminated and sustainable, prosperous no-growth economies are now a feature of nearly all world societies.  The new economic model has resulted in the development of economic systems that depend more on reuse of existing resources than depletion of natural resources so that environmental pollution has been reduced to very low levels.

PCT has also become part of the popular understanding of "how people work".  This means that people in general now have a better understanding of how to deal with each other on an everyday basis.  In particular, people are better able to deal with the inevitable conflicts that arise between themselves and others.  People now understand conflicts to be the result of conflicting goals rather than conflicting actions. They also understand that the solution to conflict does not lie in pushing harder against it.  When they find themselves in conflict, people are now more apt to look at themselves and ask, "What do I really want?" rather than look at their adversary and ask, "How can I get them to change?"  The prevalence of the PCT has not turned the world into utopia but it has reduced the level of violence in the world considerably since violence is now understood be the cause of rather than the solution to interpersonal (and international) conflict. 

Looking back over the next 50 years I see that perhaps the greatest legacy of PCT is a change in the tone of the conversation regarding the nature of human nature.  The argument between liberals who believed that all human ills were caused by society and conservatives who believed that all human ills were the result of freely made bad choices has become more nuanced. PCT shows that the difference between liberals and conservatives was simply a difference in the part of the control loop at which one focused their attention.  The liberals saw disturbance resistance as evidence of social control of behavior while conservatives saw the existence of a higher level goal as evidence of free choice.  The liberal/conservative argument has largely disappeared with the realization that both points of view were correct.  We can reduce social ills by reducing social disturbances, such as poverty, so that people can control more effectively. But we can also reduce social ills by freely choosing goals, such as moderation and kindness, that reduce conflict by reducing the degree to which we, ourselves, are social disturbances to others.

Thirty years before the beginning of these next 50 years, William T. Powers' introduced an exciting and revolutionary new view of behavior to the scientific establishment of the day. The new view was that behavior is the control of perception. Powers proposed this view at a time when the prevailing view was that behavior is controlled by perception.  Thus, when Powers' introduced his new view of behavior it was rarely understood, often ignored and sometimes angrily rejected. Now the idea that behavior is the control of perception is taken for granted.  This Festschrift is a long overdue celebration of the work and person of William T. Powers', who first presented the perceptual control view of behavior to a skeptical and often hostile audience. 

Marken, R. S. (1997) The dancer and the dance: Methods in the study of living control systems, Psychological Methods, 2 (4), 436-446

Marken, R. S. (2001) Controlled variables: Psychology as the center fielder views it, American Journal of Psychology, 114, 259-281

Marken, R. S. (2002) Looking at behavior through control theory glasses, Review of General Psychology, 6, 260–270

McBeath, M. K., Shaffer, D. M, and Kaiser, M. K. (1995) How baseball outfielders determine where to run to catch fly balls, Science, 268, 569-573.

Powers, W. T. (1973). Behavior: The control of perception. Hawthorne, NY: Aldine DeGruyter.

Powers, W. T. (1998). Making sense of behavior. New Canaan, CT: Benchmark.

Powers, W. T. (1999) A model of kinesthetically and visually controlled arm movement, International Journal of Human-Computer Studies, 50 (6), 463-581

1.6 Late at night contemplating Erving Goffman as a Perceptual Control Theorist. By Dan E. Miller

Late at night contemplating Erving Goffman
as a Perceptual Control Theorist [1]

Dan E. Miller

University of Dayton

Late one winter evening several years ago I was rereading Frame Analysis by Erving Goffman when I realized that his theoretical approach was akin to that of Powers’ Perceptual Control Theory.  Was Goffman a closeted perceptual control theorist?  Were these similarities merely coincidental?  But, as I began to identify his ideas as being based in the principles of Perceptual Control Theory, they made increasing sense to me.  Had Goffman read William Powers’ book, Behavior: The Control of Perception?  Did a connection exist between the two?  And so, this journey began.


Symbolic Interactionists have mostly escaped the obsession with behaviorism shared by many proponents of Perceptual Control Theory.  The reason for this is relatively straight forward; symbolic interactionism is based on a withering critique of behaviorism (Mead 1934).  Indeed, an elementary cybernetic theory has been in existence for over a century in the social sciences. [2]   For example, an early cybernetic explanation is apparent in Dewey’s (1896) critique of the reflex arc in psychology in which he argued that the neural circuitry of the reflex arc had been misinterpreted.  This misinterpretation resulted in the notion of a stimulus that was separate from its response.  In fact, as Dewey noted, the underlying physiology indicated a unity and continuity of coordinated action.  Separate, temporally contingent, sequential events were not consistently identifiable.  Rather, he noted that humans act purposively in order to solve existential problems.  In doing so they must continually adjust their behavior with regard to the desired outcome in order to successfully complete the act.  Dewey’s argument calls for circular (or cybernetic) logic rather than the cause – effect model of behaviorism (Shibutani 1968). 

Mead’s (1934; 1938) extension of Dewey’s ideas served as the basis of a theoretical position that was later called “symbolic interactionism.”  For Mead, mind and self are self-regulating processes arising through our actions in the natural and social world.  Implicit in the development of his ideas is a feedback loop with a comparison process and an objective (a desired future state) – all of which he included in the process of “the act.”  To Mead “the act” was what the organism was doing at the time – purposive behavior with a desired outcome.  The act was all encompassing, from its impulse to culmination.  The act, then, involved both perception and action in a circular process.  That is, one’s actions must be carried out in a way so that the response continually acts back on the organism, who selectively perceives those stimuli that enable the continuation of the act to its culmination (Mead 1938).  For Mead individuals continually adjusted their behavior as a result of an implied comparison between the present circumstances and the desired future state.    Mead’s theory of the social act is an established and universally accepted principle among interactionists.  It is the foundation upon which symbolic interactionism [3] was developed and his ideas form the basis of how interactionists think about individual and social behavior. 

Mead’s ideas flourished in sociology departments and graduate programs and symbolic interaction became a major theory within the discipline of sociology with interactionists serving as critics of the more behavioristic approaches in the discipline.  In other disciplines – notably engineering and information systems – cybernetics and control theory were being developed.  In 1943 Rosenbleuth, Wiener, and Bigelow published their paper, “Behavior, Purpose, and Teleology,” outlining the principles of a cybernetic theory of human behavior.  Wiener’s book, Cybernetics, published in 1948, further expanded the principles of self-regulating systems into the biological and behavioral sciences.  As these ideas developed, and believing a major new approach to science was afoot, a series of influential conferences were held – from 1944 to 1956 – on Circular Causal and Feedback Mechanisms in Biological and Social Systems.   Sponsored by the Josiah Macy, Jr. Foundation these conferences were attended by notables including Norbert Wiener, John von Neumann, Gregory Bateson, Ross Ashby, Don Jackson, and the young sociologist, Erving Goffman. [4]   William Powers’ important work developing the principles of Perceptual Control Theory began a few years later. [5]


The title of this paper may pose some questions to the readers of this volume.  For example, “What does Erving Goffman have to do with Perceptual Control Theory or Bill Powers?” and “Who is Erving Goffman, and why should we care about him?”  Goffman was, perhaps, the most well known and the most influential sociologist of the last half of the 20th Century.  I have established a likely connection between Goffman and Bill Powers – maybe not the persons themselves, but their ideas.  Unlike Powers, Goffman was only incidentally interested in individual behavior.  Rather, like Mead he was interested in social acts constructed by people.  His major concerns involved how the self and others interacted in the process of constructing and regulating the “interaction order”. 

Goffman did not simply describe interaction processes.  Rather, he delved into the inner workings of his actors whom he characterized as purposive, self-regulating agents.  For example, when reading Goffman one becomes aware that his humans are as often guided by what they would avoid than by what they would attain (Schudson 1984).  His actors purposively avoid situations that would engender embarrassment, humiliation, the loss of face, a loss of poise, an interruption in the proceedings, and a failure to complete the act.  In addition, Goffman’s actors help each other through those difficult situations that may end in failure or embarrassment. 

Early in his career Goffman (1959) acknowledged the significance of Mead’s ideas to his own work, but then clearly stated his intention of going beyond Mead’s discussion of individual and social action.  In his early work Goffman (1959; 1963) was clearly employing the ideas of control in his work.  Take, for example, his concept of impression management.  Actors in the process of managing the impressions that others have of self do not/cannot know with any certainty that a specific behavior will be successful.  Rather, they adjust their behavior based on the perceptual input of the responses of others to their initial behaviors.  Impression management is accomplished by controlling perceptual input. 

As it developed, Goffman’s approach was remarkably similar to the principles of Perceptual Control Theory.  It is not always easy to tell who influenced Goffman.  He was stingy giving credit and citing his influences.  It is clear that he wanted to be thought of as an independent and original thinker. [6]   Goffman never named his theory or used the names of other theories for that matter.  He paid no homage either to cybernetics or control theory, but the principles of self-regulating systems are evident in the work (Schweingruber 1994).  Did Goffman have contact with PCT ideas?  Had he read Powers’ Behavior: The Control of Perception?  As a broad reader and an early attendee of the conferences in which cybernetics was developed, he almost certainly did.  In this paper I will attempt to connect the dots and establish an elective affinity. I will compare Goffman’s theoretical concepts and methodological approaches with the principle elements of Perceptual Control Theory, namely control, perception, comparison, error, equifinality, the method of specimens, and the test of controllec variables.


Control and the Comparison Process

The central proposition of PCT holds that individuals act purposively in order to bring about a desired state.  This is accomplished by continually adjusting actions in order to make current perceptions correspond to a desired perceptual state or decrease error between the two perceptons (Powers 1973; 1989; 1992).  More simply, people behave in order to control perceptual input to a desired end.  It is the control process, how purposive action works, that is to be explained and not specific behaviors or behavioral sequences.  In accordance, Goffman had no particular interest in specific behavior for its own sake.  Rather, he was interested in outcomes or consequences of actions and how they were achieved.  For example, in his discussion of civil inattention, Goffman does not dwell on the specific behaviors that identify the situation, but rather on how the participants in a variety of situations are aware of each other’s presence and how each participant adjusts his or her behavior in order to maintain the perception of being civilly inattentive to both self and other.  Similarly, with greetings Goffman (1971) was not so interested in the behaviors that constituted a greeting, but rather with how interactants knew a successful greeting had been accomplished.

In Frame Analysis (1974) Goffman first put forward a complex and integrated theory of individual and social behavior.  In doing so, his work had developed an even more distinct similarity to Perceptual Control Theory (McPhail 2002; Schweingruber 1994; Miller 1993). [7]   As with Mead, his central concern was with “the act,” what Goffman called “guided doings.”  Remember, Goffman was interested in how things get accomplished and with what people are doing by themselves and with others.  For him, what people are “doing” is guided controlled with respect to the desired or expected culmination of the act.  In order to carry it off, “a serial management of consequentiality is sustained, that is, continuous corrective control, becoming most apparent when action is unexpectedly blocked or deflected and special compensatory effort is required (1974, 22).”  Con men do whatever is necessary to reel in a mark without spooking him.  Those marks wary of the proposition placed in front of them require constant corrective control if the act is to have consequence for the cons (Goffman 1952).

Powers focus has been on the control process involved in individual behavior.  Goffman, on the other hand, repeatedly claimed that while he was not interested in analyzing individual behavior, he found it necessary to do so.  “I assume that the proper study of interaction is not the individual and his psychology …  None the less, since it is individual actors who contribute the ultimate materials, it will always be reasonable to ask what general properties they must have if this sort of contribution is to be expected of them.  What minimal model of the actor is needed if we are to wind him up, stick him in amongst his fellows, and have an orderly traffic of behavior emerge (Goffman 1974, 2-3)?”  That minimal model included a verbal approximation of the control process as described by Powers (1973).

One must be able to act toward something in order to affect perceptual input.  When two colleagues greet each other as they pass in the hall, often they will establish eye-contact and nod.  They may, however, exchange “Good morning”(s).  Their intended action – to acknowledge each other’s presence and attention – created a situation allowing the other to return a nod so perceived by self as a reciprocal indication of civility, neighborliness, and that everything is OK.  Such greetings could involve two neighbors waving to each other over a back fence.  The principle of equifinality finds that the end state of a system may be reached from different initial conditions and in different ways.  Action cannot be understood simply by describing behavior; it is primarily the perceptual process of controlling a variable(s) toward some desired outcome. [8]   It was the establishment of neighborliness and civility that constituted the projected outcome.  The exact behavior involved is relevant only to the extent that it worked; it is pragmatically significant.

Perceptual control works only if a mental process comparing how the present situation looks (sounds, feels) with regard to how we want it to look (sound or feel).  Behavior is continually adjusted based the error signal derived from the comparisons in a manner so that the error signal moves toward zero with the perceptual input of a successfully completed act – reaching a desired outcome.  However, people seldom engage in purposive behavior that goes smoothly.  Rather, actions often are disturbed or interrupted.  In order to complete the act, the individuals concerned must take action in order to overcome the disturbance.  For Goffman, when the interaction between two people begins to unravel – for example when a man calls his date by the wrong name – the other may act to protect the offending person’s face by ignoring the gaffe or by making a little joke of it.  The issue isn’t how control is reestablished, but that it is.

Goffman was aware that accomplishing an activity requires continual attention to what’s happening now in relation to the desired situation, and that adjustments often are necessary and that these, too must be monitored in order to see if progress toward the desired state is being made. 

Goffman (1974) called the cognitive organization of situations – frames.  To Goffman raw experience was perceived through frames, thus transforming meaningless raw data into something meaningful and workable in terms of the present situation and the desired outcome.  Frames allow people to locate, perceive, identify, and label experience.  He argued that these frames were essential for successfully identifying and understanding social situations, and for the successful construction of guided doings.  Not only do people learn to read situations – to know “What’s happening here?” – but also to identify the end state of the act (the desired outcome). 

Hierarchical Organization

PCT proposes that individuals (living organisms) are composed of hierarchies of control loops – from autonomic activities like temperature regulation, to non-symbolic purposive action such as walking, to symbolic purposive action such as playing a game of chess or conversing with a friend (McPhail 2002).  These hierarchies, while obviously based on the presence of complex neural networks, are built-up through the individual’s successful participation in the external environment – including, of course, the social and symbolic worlds of daily life.

For Goffman, guided doings are organized on several levels at the same time.  He notes that while playing a game of chess one must manipulate the natural world, overcoming its constraints and obstacles, as one acts in “special worlds” of chess and in the specific interaction encounter – the game and opponent at hand.  Here, Goffman invokes a hierarchical control structure similar to that espoused in PCT – from non-symbolic to symbolic to systems levels.  One must maintain poise and “stay” in the game, observe the eyes and movements of other, carefully move one’s own chess pieces in the context of anticipating the opponent’s response and one’s subsequent move, thus presenting self as a competent and self-assured chess player.  The chess player knows this is happening by monitoring the non-verbal behavior of the other.  Through perceptual input one assesses the degree to which the impressions of the other have been successfully managed, and the player gains insight into the other player’s strategy as it relates to self and the game as it is presently situated.

At the top of Goffman’s hierarchy is the self, a fragile system-level process that is nearly totally dependent on how one perceives others perceiving and acting toward it.  Self is that stage of the hierarchy through which symbolic behavior is orgnized, initiated, and regulated – keeping the situation organized and moving toward desired ends (1959; 1967; 1983).  For Goffman (1956; 1955) people went through life trying to keep from being embarrassed, humiliated, and/or shamed.  In the terminology of PCT, the absence of embarrassment and humiliation are controlled variables at the highest level of the hierarchy.  Embarrassment is “the flustering caused by the perception that a flubbed performance, a working consensus of identities, cannot, or in any event will not be repaired in time (Silver, Sabini, and Parrott 1987)” and the situation falls apart.  A new advertising agent gets flustered and flubs a presentation to a potential client.  The situation ends with the embarrassed agent leaving the board room.  Such embarrassment not only draws attention to the adequacy of self, but also the interaction has broken down with a potentially devastating effect on the relationship between the agency and the potential client. 

Goffman felt that interaction participants were morally obliged to look out for each other – to do one’s part to control the variables that may lead to embarrassment and the breakdown of order.  This “golden rule” of interaction was necessary for survival of the social order and preservation of self.  Therefore, commitment to a working consensus in social situations has moral dimensions.  This commitment is to the interaction order, that fragile and mutable boundary between chaos and survival and to the mutual protection of self.  In the interaction between the advertising agent and client, the client could have helped the agent by initiating a course of action that would put the rookie agent at ease, helping her save face and keeping the sales interaction moving along.    By controlling perceptual input – the poise of the agent and the sense that the social act is progressing toward completion – the interaction order is sustained and selves protected.

Embarrassment stems not from a flubbed performance, but rather from the responses of others as perceived by self.  It is not our behavior that embarrasses us.  Rather, it is our perception of how others see us as the kind of person who flubs his performance and fails to maintain poise in a deteriorating situation.  Goffman suggests participants in interaction have a moral obligation to sustain their own and each other’s claims to relevant identities and that embarrassment emerges if expressive facts threaten or discredit the assumptions a participant has projected about his identity.  That is, we are obliged to help each other overcome disruptions and maintain the interaction order.  We may fumble all over ourselves in our negotiations with potential business partners, but unless the fumbling is perceived or perceived as troublesome, no damage is done.  The negotiations proceed.  Indeed, “face-to-face interaction requires just those capacities that embarrassment may destroy (Heath 1988, 138).”

The Test of Controlled Variables

How do perceptual control theorists know when individuals are controlling a particular perceptual variable?  They have devised a test.  By disturbing/disrupting an individual’s behavior it is possible to witness (observe and measure) “compensatory actions taken by the individual allowing him to continue his original line of action.”  The disruption makes public and observable the act of control as well as the variable(s) being controlled.  Goffman’s approach to the test has been called the method of revelation through disruption (Collins and Makowski 1989).  Clearly interested in the disruptions, flubs, gaffes, and muffs that momentarily stop or disrupt the flow of face-to-face interaction, Goffman focused on how such breaches are repaired – in how perceptions of order are re-established, “by what compensatory methods the interaction was remediated (1974, 3).”

In order to understand the routine competence and taken-for-grantedness of interaction Goffman looked to situations in which interaction was breached. [9]   For Goffman breached interaction was a significant method of data generation “because we [the participants] are gleaning one another’s intentions and purposes from appearances in interaction. We are constantly in a position to facilitate this revealment, to block it, or even to misdirect our viewers (Goffman 1983, 3).”  By studying face-to-face interaction, the disruptions can help the analyst identify which variables are being controlled (or temporarily not controlled).  Also, by creating a disturbance in social situations it is possible to gain critical insight on how organized social activities are constructed and maintained by their participants.  The public nature of the disorganization and subsequent reorganization provides the researcher with data pertinent to the question addressed. 

As with “the test” in PCT, breached interaction may deflect one or more people away from their purposive activity and instead lead to a new course of action or interaction.  For example, panhandlers take on a difficult task.  They must disturb the on-going behavior of a person walking down the street, gain his attention, introduce the topics of money and need, and convince the person to fork over a reasonable amount.  The accomplished panhandler does not rely on a script or a stepwise recipe.  Rather, he will take his cue to pursue a person given the slightest indication of connection – a break in formation, eye-contact, a slowing down – and continually adjust his behavior to the attention and responsiveness of the mark, behaving in a way to maintain a level of interrelatedness that will lead to getting some money.  Here we can witness the successful completion of a social act – and most certainly we can witness unsuccessful social acts that breakdown

Perspectives and Methods

The principle approach in Perceptual Control Theory to the study of human behavior is to construct computer models and demonstrations.  On the other hand, Goffman was a naturalist and a radical empiricist.  He observed from the standpoint of a detached but present and (perhaps) reluctant participant.  He wrote from that perspective, a habit that often obscured the control being exhibited by those he was observing.  Goffman’s practice was to take copious notes of a wide variety of situations, actions, and encounters.  In addition, he gathered and collected written accounts of social situations and the thoughts and perceptions of actors within them.  Goffman did not gather large, representative samples of situations and interactions.  Rather, he gathered diverse specimens (Runkle 1990) of situations and interactions which he, then, described in great detail. 

The specimens Goffman collected were used to illustrate more general processes.  The specimens served as evidence of more pervasive and complex processes.  For example, he described in great detail how people changed their facial expressions as they came into the visual presence of others, and in doing so acted to manage the impressions the immediate others had of self.  He demonstrated how such actions were examples of how people maintained civility and neighborliness.  Even if a neighbor witnessed the change in facial expression of an approaching visitor from a resigned frown to an expectant smile, the neighbor would almost always act as if the warm smile was authentic.  And, if the visitor was aware that the neighbor had witnessed the purposive transformation, yet acted toward the visitor as if his friendliness was authentic, he would continue the ruse unabated in controlled mutual pretense.  It is clear that Goffman’s method of collecting specimens of interactions and social situations served him well.  He was able to sample to his emergent theory.  A representative sampling or randomizing approach would not have been appropriate to the questions he addressed, nor would these approaches to generating data have revealed the richness of detail found in his specimens. [10]

Social Interaction

In Behavior in Public Places Goffman studied how people manage co-presence in social situations varying from unfocused to focused interaction.  He noted that social interaction required its participants to act so that “our actions will render our behavior understandably relevant to what the other can come to perceive is going on (Goffman 1983,51).”  Working consensus is developed and through past successes of working consensus individuals come to anticipate the perceptions of others and their subsequent action. 

Both Mead and Goffman discuss an internalized recognition (memory) of the elementary structure of social acts from the perspectives of those involved who know what is expected of each given the social act.  While socialized individuals have cognitive models of social acts successfully organized, exactly what behavior is called for is not scripted.  Rather sequences are generalized as abstract categories of response.  For example, with greetings, eye-contact followed by reciprocal nods is functionally identical to a “good day” – “good day” sequence.  For each the issue is a successfully accomplished greeting. 

When two or more people come into each other’s sensory space they establish “co-presence,” a situation in which each “must sense that they are close enough to be perceived in whatever they are doing, including their experiencing of others, and close enough to be perceived in this sensing of being perceived (1963,17).”  Often, strangers in co-present situations are aware of the actions of other(s) but remain publicly inattentive to them.  This civil inattention is difficult to manage, calling for attention to detail and action.  One is aware that others are present and of what they are doing.  The person makes it known to the co-present others that he is aware of the situation – that they are in each other’s co-presence – that they are accessible, but not available.  That is, the others are not a direct focus of attention or a direct participant in the on-going line of action.  This situation is managed by the participants through the control of perceptual input of accessibility without availability as acknowledged by each other.

A modestly more focused form of interaction for Goffman, a queue is “… one of the most human, most moral of all social encounters precisely because it has the least external organization and requires the purest commitment to the interactional order for its own sake (Goffman 1983).” 

“For instance, a queue has a particular and characteristic spatial organization.  It has boundaries, and would-be queue members must observe these boundaries, otherwise they may not hold a ‘place’ in it.  If you stand too far to the side or too far back from the next person, questions may arise as to whether you are really in the queue or not.  To engage in queuing, thus, participants must join in sustaining a spatial arrangement and this must be done through a kind of interaction that can in fact be said to be governed by a jointly sustained attentional focus.  Such a focus is rarely formulated as such, …  Nevertheless, it is something to which all members of the queue are attentive, for it is easy to observe how they co-operate with one another to keep it going (Kendon 1988,26).”

In his Presidential Address to the American Sociological Association Goffman (1983) summed up his life’s work by laying out how naturalistic research on face-to-face interaction fit with the larger projects within sociology.  In it he noted how his work fit and how it was to be understood.  In his conclusion Goffman noted that sociology should be useful – that it should be able to help make a better society.  To this end he wrote, “If one must have warrant addressed to social needs, let it be for unsponsored analyses of the social arrangements enjoyed by those with institutional authority – priests, psychiatrists, school teachers, police, generals, government leaders, parents, males, whites, nationals, media operators, and all the other well-placed persons who are in a position to give official imprint to versions of reality (1983, 17).”  Goffman’s ideas along with those of Perceptual Control Theory to which I find a strong affinity are essential for such analyses. 


I do not think it is an accident that those sociologists and social psychologists who were influenced by Goffman found Perceptual Control Theory.  While Goffman and Powers never met, a direct linkage between them may be present.  The parallels are remarkable and Goffman’s work is lovely in light of Powers writings on Perceptual Control Theory.  However, this is not the end of the story.  The  influence of PCT in sociology continues and grows (though slowly) through the work of Clark McPhail, Charles Tucker, Peter Burke, Kent McClelland, David Schweingruber, Bob Hintz, and others. [11]  

The movement will not be an easy one as most social scientists identify a control system as a metaphor and not as actual socio-neurological processes.  As pragmatists, symbolic interactionists will not believe that perceptual control theorists have reached anything close to scientific law.  Instead, the pragmatists will employ PCT as a useful approach to solving practical and theoretical problems.  The approach to theory taken by symbolic interactionists and pragmatists is not accepted sympathetically by Powers and many of his followers.  In fact, the pragmatic inclusion of PCT into other social scientific approaches is a matter of great concern and consternation to the principal agents of PCT who insist on a kind of “purity.” 

We have come a long way since Dewey’s and Mead’s early formulations of a rudimentary perceptual control model.  Powers has advanced the model of the controlled act to its present elegant complexity.  While Powers’ focus was on the purposive action of individuals, Goffman focused on the purposive action of two or more individuals fitting their behaviors together to form concerted action.  They didn’t know each other, but their very useful and powerful ideas mesh nicely, and those ideas are alive in the social sciences, helping us to roll the rock up the mountain.


Burke, Peter and Tsushima, T.  1999.  Levels, agency, and control in parent identity. Social Psychology Quarterly 62:173-189.

Collins, Randall and Michael Makowsky.  1989. The Discovery of Society, 4th Ed. New York: Random House.

Dewey, John.  1896.  The reflex arc concept in psychology. Psychological Review 3:357-370.

Drew, Paul and Anthony Wooten.  1988.  Erving Goffman: Exploring the Interaction Order.  Boston: Northeastern University Press.

Heath, Christian.  1988.  Embarrassment and interactional organization.  In P. Drew and A. Wooten (eds.), Exploring the Interaction Order.  Boston: Northeastern University Press.

Garfinkel, Harold. 1967. Studies in Ethnomethodology. Englewood Cliffs, NJ: Prentice Hall.

Goffman, Erving. 1952.  On cooling the mark out: some aspects of adaptation to failure. Psychiatry 15:451-463.

1955.  On face-work: an analysis of ritual elements in social interaction. Psychiatry 18:213-231.

1956.  Embarrassment and social organization.  American Journal of Sociology 62:264-271.

1959.  The Presentation of Self in Everyday Life. New York: Anchor Books.

1963.  Behavior in Public Places: Notes on the Social Organization of Gatherings. New York: The Free Press.

1967.  Interaction Ritual: Essays on Face-to-Face Behavior. New York: Anchor Books.

1971.  Relations in Public: Microstudies of the Public Order. New York: Harper Colophon.

1974.  Frame Analysis: An Essay on the Organization of Experience.  New York: Harper Colophon.

1983.  The interaction order.  American Sociological Review 48:1-17.

Kendon, Adam.  1988.  Goffman’s approach to face-to-face interaction.  In Drew and Wooten (eds.), Erving Goffman: Exploring the Interaction Order.  Boston: Northeastern University Press.

Manning, Philip. 1992.  Erving Goffman and Modern Sociology.  Stanford, CA: Stanford University Press.

McClelland, Kent. 1994.  Perceptual control and social power.  Sociological Perspectives 37:461-496.

McPhail, Clark.  1994.  The dark side of purpose: individual and collective violence in riots.  The Sociological Quarterly 35:1-32.

McPhail, Clark and Charles W. Tucker.  1990.  Purposive collective action.  American Behavioral Scientist 34:81-94.

Mead, George Herbert.  1938.  The Philosophy of the Act.  Chicago: University of Chicago Press.

1934.  Mind, Self, and Society.  Chicago: University of Chicago Press.

Miller, Dan E., Robert A. Hintz, Jr. and Carl J. Couch.  1975.  The elements and structure of openings.  Sociological Quarterly 16:479-499.

Miller, Dan E. and Robert A. Hintz, Jr.  1997.  The structure of social interaction.  In D. E. Miller, M. A. Katovich, and S. L. Saxton (eds.), Constructing Complexity:  Social Interaction and Social Forms. Greenwich:  JAI Press.

Powers, William T.  1973.  Behavior: The Control of Perception. Chicago: Aldine.

1989.  Living Control Systems: Selected Papers of William T. Powers. Gravel Switch, KY: The Control Systems Group.

1992.  Living Control Systems II: Selected Papers of William T. Powers. Gravel Switch, KY: The Control Systems Group.

Rosenbleuth, A., N. Wiener, and J. Bigelow. 1943. Behavior, purpose, and teleology. Philosophy of Science 10:18-24.

Runkel, Philip J.  1990.  Casting Nets and Testing Specimens: Two Grand Methods of Psychology.  New York: Praeger.

Schweingruber, David.  1994.  Frame Analysis and perception control theory.  Presented             at the annual meeting of the Midwest Sociological Society, St. Louis, MO.

Schudson, Michael.  1984.  Embarrassment and Erving Goffman’s idea of human nature.  Theory and Society 13:633-648.

Shibutani, Tomatsu.  1968.  A cybernetic theory of motivation.  In W. Buckley (ed.), Modern Systems Research for the Behavioral Scientist. Chicago: Aldine.

Silver, M., J. Sabini, and W. G. Parrott.  1987.  Embarrassment: a dramaturgic account Journal for the Theory of Social Behavior 17:47-61.

Wiener, Norbert.  1948.  Cybernetics.  Cambridge: MIT Press.

Dan E. Miller, Ph.D. is a professor of sociology at the University of Dayton.  A past President of the Society for the Study of Symbolic Interaction he has been actively involved in bringing the ideas of perceptual control back into symbolic interactionism.

[1] An early draft of this paper was first read at the Control Systems Group meetings held in Durango, CO in July of 1993.  I would like to thank Clark McPhail and David Schweingruber for their incisive comments on the paper.  A later version of this paper was read at the annual meeting of the Society for the Study of Symbolic Interaction in Chicago, 2002.  Address all inquiries to: Dan Miller, Department of Sociology, University of Dayton, Dayton, OH 45469-1442.

[2] After World War II scientists who had been working with computers, cryptography, and game theory developed a general theory that Norbert Wiener in 1948 named cybernetics – Greek for steersman – a direct reference to the self-regulated, purposive actor.

[3] Symbolic Interaction(ism) was coined by Herbert Blumer in 1937.  Before that the theoretical perspective was known as Social Behaviorism.  We are not sure what Mead called his perspective. 

[4] Kendon (1988) provides evidence that Goffman attended one of the last Macy Conference with Gregory Bateson.  Indeed, Bateson’s influence on Goffman is evident in his work on mental illness and as an observer in mental hospital wards (Goffman 1963; 1967).

[5] It is clear that Powers’ ideas were grounded in cybernetics and engineering control theory and not from the ideas of Dewey and Mead.  Both Mead and Powers’ ideas were sharpened by their critiques of behaviorist principles.

[6] Goffman did not always cite his influences.  Clearly, he read voluminously, but from reading him one may draw the conclusion that he was more influenced by a spy novelist or an etiquette columnist than by a scholar whose work his closely resembles.

[7] “The interaction order” was Goffman’s (1983) posthumously published Presidential Address to the American Sociological Association.  In it he connected his life’s work into a coherent and powerful theoretical statement that was, I believe, present in the 1974 work, Frame Analysis.

[8] Late in 1973 Clark McPhail suggested that my colleague Bob Hintz and I read Powers’ book, Behavior: The Control of Perception.  While McPhail was one of the first sociologists to recognize the importance of Powers’ ideas, he was not alone.  Powers’ book was listed in Erving Goffman’s estate.

[9] The concept of the breach as a methodological device for the study of social interaction was elaborated by Harold Garfinkel (1967).  He defined breaching as an intrusion intended “to produce and sustain bewilderment, consternation, confusion; to produce socially structured effects of anxiety, shame, guilt, and indignation; and to produce disorganized interaction … (to) tell us something about how the structures of everyday activities are ordinarily and routinely produced and maintained (1967,38).”

[10] Goffman’s use of the method of specimens was roundly criticized by some sociological colleagues.  Their major complaints were that his work was disjointed and nonsystematic.

[11] A sampling of the works based on or influenced by Powers’ ideas include McPhail (1994), McPhail and Tucker (1990), Schweingruber (1994), Burke (1999), McClelland (1994), Miller, Hintz, and Couch (1975), and Hintz and Miller (1997).

1.7 What is controlled in speaking and in speech recognition? By Bruce Nevin

What is controlled in speaking and in speech recognition?

By Bruce Nevin


Anyone who comes to PCT from a field of science that touches on behavior is challenged to rethink that field in terms of negative feedback control. At the 2002 CSG Conference, I demonstrated how a linguist interacts with a speaker of an unknown language, and it was generally agreed that I was testing for controlled variables. The process that I demonstrated results eventually in a representation and a description from which anyone suitably trained in the field could reproduce the utterances so represented. However, it does not identify what perceptions a native speaker of the language is controlling.

Let me emphasize the point: A phonemic representation and phonetic description makes it possible for a (suitably trained) person to learn to control certain perceptual variables as a native speaker does, without going so far as to actually identify those variables. Such a description captures the controlled variables without identifying them—or, more precisely, without showing whether it has identified them or not.

In the search for controlled variables, some of the observable features of speech that are used in such a description can be eliminated as candidates. Those descriptors that are not directly perceptible, but are available only to an observer using special equipment, cannot be variables perceived and controlled by the speaker and listener. But that does not mean that these descriptors are irrelevant to that search. Nor, of course, does it mean that by process of elimination the residue comprises only controlled variables.

Human speech has been investigated for a long time. In the 19th and 20th centuries especially, researchers developed methods and tools for accurately measuring aspects of speech production and perception. But, in the words of a sign that used to hang in Einstein’s office, “Not everything that counts can be counted, and not everything that can be counted counts.” This prolegomena to PCT linguistics surveys some of what linguists and phoneticians have identified and measured, and considers possible other perceptual variables that are more difficult for an observer to measure. The discussion of speech sounds concludes with a summary of what appear to be the controlled variables in speaking and in speech recognition.

Features for description of speech

Just what features are best for description of speech, and how they are related and used, is a field that has been worked and reworked for many years, but never quite settled. Linguists and phoneticians describe speech sounds in acoustic terms, and they describe the means for producing speech in terms of the actions of articulatory organs such as the tongue, lips, and larynx. The earliest phonetic descriptions were in articulatory terms.

Lying athwart this dichotomy is the question of whether the difference between two feature values is phonetic or phonemic. Phonetic descriptors are necessarily universal, applicable to all human speech anywhere. The International Phonetic Alphabet, designed in the 1880s and refined in a number of revisions since then, suffices for phonetic description of every human  language. Vocabulary, however, is idiosyncratic for each language. Even words borrowed from one language into another are subjected to reinterpretation, phonetic as well as semantic and syntactic. A fortiori, the phonemic contrasts between utterances, and the detailed means for keeping similar words distinct from one another, must be established for each language.

Nevertheless, stemming from work by the Prague Circle in the 1930s and ‘40s, there has been a long campaign to establish “universal alphabet” of features that are phonemic descriptors for all languages. The initial proposals defined acoustic features, in recognition of the predominantly social nature of language. This turned out to be problematic, and actually unworkable, and more recent formulations have included more and more features that are defined in physiological terms, harking back (largely without acknowledgement) to systems of features developed by Kenneth Pike in the 1940s and by Charles Hockett in the 1950s. [1] The motivation for this has been an anti-empiricist dogma that a child surely cannot learn language simply on the basis of exposure, and that language learning depends crucially upon innate structures ‘hard wired’ in the human genome and in the human brain, which are then presumed to be the locus of these universal features. Here is not the place to address this issue—I have done so elsewhere. Suffice to say here that these neo-phrenologic teachings have not convinced me, and there is more than ample evidence to the contrary from e.g. statistical learning theory.

The relationship between acoustic and articulatory descriptors has to date not been resolved in linguistics. Further on I will return to this issue, and show how PCT can resolve it. We will also return to the question of identifying phonemic contrasts in a given language, which is really a question of identifying the variables that language users control in distinguishing different syllables and words from one another.

Sound spectrum

We will look at the acoustic side of things first. Speech sounds can be displayed as a sound spectrum, with time on the x axis and frequency on the y axis, from 0 to as high as about 8 KHz. The presence of acoustic energy at a particular frequency level is indicated by a darkening of the image. Local energy peaks in the spectrum appear as dark bands in the display. At about 0.4 sec in the display shown in 0 there are local peaks at about 700 Hz, 1100 Hz, and 2800 Hz. Long before the advent of sound spectrograph equipment in the 1940s these concentrations of energy in the sound wave were recognized by phoneticians, such as Alexander Graham Bell, and termed formants because of their constitutive role in speech sounds.

Figure 1.     A wide-band spectrogram of the syllable [ba] (as in “bah bah black sheep”). Bands on the left margin are set by the spectrograph to help with visual estimation of frequencies. [L&B Fig. 5.2 p. 56] [2]

The lowest peak at about 140 Hz corresponds to the fundamental frequency of vibration of the vocal folds in the larynx, the ‘pitch’ of the voice. The pitch varies somewhat even over this short time span. Since each vertical striation corresponds to one pulse of air pressure through the vocal folds, the average fundamental may be more directly measured by counting the number of vertical striations in the display over a given interval of time. There are about 14 in a 0.1 sec. Interval, hence the fundamental is estimated to be about 140 Hz.

What we saw in Figure 1. was a broad-band spectrogram. In a narrow-band spectrogram of the same syllable [ba] the individual harmonics of the fundamental are instead visible (Figure 2 ). These terms have their origin in the way that the spectrograms are made.


Figure 2.     A narrow-band spectrogram of the same sound [ba] analyzed in Figure 1. The arrow labeled X points to the fifth harmonic of the fundamental frequency (pitch). [L&B Fig. 5.3 p. 58]

The spectrogram is the output of a ‘stack’ of filters, each of which lets a specified range of frequencies pass. In the original spectrograph devices a tape loop was played and with each repetition the filter range was moved up a notch. At the same time an electrode was moved upward on the surface of a drum rotating about a vertical axis. Specially prepared paper was fixed to the drum. Intensity of sound was transformed to intensity of an electrical spark, and the portion of paper through which the spark passed was darkened proportionally. If each filter lets a narrow band of frequencies pass, then each harmonic of the fundamental frequency is indicated by a separate striation from left to right when the paper is unwrapped from the drum—the wide-band spectrograph seen in 0 , where the individual harmonics become obscured because their energy is averaged over the entire band. When the filter bandwidth is reduced, the individual pulses of the fundamental frequency become visible (Figure 2).


Figure 3.     The sound spectrogram (A) of the phrase to catch pink salmon is compared here with the simplified representation of the same phrase that was used as input for pattern playback speech synthesis [L&B Fig. 7.2 p. 142, after Cooper et al. 1952 (see L&B for reference)]

Formants and pitch

The frequency ranges of the formants are independent of the ‘pitch’ of the voice, that is, the fundamental frequency of the glottal pulse. One consequence is that the number of harmonics that fall within a formant is a function of pitch.

Recall that the vocal tract is a filter that lets certain harmonics through and damps other harmonics. For a deep voice, there are more harmonics within the band or ‘window’ of a given formant, and for a high-pitched voice there may actually be no energy present in one or another of the formants because no harmonics of the higher fundamental frequency fall within that range. A readily accessible example is your ability to recognize vowels sung at a high pitch in a language that is unfamiliar to you (so that redundancies of speech cannot assist judgment).

Surprisingly, error rates in perception of speech sounds increases only very slightly as the frequency of phonation increases. It seems that in these circumstances, where there is no acoustic energy actually in the formant because it fails to coincide with any harmonic of the fundamental, we perceive a formant as a ‘window’ that would let speech energy pass if there were any present. Perhaps the listener reconstructs the formant in relation to other vowels for which there is energy in the formant. This may also be a value of vibrato in the singing voice, moving harmonics up and down through the formants. Note, however, that a high soprano voice may actually be pitched higher than the first two formants. Thereby hangs an interesting, if somewhat tangential, experiment.

Experiment: Play a recording of a voice pronouncing various syllables with a fundamental frequency higher than 2500 Hz, and see if hearers can reliably recognize the vowels. (The E three octaves higher than the E above Middle C on the piano is at 2637.7 Hz.)

As a first investigation in this direction, Figure 4. shows a spectrogram of the vowel [i] produced at a pitch steadily rising from just under 400 Hz to almost 900 Hz.

Figure 4.     Sound spectrum of the vowel [i] as produced by my daughter, Katrina, beginning at a high pitch (394Hz) and rising to about 880 Hz. According to the formant scale (in red on the left), the pitch (blue line) would be plotted between the two dotted red lines, but instead it is plotted according to the scale in blue on the right, with a base at 75 Hz and a ceiling at 1000 Hz. (This is a composite of two images from the Praat speech analysis program. Note the break in phonation at about 700 Hz.)

The locations of formants for this speaker are shown in 0 . This tells us that the pitch of the production shown in 0 rises above the first formant for [i] less than halfway through, but never approaches the second formant. However, a pitch of 850–900 Hz. could easily rise above the second formant for other vowels, such as [a] and [u].

Figure 5.     Spectrogram of the vowel sequence [iau] for the same speaker low in her pitch range, at about 207 Hz. For the vowel [i], the first formant is at about 575 Hz, and the second is at about 2500 Hz.

There may also be noise in the vocal signal apart from the glottal pulse of ordinary phonation. Thus, despite the high pitch of an infant’s voice, it is possible to identify formants because of its ‘breathy’ quality.

If subjects can reliably recognize vowels where the fundamental frequency is higher than the formants, then they must be using acoustic information other than that which suffices for synthesizing speech. L&B suppose that “[h]uman listeners appear to derive formant frequencies through a procedure that makes use of their unconscious, internalized knowledge of the mechanisms of speech production” and note further:

It is possible to program digital computers to ‘recognize’ vowels using this sort of procedure (Bell et al 1961). The computer program has access to a memory in which the acoustic consequences of various possible vocal tract shapes are stored. The computer systematically generates internal spectra using this memory. The internally generated spectra are then matched against the spectra of the incoming speech signal. The process does not select the individual formant frequencies on the basis of a single local energy maximum, but rather on the match between the total spectrum of the incoming speech signal and the internally generated signal specified by all three formant frequencies. (L&B p. 42).

Formants and body size

The frequency ranges of the formants are not determined by the fundamental frequency or pitch of the voice, but they are at a higher frequency for women and children than they are for men, so that there is a rough correlation with pitch. The frequency ranges of the formants are determined by the length of the vocal tract. As the length of a tube changes, its resonant frequency changes, just as you get a lower or higher pitch by blowing across the necks of bottles of different size, or of bottles with different amounts of water in them to vary the length of the air column.

The locations of the formants within the range for the given vocal tract are determined by the relative lengths of two tubes, the smaller-diameter pharyngeal tube and the larger-diameter oral cavity. You lower the frequency of F2 by moving the narrowest constriction (made with the tongue) toward the back of the oral cavity, and also by protruding and rounding the lips as noted above, lengthening the larger tube Moving the closest constriction by the tongue forward in the mouth raises the frequency of F2. To increase the frequency of F1 you increase the cross-section of the oral cavity by lowering the tongue and jaw, and conversely by raising the tongue and jaw, thereby decreasing the cross-section of the oral cavity, you lower the frequency of F1.

There is interesting evidence in certain speech mannerisms that people control perceptions of their own voices by doing things that vary the length of the vocal tract. As many linguists and phoneticians have noted, young boys wishing to sound more manly often speak with lips protruded, as do women mimicking (perhaps disdainfully) the speech mannerisms of men. Protruding the lips lengthens the vocal tract, and therefore lowers the frequency ranges of the formants. You may also note that speaking with the lips spread minimizes the length of the vocal tract, producing a ‘smiling’ voice quality that is generally perceived as non-threatening. (It seems to me that in both cases there are adjustments at the rear of the oral cavity made with the tongue, pharynx, and larynx that also affect the length of the vocal tract, but I know of no studies showing this.)

Figure 6.     Average values (plotted on logarithmic scales) of formant frequencies for adult male speakers, adult female speakers (triangles), and adolescents (circles). The vowel points are connected to delineate the ‘vowel space’ of each class of speakers, showing how the vowel space for adult males is lower in frequency than the vowel space for adult females, which is lower in turn than that for adolescents. [L&B Fig. 8.9 p. 178, after Nearey (1978), data and averages derived by Peterson & Barney; see L&B for reference.]

In every culture there are ritual phrases of greeting that are highly predictable and therefore practically empty of information. “Hi, how are you?” “Fine, and you?” One function of these customary greetings appears to be to enable each listener to attune to the formant range of the other speaker. Note, for example, that the [U] of hood in women’s speech is very close to the [u] of who’d in the speech of adolescents (and close to that of men on the other side).

Formants and the vowel space

Figure 7. shows how configurations of the vocal apparatus, both in sagittal section and in a plot of cross-sectional area at various distances from the larynx, correlate with the formants. (Formants above F2 are shown in the graphs of transfer functions, but are not labeled.)

Figure 7.     For each of the apical vowels [i], [a], and [u], the mid-sagittal section of the vocal tract, cross-sectional area functions (for transverse cross-sections of a 17 cm vocal tract), and acoustic transfer functions (in which the vertical dimension is the transfer ratio in dB) [L&B Fig. 8.7 p. 176].

The length of the vocal tract and the limits of movement of these articulators thus determine what may be called the ‘vowel space’ for a given speaker. By plotting F1 on the x axis and F2 on the y axis we can identify locations of vowels in a ‘map’ that correlates with the locations of the narrowest constriction by the tongue in the oral cavity. The conventional diagram used by the International Phonetic Association and others is shown in 0 .

0 ).

Figure 9.      Radiographic data derived from an X-ray movie of a speaker producing the English front vowels [i], [I], [e], [E]. The tongue contours were sketched from motion picture frames that were exposed during the most stable part of each vowel’s articulation. The contour of the mandible is also sketched. [L&B Fig. 8.4 p. 168]

Figure 10.     X-ray tracings for another speaker. Note pairwise swapping of vowels from expected positions.

Figure 11.     X-ray tracings for a third speaker.While the jaw positions are as expected (albeit with the two highest vowels virtually indistinguishable), tongue configurations are quite inconsistent with perceived vowel height.

The Mel scale for pitch perception

Figure 12.     The Mel scale, which relates perceived pitch to frequency. [L&B Fig. 7.5 p. 155 (adapted from Gunnar Fant, Speech, sounds and features MIT Press 1973)]

A frequency of 1000 Hz (1 KHz) is a pitch of 1000 Mel, but 500 Mel is 400 Hz, and 2000 Mel is about 3100 Hz.

Figure 13.     Acoustic classification of the vowels of American English in terms of Mel values of first formant and ‘equivalent’ second formant frequencies. [L&B Fig. 8.11 p. 182]

Note how the vowels are distributed fairly evenly along the periphery of the vowel triangle. (Acute and grave are acoustically defined features that have been used to differentiate two classes of ‘peripheral’ vowels.)

Figure 14.     Acoustic classification of the vowels of Swedish in terms of Mel values of first formant and ‘equivalent’ second formant frequencies. [L&B Fig. 8.10 p. 180]

The three apex vowels [i], [a], and [u] are easiest for hearers to recognize (fewest errors by a factor of 10). Most languages have other vowels in addition to these three, but universally every language includes these three vowels. They are maximally distant from one another in the ‘vowel space’ for a given vocal tract. Further, additional vowels in any given language are spaced more or less maximally from one another and from the apex vowels; exceptions, such as the [¬] vs. [O] in the Swedish vowel space, or the [Ã] vs. [] in the American English vowel space, are typically unstable, subject to dialect variation and in process of change. This maximization of separation in the vowel space has the effect of controlling phonemic contrast.

A fly in the ointment

The correlation of formant coordinates with positions on the vowel triangle is very pretty, as far as it goes. There is, unfortunately, a problem on the articulatory side. Speakers do not all control vowels in a way that so nicely conforms to the vowel triangle as that shown in 0 . For some speakers, tongue contours recorded in this way are little differentiated, or even change the sequence from to [i], [I], [e], [E] to [i], [e], [I], [E] (L&B Figures 8.2 and 8.3).

The perception of the relationships among vowels as being higher, lower, etc. may be fundamentally an acoustic perception. However, this does not prove that vowel production is the control of auditory perceptions alone. We will return to this point.


Figure 15.     Three series of CV (consonant-vowel) syllables where C = [b], [d], and [g], represented in synthetic spectrograms using only F1 and F2. When produced in a pattern-playback device these are perceived as the voiced stops before various vowels.  [L&B Fig. 7.3 p. 144]

The perceptible indications of a consonant usually include an interruption (or in some cases attenuation) of the glottal vibration, and they can involve aperiodic noise of sounds like the [s] of salmon in 0 , but these very short transitions in the formants are what in almost all cases distinguish consonants from one another.

Quantal sounds

The formant frequencies can vary continuously, but there are some conditions which favor a more discrete differentiation, such that a certain amount of imprecision in articulation makes little or no difference for recognition. Sounds produced in these regions are termed ‘quantal’ sounds. [3]

The acoustic signature of a consonant is a relatively rapid transition of formants (0 ), in most cases an interruption or attenuation of the energy in the acoustic signal, and in many cases the characteristic aperiodic noise accompanying the release of closure or the turbulence of near-closure for a fricative or sibilant sound. On the face of it, you would expect a continuous gradation of different consonants corresponding to the infinite differentiation of places at which the tongue and lips can locate the narrowest constriction in the oral cavity. However, the claim is made that the languages of the world use at most seven different places of articulation, although no language uses all seven. Stevens (1972) identified ‘quantal’ regions within which small articulatory differences make no acoustic difference. (Ladefoged and Maddieson [4] prefer the term ‘modal’.) At the boundaries of these regions, however, a small shift makes a large change from the acoustic pattern that is characteristic of one quantal region to the acoustic pattern that is characteristic of the adjacent quantal region. Each of these acoustic patterns is characterized by well-defined peaks in the acoustic spectrum. These physiological and acoustic effects provide ‘affordances’ for pronouncing consonants.

Stevens’ theory of the ‘quantal’ properties of speech also refers to the ‘apex’ vowels [i], [a], and [u] as ‘quantal’ vowels. For example, for [a], the cross-sectional area of the oral cavity is about 10 times that of the pharyngeal tract, and the transition of pressure waves from the latter to the former is at about the midpoint of the vocal tract. If the transition point is moved a centimeter or so forward of center, then F1 is generated in the pharyngeal tube and F2 in the slightly shorter oral cavity; if the transition point is moved a centimeter or so back of center, then the sources of F1 and F2 are reversed, but it makes no difference to the acoustic waveform, F1 and F2 are still very close to one another at the same frequencies (Lieberman 173–174). Thus, a considerable lack of precision in articulation makes little or no acoustic difference.

There are other factors favoring the vowels at the apex of the vowel triangle. For the vowel [i], F2 is at its high extreme, near F3, so that the two formants together constitute a peak of energy in the acoustic spectrum; for the vowel [a] F1 is at its high extreme and F2 converges relatively low next to it, forming a spectral peak in the low-mid range; and for the vowel [u] F1 and F2 are both low. Perhaps more importantly, kinesthetic and tactile perceptions are more salient at the apices of the ‘vowel triangle’: you cannot open your mouth wider than for [a], nor thrust your tongue up and forward farther than for [i], nor up and back farther than for [u] without producing air turbulence and the sound of a fricative consonant instead of a vowel. With the vowel [u] there is also a limit to how closely the lips may be approximated without producing a bilabial fricative [B].

As may be seen in 0 , for the vowels [o] and [] the first two formants are also close together. However, with this pair of vowels, relatively slight changes in configuration of the vocal tract affect the acoustic signal and the perceived vowel.

Other acoustic and articulatory variables

In addition to place of articulation there are other variables which speakers of one language or another control to distinguish different utterances from one another. The most common of these are:

  • Fricative: An articulator approximates closure, permitting air to pass with noisy turbulence. Thus corresponding to [b], [p] there are bilabial [B], [Φ] heard e.g. in the English of speakers from the Indian subcontinent. The more familiar [v], [f]  of American and British English are labio-dental sounds.
  • Nasal: The velum may be raised, closing off the nasal cavity from the oral cavity, or it may be lowered, adding the resonance of the nasal cavity to whatever sound is currently being produced. Thus, corresponding to [b] there is [m], corresponding to [d] there is [n], corresponding to [g] there is [N], and so on. Vowels may be nasalized, as in French and Lakota/Dakota.
  • Voice: Vibration of the vocal folds may begin (or resume) at various points in time relative to the occlusion of the oral cavity for the consonant. The most common distinction is that of [p] vs. [b], [t] vs. [d], [k] vs. [g], and so on, but the timing of voice onset varies from one language to another even for these pairs. In English, initial b is actually voiceless, and voice onset for English p is delayed considerably past release of the oral occlusion. Spanish p is much like English  b, with voice onset close to the oral release. It contrasts with fully voiced Spanish b.  Vietnamese has ‘pre-voiced’ oral stop consonants, produced by expanding the pharynx and lowering the larynx to provide space for vibrating air to go before releasing the oral occlusion.

Other combinations with laryngeal activity are fairly common: Sounds that are usually voiced may be devoiced, whispered, accompanied by ‘creaky voice’ (‘laryngealization’) or outright closure of the larynx (‘glotallization’). Other articulatory combinations occur: For laterals like English [l] the oral cavity is closed by the tip of the tongue but open at the sides of the tongue. Labiovelars like [kw] combine occlusion at both the front and back of the oral cavity. An excellent survey, including exceptionally clear sound spectrograms, may be found in Chapter 10, “Multiple articulatory gestures”, in Ladefoged & Maddieson (op. cit.).

Matrix of features

It can be readily seen that the combination of variables such as voicing and nasalization with place of articulation afford a natural cross-classification of speech sounds, e.g.

                    p        t        k
                    b        d        g
                    m        n        

In this array, all the consonants in a column have their oral occlusion at the same place, and all the consonants in a given row are alike with respect to voicing and nasality. The consonants of every language fall into an array of this sort. There is what is called a ‘universal tendency’ to have no gaps in the array, that is, consonant systems universally tend to be symmetrical. This is often taken as prima facie evidence that what people are controlling are not the sound segments (as represented by letters) but the features that the segments in a given row or column have in common.

Gestural ‘score’

By extensive computer modeling and simulation, researchers at Haskins Laboratories at Yale have developed a ‘gestural’ approach, [5] exemplified in 0 for the word “pawn”. 0 shows a ‘gestural score’, a graph of aperture changes for each of five ‘articulator sets’, the velum, tongue tip, tongue body, lips, and glottis. In a ‘gestural score’, each line represents states of an articulator, and the vertical alignment of transitions from one state to another represents the coordination and relative timing of gestures by the several articulators. (Note that a higher position on the y axis generally signifies a more closed aperture for all articulators, except that for the glottis the sense is reversed.)

Figure 16. A representation of a ‘gestural score’ for the word pawn [http://www.haskins.yale.edu/haskins/MISC/RESEARCH/GesturalModel.html]

The labels abbreviate numeric specifications in a ‘gesture dictionary’. For example, in the label {clo alv} for the tongue tip, “{clo} stands for -3.5 mm (negative value indicates compression of the surfaces), and {alv} stands for 56 degrees (where 90 degrees is vertical and would correspond to a midpalatal constriction)” (op. cit.).

Timing of gestures relative to one another is critical. “For example, the tongue tip {clo alv} gesture and the velum {wide} gesture (for nasalization) are phased such that the point indicating 0 degrees—onset of movement—of the tongue tip closure gesture is synchronized with the point indicating 240 degrees—achievement of goal—of the velic gesture.” A side effect of this particular overlap of gestures is the nasalization of the latter portion of the vowel, which is not distinctive in English.

While the Haskins ‘gestural score’ has proved an efficient and informative representation for analysis and synthesis of speech, it does not identify variables that the speaker and listener can perceive, only variables that the phonetician can perceive by means of X-ray photography and so on. The speaker and listener surely do not perceive millimeters of aperture, any more than an airplane pilot perceives the angles of control surfaces. And of course the parameters measured for the gestural score do not correlate with myographic and other measures of muscle effort, which reflect the influence of disturbances: e.g. if the speaker is lying down or hanging upside down efforts change considerably to resist the effect of gravity on the articulators, while achieving the same apertures with them.

Acoustic features may be expected to correlate well with what the listener hears, and with the speaker’s perception of the sounds of her own speech. For this, the Mel scale is evidently more relevant than straightforward frequency measurement. However, listeners cannot see inside the vocal tracts of speakers or of themselves. To be relevant, articulatory descriptors must be mapped to corresponding kinesthetic and tactile perceptions. The relationship between control of auditory perceptions of the sounds of one’s own speaking and control of kinesthetic and tactile perceptions in the oral cavity helps to explain why this has been such a vexed issue.

Relation of auditory and articulatory variables

The relationship between acoustic and articulatory descriptors has to date not been resolved in linguistics. I believe that PCT can resolve this issue.

There is obvious evidence that people control auditory perceptions in speaking. If they notice that they have mispronounced a word, they say it again, usually with exaggerated precision. Evidence for control of tactile and kinesthetic perceptions is less obvious.

In an experiment I recorded myself reading a text with auditory feedback masked. If solely auditory perceptions were controlled, one would expect the accuracy of articulation to break down, much as accuracy in placing objects would deteriorate in the absence of light. That my pronunciation did not change over the course of reading a long text demonstrates that perceptions other than auditory perceptions are being controlled.

Philip Hoole, [6] working with a patient who had no sensation in the oral cavity, strongly disturbed articulatory control (with a bite block), overwhelmed auditory control (with 30 dB white noise), and both together. He found that “in the absence of oral afferent information articulatory compensation was forced to rely on auditory feedback” with consequent delay in ‘compensating for’ the bite block. The assumption is that control of speaking depends upon neural signals from the oral cavity, as well as auditory input. This appears to demonstrate that acoustic input can be used to re-set reference values for whatever non-acoustic input is being controlled, and it is this which accounts for the delay in ‘compensating’ for the bite block.

It appears to me that tactile and perhaps kinesthetic (e.g. tongue and lip configuration) perceptions are controlled as means of controlling speaking. The reference values for these perceptions for a given phoneme (or perhaps for a given syllable) are set by control of auditory perceptions. Vowel articulation affords less tactile feedback than consonants do, because there is less contact of articulators. In Hoole’s experiments, vowels were more disturbed than consonants, of the vowels u was more disturbed than a and i, and of the consonants s was more disturbed than consonants like and l. These results suggest that of the non-auditory perceptions of speech, tactile perception is easiest to control, followed by perception of the configurations of the tongue, lips, and jaw. A prediction follows that could be tested, that susceptibility to disturbance is proportional to the degree of aperture in the articulation of speech sounds. Anecdotally, I can say that sibilants sS (“sh”) are difficult to control when the tongue tip is numbed with novocaine, and pbfv are somewhat difficult to control (and ‘feel’ awkward) when the lips are numbed.

In addition to vowels and consonants there are ‘suprasegmental’ phonemes of pitch, accent, intonation contour, and the like. These are not merely expressive intonations, but formal requirements for the given language. Like the segmental phonemes, they are learned by the members of a speech community as arbitrary and discrete elements.

Disturbances to control of speaking

To test for controlled variables we require gentle disturbances that can be controlled and measured by an observer. The oral cavity and speech organs are rather well protected from such disturbances.  ‘Coarticulation effects’—where control actions for adjacent phonemes interfere with one another—are indeed disturbances, but they are not controlled by the observer and can be measured only with respect to ‘ideal’ control actions for the ‘same phonemes’ without such interference (because of the mutual disturbance). If the complexity of the problem is not immediately apparent, then as a starting point consider (in 0 ) the diversity of the formant transients for the ‘same’ consonant, depending on the adjacent vowel.

Ladefoged observed that muscle efforts must vary to produce the same articulatory and acoustic result when the speaker is hanging upside down. (This obviously also applies to any change of orientation and to inertial forces under acceleration.) However, to my knowledge there have been no studies calculating the effect of gravity on the tongue, jaw, and other articulators.

On the other hand there is a substantial literature involving disturbances that overwhelm control in one or another aspect.

Many studies, such as that of Hoole cited earlier, have been done with a dentist’s bite block preventing closure of the jaw past a certain point. This changes the position from which the muscles of the tongue act during speaking. Ordinarily, jaw height changes together with tongue height. If control of jaw height is overwhelmed with a bite block, then tongue height can be controlled with the muscles of the tongue alone. There are a dozen or so muscles in and connecting into the tongue (Gray’s Anatomy 323–327). Hoole also reported masking auditory perception with white noise.

As a kind of exotic extreme, Madelaine Plauché and John Ohala at UC Berkeley investigated the effect of cosmetic body piercing of the lips, tongue, and cheeks. They interviewed ‘piercees’ and the professionals (piercers and physicians) who treat them. Acoustic and impressionistic auditory analyses were made of the speech of a subset of ‘piercees’ and discussed the results in a series of case studies. “Piercings may affect the ability of speakers to make certain articulatory gestures and may alter normal speech aerodynamics but in most cases, speakers find alternative production strategies to make speech sound right, suggesting that speakers aim not only at an articulatory target, but also an acoustic one.” [7] For example, one woman had progressively enlarged a hole in her cheek (held open by an ‘eyelet’), and when the aperture reached 6mm she found that the loss of air pressure interfered with consonants like p. Her ‘alternative strategy’ was a ‘tongue-in-cheek’ maneuver to block the opening with the side of her tongue while producing those consonants. This may have interfered with pronunciation of nearby sounds, as e.g. the t in pat, but the article makes no comment on this.

The Motor Theory of Speech Perception [8] says that what is common to the perception and production of speech is “the intended phonetic gesture of the speaker, represented in the brain as invariant motor commands that call for movements of the articulators” (Liberman and Mattingly, 1985, p.2). In the elaboration of this theory, speech perception is inherently distinct and independent from general-purpose auditory perception. A special phonetic system processes auditory input first, and the general auditory system subsequently processes the residue in the acoustic signal. The phonetic processor can affect the functioning of the general auditory system (by either removing information from the signal or leaving it in), but not vice versa. Chris Darwin has done some work on the perception of a tone as either integrated with a vowel formant (making a change in the perceived quality of the vowel), or as a sound external to speech. [9]

It is not difficult to see how to reframe this in terms of a pandaemonium of control systems. In a typically noisy environment, a given sound might be recognized, for example, as the noise component of s and at the same time be recognized by another control system as the hiss of a drop of water on a hot surface. Each of the two systems sends upward a signal whose import is that the given sound has been recognized, though in one an s has been recognized and in the other a hiss. Which of them ‘wins’ depends upon which of these signals is successfully controlled by other systems. What this amounts to subjectively is fitting the perception consistently into the context of someone speaking, or into the context of water dripping onto the hotplate. [10]

There is some discussion of feedback in speech production. John Houde, whose name came up on CSGnet in connection with real-time disturbance of the speech signal, has funding from the James S. McDonnell Foundation to investigate “the role of auditory feedback in speech production. [11]

MacNeilage, P. (1979) Feedback in speech production: current issues. Paper read to the International Symposium on the Cognitive Representation of Speech, Edinburgh, July 29 to August 1st ‘79

Phonemic and Postural Effects on the Production of Prosody
www.lpl.univ-aix.fr/projects/aix02/ sp2002/pdf/moebius-dogil.pdf

Perkell, J. (1979) Feedback in the control of speech production. Paper read to the International Symposium on the Cognitive Representation of Speech, Edinburgh, July 29 to August 1st ‘79

Perkell, J.S.; Guenther, F.H.; Lane, H.; Matthies, M.L.; Perrier, P.; Vick, J.; Wilhelms-Tricarico, R.; Zandipour, M. 2000. “A theory of speech motor control and supporting data from speakers with normal hearing and with pro-found hearing loss.” Journal of Phonetics, 28(3), 233–272.

Cong. Phonetic Sciences Tallinn, 1987, Volume 4, 16:19 Paper Se 57.1



 archive.ling.ed.ac.uk/documents/disk0/ 00/00/01/52/taal00000152-01/paper.pdf

[PDF]The Control of Token-to-Token Variability: an Experimental and 
 www.zas.gwz-berlin.de/mitarb/homepage/ fuchs/mooshammer_et_al.pdf

Measuring and Analyzing: a paradigm for speech research

 www.icp.inpg.fr/OHLL/lesPubliRapports/SPS5-palato.pdf – Similar pages

Institute of Phonetic Sciences,
fonsg3.let.uva.nl/Proceedings/Proceedings24/ Proc24Pols_corrton.html – 33k – Cached – Similar pages


Mark Tatham, Model Building in Phonetic Theory

Robert F. Port, Is There a Universal Phonetic Space?:  Why Apriori Phonetic Transcription is Not Possible

Louis C.W. Pols, Acquiring and implementing phonetic knowledge


We are so accustomed to our alphabetic writing systems that a segmentation of speech into letter-like phonemes seems both obvious and necessary. Examination of a gestural score like that in 0 calls this into question. There are cases where two vowels are acoustically indiscriminable but consistently different in articulation (e.g. lip rounding [12] or advancement/retraction of the tongue root [13] ). The given articulatory feature extends over all the vowels of a specifiable domain, typically per word, in what is called vowel harmony, and so native speakers are never in doubt which vowel to articulate or which they are hearing. See

Other sorts of evidence

There is various indirect and naturalistic evidence for controlled variables. Two examples are language change through time, and perception of one language by speakers of another.

A classic example of identifying controlled variables from the evidence of cross-linguistic confusions is The Pronunciation of Greek and Latin: The Sounds and Accents by William H. Sturtevant. One source of evidence was misspellings of Greek and Latin words in inscriptions made by foreigners. More recent research was surveyed by Winifred Strange in 1995. [14]

The literature of language change and linguistic reconstruction is too vast to summarize here. Historical and comparative linguistics is one of the great, if currently unsung, achievements of 19th century science. By way of illustration of the efficacy of its methods, suffice to say that words have been reconstructed in Latin which were not attested in any records, but which were later found written in inscriptions discovered by archaeologists.

Other redundancies of language in speech recognition

The pandaemonium process referred to earlier obviously is not limited to placing a given perceptual signal into a phonetic or phonemic context. It is easy to confirm that we perceive the intended word even when part or all of it is missing from the input, reconstructing it on the basis of syntactic, semantic, stylistic, and other dependencies.

[This is an incomplete draft, alas. [15] 

[1] Hockett, Charles F. (1955) A Manual of Phonology, Baltimore: Waverly Press. Pike, Kenneth L. 1943. Phonetics: a critical analysis of phonetic theory and a technic for the practical description of sounds. University of Michigan publications. Language and Literature, 21. Ann Arbor: University of Michigan.

[2] The abbreviation L&B refers to Lieberman, Philip & Sheila Blumstein, Speech physiology, speech perception, and acoustic phonetics, Cambridge Studies in Speech Science & Communication, CUP 1988.

[3] Stevens, Kenneth N. (1972) “Quantal nature of speech”, in E. E. David Jr. and P. B. Denes (eds.), Humand communication: A unified view.

[4] Ladefoged, Peter, & Ian Maddieson, The sounds of the world’s languages, Blackwell, 1996

[5] See the overview at http://www.haskins.yale.edu/haskins/MISC/RESEARCH/GesturalModel.html and the references, especially those by Browman and Goldstein.

[6] “Bite-block speech in the absence of oral sensibility”, Proc. 11th International Congress of Phonetic Sciences Tallinn, 1987, Volume 4, 16:19 Paper Se 57.1 www.phonetik.uni-muenchen.de/~hoole/pdf/bb_tallinn.pdf. Hoole is at the Institut für Phonetik und Sprachliche Kommunikation, Ludwig-Maximilians-Universität München, Germany.  In other research (reported in notes for an oral presentation, “Tongue-jaw trade-offs and naturally occuring perturbation” www.phonetik.uni-muenchen.de/~hoole/pdf/dfgsp_fol1.pdf), Hoole placed sensors at various places on the tongue and elsewhere in the oral cavity and plotted their positions for loud and normal speech with various consonants between pairs of vowels iau. Loud speaking has an effect similar to a bite block, due to wider opening of the jaw to allow greater volume of sound to emerge, but is not controlled or so easily measured by the observer. This document is of particular interest because it presents data for individual speakers, albeit not detailed numeric data.

[7] Madelaine Plauché & John Ohala, Phonetic consequences of body piercing of the speech organs” http://ist-socrates.berkeley.edu/~mcp/Piercing.pdf, emphasis in original.

[8] Liberman, A. M., & Mattingly, I. G. (1985). The motor theory of speech perception revised. Cognition, 21, 1-36.

[9] Darwin, C. J. 1984. “Perceiving vowels in the presence of another sound: Constraints on dormant perception.” Journal of the Acoustical Society of America76, 1636-1647.

[10] Of course the same sound might be recognized as another speech sound (f or even the release of an oral stop like t) or another environmental sound. Several relevant studies are at http://www.sonicpuzzle.com/Shafiro_CV.html.)

[11] An abstract is at http://www.jsmf.org/old_site/programs/mcpewprograms/CNS%20WEB%20ABS/98%20cns%20abs/98%20Houde.htm

[12] Fred Householder, “Vowel overlap in Azerbaijani” in A. Valdman (ed.) Papers … in the memory of Pierre Delattre (Moutin, 1972).

[13] Leon Jakobson, Vowel harmony in Dholuo, unpublished PhD dissertation, UCLA Dept. of Linguistics 1978.

[14] Strange, Winifred. 1995. “Phonetics of second language acquisition: past, present and future”, in Elenius, K. & P. Branderud, (Eds.) Proceedings of the XIIIth International Congress of Phonetic Sciences. Stockholm, Sweden, 13-19 August, 1995. Stockholm: KTH / Stockholm University. Vol. 4. pp. 84-91.

Strange, Winifred. (1995). Cross-language studies of speech perception: A historical review. In Strange, W. (ed.) Speech perception and linguistic experience: Theoretical and methodological issues in cross-language speech research. Timonium, MD: York Press (pp. 3‑45).

[15] Progress from here to Markov-chain dependency of phoneme-successors for sentence, word, and morpheme boundaries, statistical learning for morpheme classes and sentence forms, dependence on dependence for operator grammar and linguistic information, sublanguage for informational precision in closely specified domains. Missing too, still, is a discussion of conventionalization among participants in a public, underwriting all of this.

1.8 How Perceptual Control Theory Began: A Personal History. By Mary Powers

How Perceptual Control Theory Began:

A Personal History
Mary A. Powers

    The beginnings of PCT lie in two major developments of the 1920s and 1930s:  H. S. Black’s concept of negative-feedback control in electronics and Walter B. Cannon’s concept of homeostasis in biology.  These were brought together in the early 1940s, primarily by Norbert Wiener, a mathematician, Julian Bigelow, an engineer, and Arturo Rosenblueth, a co-worker with Cannon.  In 1943, they published the first paper relating engineering control theory to neurophysiology.

    Although Wiener called his 1948 book Cybernetics, or Control and Communication in the Animal and the Machine and believed in the importance of control theory as a key to explaining some phenomena of living systems, he was far more interested in communication engineering and information theory.  This bias was shared by the participants in the 10 Macy Conferences that preceded and followed the 1948 publication of Cybernetics.  Many of the people who attended these conferences were prominent figures in the life, social, and behavioral sciences, mathematics, physics, and philosophy.  Though officially titled “Cybernetics:  Circular Causal and Feedback Mechanisms in Biological and Social Systems,” the meetings were primarily concerned with issues of information and communication.

    Enter Bill Powers, an ex-Navy electronic tech and college physics major in his 20s (hardly the sort of person who got invited to the Macy Conferences) who had then what he has now:  an irresistable urge, when confronted with something unfamiliar but interesting, to grab a pencil and a piece of paper and start figuring it out.  What was interesting to him in Cybernetics was not communication, but rather the idea that the nervous system seemed to be a control system.  He thought this was an enormously exciting idea, and he couldn’t wait to see where the big scientific guns and gurus would carry it.  Because he couldn’t wait, he started figuring it out for himself, but he was sure for many years that someone far more competent than he would be coming along with a more extensive and profound analysis.  That someone, we now know with 20-20 hindsight, turned out to be himself, 20,30, and now 40 years later.

    Why Bill Powers?  My purpose here is to suggest a few of the variety of characteristics and circumstances that made him uniquely the person to develop PCT.  Since he is a private person, I intend to avoid getting too personal, with the one exception that for the rest of this paper I’m going to call him Bill.

    One place to begin is with the satisfaction Bill has always found in figuring out how things work, as mentioned above.  This contributed to a professional career at a technical level, with little aspiration to rise beyond the actual hands-on design and construction of control systems and development of computer software into the heady realms of administration and paperwork.  And his real career lay elsewhereósince working at the lower levels of an organization means, usually, being able to walk out the door at five o’clock and leave it all behind, the evenings and weekends that others might have spent furthering their professional ambitions were free for PCT.

    But sticking to the technical level also meant looking at a lot of emperors and finding them naked.  There is a good deal of difference between talking about control systems metaphorically, philosophically, and theoretically, and dealing with them on a practical basis, when you’re in there soldering wires and making the damned things work.  And Bill made a number of control systems work very nicely indeed.

    While this sort of experience contributed to the solidity of the foundations of PCT, PCT at the same time contributed to Bill’s successful design of control systems:  he would imagine “taking the point of view” of the control system he was designingóif I were this system, what would I be able to perceive, what would I need to perceive, what would “really” be going on?  This worked so well that he was convinced he was cheating, fudging over gaps in his technical expertise by using control theory as he was developing it to explain living systems (of course one person’s cheating is another person’s insight).

    Another circumstance fostering Bill’s approach to living control systems was his coming of age in the era of analog computers.  The digital computer as a metaphor for the workings of the nervous system was immediately more attractive to many than the telephone switchboard it replaced, but in Bill’s eyes, it is false at its base.  His programs, although digital, are designed to simulate the actual analog functions of the brain, not, as in Artificial Intelligence, to produce brain-like results by whatever means.  The contributions his analog models might make to neuroscience have yet to be explored.

    While Bill wanted his model to be plausible and workable from the physiological ground up, his main interest was psychology.  What he knew of psychology when he began was whatever was taught in undergraduate courses around 1950.  Behaviorism held the high ground as far as psychology as a science was concerned.  The therapeutic community was largely Freudian, with a dash of humanistic psychologyóCarl Rogers and Fritz Perls, and later Abraham Maslow.  The treatments available for psychosis were lobotomies and electric shock.  Bill was interested in psychology for personal reasons, as almost everyone is, and like many young engineers and other technically inclined people, he discovered what seemed to be a far more fruitful approach in the pages of what for many of us was our favorite magazine, Astounding Science Fiction.  Many people were drawn to Dianetics because, unlike behaviorism, it didn’t try to do away with the mind; in fact, all the action was in the mind, accessing and dealing with memories, in a very straightforward and routinized manner.  There was an appealing technical flavor to it.  Like others who went into Dianetics, Bill got out when L. Ron Hubbard’s grandiosity, greed, and paranoia turned off youthful enthusiasm, and when it seemed that this “new science of the mind” was not all that it was cracked up to be.

    Soon, the first wave of disillusioned Dianetikers went back to work or school and went on with their lives (I kept running into them at the University of Chicago in the early ’50s, and there are fouróthat I know ofóalive and well in the CSG).  Bill, who had read Cybernetics by that time and thought it to be a much more promising approach than Dianetics had turned out to be, went to work as a medical physicist, and he discovered to his delight that his bosses knew about, used, designed, and could teach him about control systems.  This means he did not approach the subject of living systems from the point of view of a control engineer, but rather as a student of control theory, applying what he was learning to both artificial and living systems at the same time.  This, I think, is the source of his realization that the reference signal, which in artificial systems is set externally to the system and labelled “input,” is, in living systems, internal, and not an externally accessible input at all.

    Together with Bob Clark, another physicist, and later Bob McFarland, a psychologist, the first model of hierarchical living control systems was worked out.  It was published in 1960 as “A General Feedback Theory of Human Behavior,” which presented a six-level hierarchical model.  By this time, Bill had left his job and begun graduate work in psychology at Northwestern University, and the association with Clark and McFarland ended.  The graduate work ended after one year, done in by total incomprehension on the part of the faculty as to what on earth Bill’s master’s thesis proposal was about, by wifely financial panic, and by an appealing job offer from the Northwestern astronomy department.

    “Feedback theory” was the name of the game for many more years, as a book slowly took shape, was dropped into the wastebasket, was written again, and then again.  As this went on, the emphasis shifted from the one immediately obvious component that makes control systems unique, namely feedback, to the overall system of which feedback is a part, and ultimately to that aspect of a living control system that makes it so radically different and so difficult to understand, the control of perception.  The only possible way to know what is happening, or what one is doing, or the effects of either on the other, is by perception.

    How does a person entirely alone develop a science, without money, a lab, or colleagues?  One answer, of course, is that all the equipment was readily at hand.  Between children, a dog, a clunky computer, and above all, himself, there was more than enough to observe and think about.  The nature of much of that observation was unique, however, and involved a form of introspection in which one does not think about thoughts, but about what one is seeing:  What perceptions are necessary to see an object, or movement?  From what perceptions does logic emerge, or principles?  Thus the six levels of 1960 became nine by 1972, and they now number 11.  Bill is the first to admit that the levels he sees are personal, and possibly idiosyncratic, and it is with some dismay that he sees them taken as a final word on the subject, copied down and memorized.  But the main point here is that the levels, and much else about PCT, were derived from experience; the theory had to explain not just the performance of subjects, of others, but how the world looks from the only point of view available to anyone, from the inside.

    The main thing that Bill has been able to bring to his work, then and now, is a mind with no strings attached except his own initial feeling that control theory could answer some of his questions.  It is from that stance that he has read books, taken courses, and otherwise absorbed what was already available in the life sciences.  Learning what other people have done has never meant accepting either their premises or their conclusions.  As an outsider, he has never had to conform to any particular school of thought or please any particular community of scholars.  When confronted with such pressure (as with his master’s thesis), he has simply walked away and continued on his own path.

    I think it took many years for Bill, and for the other people who have become committed to PCT, to fully realize how radically different control theory is from the rest of the behavioral sciences.  There is something about PCT that offends just about every point of view:  behaviorists, cognitive scientists, dynamic systems analysts, roboticists, cyberneticists, and even control engineers seem equally unimpressed, or baffled, or annoyed.  Well-meaning attempts to integrate control theory into the mainstream have succeeded only in confusing the issue with inaccuracies and gratuitous embellishments.  The concept of PCT is expressed as principles which contradict many fundamental assumptions:  that behavior is the end point in a chain of events, that the braincalculates necessary outputs, that the concept of purpose is unnecessary to explain behavior, that reference signals (if they exist at all) can be imposed from outside, that feedback can be given or withheld, that self-regulation is a conscious process only and has nothing to do with homeostasis, and so on, and on.

    In 40 years, Bill and his colleagues have developed a rich and comprehensive theory which encompasses and resolves many issues in the behavioral sciences.  I will never forget the astonishment, joy, and relief on Bill’s face as he looked around at the people gathered at the first CSG meeting in 1985, when he really felt for the first time that control theory was not a lonely and eccentric obsession, but rather a shared enterprise that might, just might, change the behavioral sciences forever.  That hope, unfortunately, is still discouragingly far from being fulfilled, but at least it is clear that PCT no longer exclusively depends on the unique life, talents, and circumstances of a single person.

2. Personal Reflections

2.1 Bill Powers, You are the Man. By David M. Goldstein

Bill Powers, You are the Man.

David M. Goldstein, Ph.D.
A personal note.

Bill Powers is my most important teacher outside of my parents, even though I never had him in a classroom setting. I am dedicating my website to him. See: http://www.dmghelpcenter.com.  This website documents how I apply PCT to Psychological Therapy situations, as well as tells people about me. Hopefully, Bill will not experience too many error signals when he reviews the content of the website. I simply want to join the voices of other people who express appreciation to Bill for all his good work over the years.

Sorry, I couldnít be at the 2003 meeting of the CSG in LA to do this in person.

Bill Powersí PCT is the only theory I know about in Psychology that applies to all levels of living things from the cell to the person to groups of people.  It has kept my interest and allegiance since I learned about it many years ago, despite many disturbances.

I was first exposed to William T. Powers in graduate school at the University of Connecticut, 1970-1974, by Dr. Michael Turvey, who presented it as part of his course on Perception.  None of us in the class, which included some bright people, understood what the heck he was talking about, including Dr. Turvey.  One of my other teachers, Dr. David Zeaman, described Bill Powers as a ìsnake oil salesman.î I think that he was responding to Billís openness to views that Dr. Zeaman considered to be mutually incompatible.  You can see how well I listened to them and how well PCT was received in academic circles at the time when I first learned about it.

I was later lured back into learning more about PCT from my graduate students at Stephen F. Austin State University, who were studying Piaget with me and Powers with Dr. Tom Bourbon.  There are some interesting comparisons one can make between Piaget and Powers.  The Piagetian ìschemeî, which most people find hard to understand, relates to the PCT control system.  The perceptual levels can be related to stages of intelligence.  Bill has a chapter where he spells out the developmental aspects of PCT.  Dr. Franz Plooiz has applied PCT to the development of chimpanzees in the wild and human infants.

Bill came and visited Tom and me and gave a talk to our students in Nacogdoches TX.  He described Nacogdoches as ìThe backwaters of the world.î He slept in my house and met my 4-year- old daughter Sara, who is now 27 years old and has just finished her Ph.D. work in Developmental Psychology.  Sara, by the way, served as a volunteer to do the Method of Levels (MOL) with me before an interview for admission to a graduate school; I have a transcript of the discussion that took place.  If it was good enough for Piaget to use his children, it is good enough for me.

I attended several of the CSG meetings, one in Philadelphia, PA, several in Haimowoods Wisconsin, one at the University of Indiana in PA. and one in St. Louis, Missouri.  Out of one of the Haimowoods meetings, Dr. Dick Robertson and I evolved the research to study the self-image from a PCT perspective, which was finally published thanks to Dr. Martin Taylor.  I recently learned that Dr. Brian Thalmyer did a doctoral dissertation which was based on this PCT approach to self-image work.

At one of these Haimowoods meetings, I did a Q Methodology study of the way people at the conference perceived Bill.  [At some point, I will write up this study.  It would be interesting to redo the study today and compare the findings using the same items.  Summarize it.]

When I moved to New Jersey, I communicated with Bill via email about a self-analysis I was performing using MOL and Q Methodology.  This led to the concept of an Observer who was probably part of the Reorganization System.  He was a great teacher, and therapist.  This gave me the confidence to apply the MOL in a therapy session with a real patient.  I did and the case was presented at the St. Louis CSG Conference and was accepted in the journal of Clinical Case Studies and will be published in 2003 or 2004; Sara is my co-author.

I had the pleasure of visiting Bill and Mary in Durango, Colorado when I was attending a workshop nearby.  The high mountains seem an appropriate place for Bill and Mary to spend their latter years.

I am confident that PCT will be The Psychological Theory of the future.  It would be nice if Bill gets a chance to see a glimmer of this.  He has already had more of an impact on other people than most Psychologists.  He has certainly influenced me and the way I do Psychological Therapy.  Bill Powers, You are the Man! Thank you, thank you, and thank you.


2.2 Reflecting Back On My Father. By Barbara Powers

Reflecting Back On My Father

Barbara Powers

My dad has always been a highly intelligent, very creative man. He started setting computers up at home long before this became commonplace. His repertoire includes designing observatories, writing science fiction stories, and helping to develop closed-circuit television. The list goes on.

For some reason, one thing that stands out in my mind is that we had a “commercial killer” in our house. (Bear in mind this would have been back in the 60’s.) A small block of wood with a button smack in the middle resided on the arm of the living room chair. A wire ran from this across the room to the back of the TV, making some mysterious (to us) connection there. When a commercial came on, “Wham!” Down went a hand on the button, off went the sound! The only pain was having to move that wire all around when we needed to vacuum. We did still have to get up to change the channels (aw, shucks), but no more annoying commercials!

Dad was already nearing the completion of his first book about Control Theory by the time I was old enough to begin to understand his work. We kids (and our friends!) were often his unwitting guinea pigs. Each time we would walk through the front door after a day at school, his form would appear in the doorway, finger beckoning, a grin on his face, a new experiment set up and ready to go on the computer in the back room. Anyone who entered the house was fair game. But little did we know that in those increments of five minutes we each spent regularly in front of that screen, we were contributing valuable data for a much larger project.

In retrospect, I realize we enjoyed the infancy of video games for awhile. Swing the pendulum up, and keep it balanced in an upright position. Try to make your dot move in a circle around the stationary one in the middle, fighting the unseen forces playing havoc with your point of light. Play ping-pong with the computer, never mind that the ball does not seem to behave at all like the one downstairs at our real ping-pong table.

And to this day, I still remember the multiplication lessons. Some of the rooms I used for reference, the objects I associated answers with, still appear briefly in my head. I can recall images of numbers hanging around the attic, or pasted on the teapot in the kitchen. I doubt anyone else in my class, or many others, had experienced math lessons quite like these, and I have to say they were very effective. Learning for me comes easier using a method of association, but I will never know if I was this way all along, or because of my father’s influence. At any rate, it is a technique I use in many ways.

I am very proud of my father’s accomplishments. Most of all, I am glad that he will have the joy of experiencing your respect and appreciation for his hard work. I am certain this surprise recognition you are arranging will be a most wonderful reward for Dad.

Thank you,
Barb Powers

2.3 Letter to William T. Powers. By Philip J. Runkel

Letter to William T. Powers

Eugene, Oregon
July 2003

Brother Bill:

I have thought about what I could put on paper that might be suitable for your festschrift. I thought, for example, of writing an essay on what it means to be called an expert—what an expert can and cannot do for the rest of us. After all, you and I are both experts. Unfortunately, I found myself doubting I could finish writing a decent piece by June.

I think both my last two books—Casting Nets and Testing Specimens” and “People As Living Things—can be interpreted as appreciations of your contributions to science and to good will toward humankind. For your festschrift, I quote below the coda from Chapter 18 of “People As.”

Phil Runkel


Having [read] this far, I think you can take a deep breath, look around in your thoughts, and see the grandeur of perceptual control theory.

I began reading the writings of W. T. Powers and his followers about 1985. As I read and pondered, I found my previous views undergoing wrenching and even frightening changes. I found myself having to disown hundreds, maybe thousands of pages of my writings that I had broadcast to my peers with pride. I found, then, that I could see order among my previous confusions about psychological method. The sword that cut the Gordian knot––that cut through my gallimaufry of methodological embarrassments––was the distinction between counting instances of acts, on the one hand, and making a tangible, working model of individual functioning, on the other. That idea, which in retrospect seems a simple one, was enough to dissipate about 30 years of daily dissatisfaction with textbook methods of psychological research. That simple bifurcation is what I wrote about in my 1990 book.

The idea that permits making tangible, working models is, of course, the negative feedback loop. And that, in turn, requires abandoning the almost universally unquestioned assumption made by most people (including psychologists) of straight-line causation––which, in turn, includes the conceptions of beginning and ending. Displacing that theoretical baggage, the negative feedback loop requires circular causation, with every function in the loop performing as both cause and effect. That, in turn, implies continuous functioning (beginnings and endings are relegated to the convenience of perception at the fifth level). One cannot have it both ways. Living creatures do not loop on Mondays and straight-line on Tuesdays. They do not turn the page with loops while reading the print in linear cause-to-effect episodes. William of Occam would not approve.

Powers did not invent the loop. It existed in a few mechanical devices in antiquity and came to engineering fruition after electrical devices had become common. Some psychologists even wrote, haltingly, about “feedback.” But the manner in which living organisms make use of the feedback loop––or I could say the manner in which the feedback loop enabled living creatures to come into being––that insight was Powers’s alone. That insight by itself should be sufficient to put Powers into the pages of the history books as the founder of the science of psychology. Historians of psychology will, I think, come to name the year 1960 (when the two articles by Powers, Clark, and McFarland appeared in Perceptual and Motor Skills) as the beginning of the modern era. Maybe the historians will call that year the Great Divide. The period before 1960 will be treated much as historians of chemistry treat the period before Lavoisier brought quantification to that science.

Using the negative feedback loop as the building-block of PCT enabled Powers to show how mathematics could be used in psychological theorizing. Powers’s true use of numbers made it possible at last to test theory by the quantitative degree to which the data from any single individual approach the limits of measurement error, as in other sciences.

Even making a science possible was not enough to fill the compass of Powers’s vision. He saw the unity of all aspects of human perception and action. He saw that there was not a sensory psychology over here, a cognitive over there, a personality in this direction, a social in that, and so on, but simply a psychology. He gathered every previous fragment into one grand theoretical structure––the neural hierarchy. The nature of the particular levels is not crucial. What is crucial is the idea of the enabling effect of organization by levels––the enabling of coordination among actions of all kinds. Previously disparate psychologies with disparate theories can now all begin with the same core of theoretical assumptions. Though it will take a long time to invent ways of testing the functioning of the hierarchy at the higher levels, I find it exhilarating to realize that Powers and others have already built models having two or three levels organized in the manner of hierarchical control and that those models actually work.

The neural hierarchy is far more than a listing of nice-sounding categories. The theory itself tells how we can recognize the relatively higher and lower placements of levels. It tells us, too, some of the kinds of difficulties to be anticipated in doing research at the higher levels. That kind of help from early theory is a remarkable achievement.

I have mentioned three momentous insights: (1) that the negative feedback loop is the prerequisite for life, (2) that numbers should be used to show the approximation of model to human individual, and (3) that control grasps more aspects of the environment through its hierarchical structure. For any one of those three momentous insights, I think Powers deserves a bronze statue in the town square. To put all three together in one grand system concept is the kind of thing that happens in a scientific field once in a century or more.

After more than 15 years of reading, conversing, writing, and thinking about PCT almost every day, I still feel the way Lewis and Clark must have felt when they began rowing their boats up the Missouri River. I know the general nature of the territory, I know that much of what I will come upon will be astonishing and baffling, and I know that every mile of the journey will be hard going. As I write this book, most parts of which are simply elaborations of the three simple ideas set out above, I find time and again that I must take an hour or a day to struggle with ways of keeping the words as simple as the idea. The ramifications of those simple ideas are multifarious and subtle. As I begin to describe a complication in the way those ideas work together, I find now and again that I have opened further regions of complexity for which I am wholly unprepared. Then I must take an hour or a day or a week to find my way back to firm footing. I do not feel that I am trudging along a prescribed path. I feel that I am taking every step with caution, but also with awe and exhilaration as I wonder what I might add to my understanding. I am sure, however, that I have only an inkling of the exploratory feelings Powers must have had as, day by day and year by year, he built his theory. He guided his footfalls by experimentation; I have guided mine only with thinking about the steps he took.

Pictures Related to William T. Powers and PCT History

4. Miscellaneous

5. Links

The Control Systems Group http://www.iapct.org

6. List of Contributions, Alphabetically Ordered by Author’s Name

  • Timothy A. Carey PCT —Yeah? So What?!
  • Dag Forssell What kind of science has Bill Powers wrought?
  • David M. Goldstein Bill Powers, You are the Man
  • Perry Good The Spark! A Tribute to William T. Powers
  • Joel B. Judd Language: The Control of Perception
  • Richard S. Marken Looking Back Over The Next Fifty Years of PCT
  • Dan E. Miller Late at night contemplating Erving Goffman as a Perceptual Control Theorist
  • Bruce Nevin What is controlled in speaking and in speech recognition?
  • Barbara Powers Reflecting Back On My Father
  • Mary Powers How Perceptual Control Theory Began: A Personal History
  • Philip J. Runkel Letter to William T. Powers