Hi Steven, welcome and thanks for your comments

stevenchayes wrote:As for the idea that multiple stimulus relations can be accounted for by equivalence and thus by Sidman's account -- that idea simply will not work. No one has shown experimentally that it will, and I know of no major figure even making that claim. if I can appeal to authority, here is a quote by Murray Sidman in his 2008 article in Cognitive Studies:

[my approach is] “a limited theory
in that it does not cover other kinds of relations than equivalence, as for example,
relational frame theory attempts to do” (2008, p. 331, emphasis in the original).

Murray has repeatedly and explicitly denied that his approach is an attempt to construct a comprehensive
approach to language and cognition. In that context it's a bit jarring to see the posts that seem to treat equivalence as an adequate approach to many areas in language and cognition, justifying that claim by an appeal to Sidman.

That idea is a dead horse. Empirically, it certainly seems to be. Equivalence is 40 years old and has roots that go back several decade more. If that idea is so sound, where are the data?

Fair point - as I said at the start of the thread, I don't know a whole lot about this area so I'm basically just coming at it from an interested but skeptical perspective, however it seemed to me as if Sidman's theory (although not explicitly developed as such) could potentially account for a similar range of relations as RFT. Though, it appears as if I'm mistaken on that.

stevenchayes wrote:RFT is trying to build a comprehensive contextual behavioral approach to language and cognition.
Although I think the RFT account of equivalence is
doing fine, no one has done studies on metaphor, sense of self, implicit cognition, complex forms of rule-governance,
(etc etc) using just equivalence -- and RFT is deep into such areas and many more (as some of the some on this list have pointed out). Thus it is apples and oranges to look at RFT as if it is a theory of equivalence. It is far more than that and if you want to criticize RFT a more productive angle is to get deeply into a given domain (pick one: sense of self and perspective taking; metaphor and analogy; implicit cognition; etc) and see how well the theory is doing.

You're right, RFT obviously attempts to cover more subject area than just stimulus equivalence alone, but I've only ever read about it in reference to stimulus equivalence so naturally that's all I could comment on. Stimulus equivalence seems to be a fairly important aspect of RFT though, so don't you think it makes sense to look at how RFT accounts for the data we have on stimulus equivalence?

stevenchayes wrote:There are lots of things not yet established -- that is true -- but there are no empirical sours note that I know of. If you want to deal with criticisms there are lots of things to look at and argue about. That is fine, fun, and potentially useful .. but RFT folks have hardly been silent in addressing criticisms. If you want to take these criticisms seriously can't just read the critics. You need then to go back and carefully read the original material and the studies. Very often fair readers will see in just a few hours that critics are selectively quoting; misquoting; inserting philosophical biases; ignoring data; and so on. To this moment not a single criticism has legs that can be turned into studies that I know of. If you disagree, please give me the criticism you think is telling and how it would be tested.

Indeed, that makes sense. I just thought it would be good to take an opposing view to Seeker since he seemed quite supportive of the theory so I decided to provide some views provided by the critics (and then he of course provided the counter-arguments). I haven't read all the papers and I haven't reached a personal decision on where I stand since I don't know a whole lot about the topic, but it's an interesting debate to read up on.

Out of interest, why do you think that RFT has produced such controversy? It's not a ridiculous theory, it has evidence supporting it, and it's certainly plausible enough, but it tends to generate quite a bit of heated debate, so why is this? (Or rather, why do you think this is?).

stevenchayes wrote:I don't expect me saying such things to be convincing in and of themselves -- but the data should be (and if they are not
then honest critics should say what data are missing). That takes work ... but that is just the way it is in science.
Statements like "Well what empirical evidence is there that specifically supports RFT above and beyond stimulus equivalence?" are kind of stunning. This is a vast literature and very little of it has to do with equivalence. "Prove to me its not equivalence" flips science on its head. It's your job to provide the account if you think you can, not my job to disprove your theory you've not put into written or empirical form.

My question was an honest one, I wasn't trying to make a backhanded point or anything. I was interested in reading any papers that Seeker was aware of that discussed the advantages of viewing it from a RFT perspective.

stevenchayes wrote:You want single study to try your equivalence based account on? OK. Look at Dougher, M. J., Hamilton, D., Fink, B., & Harrington, J. (2007). Transformation of the discriminative and eliciting functions of generalized relational stimuli. Journal of the Experimental Analysis of Behavior, 88(2), 179-197. This study examines the transformation of respondent and operant stimulus functions via more-than/less-than direct and derived relations.
If you think you can do an equivalence based account -- have at it.

Here is another

Dymond, S. & Barnes, D. (1995). A transformation of self-discrimination response functions in accordance with the arbitrarily applicable relations of sameness, more-than, and less-than. Journal of the Experimental Analysis of Behavior, 64, 163-184. Erratum, 66, 348.
The first study to show three patterns of derived relational responding in accordance with sameness, more-than, and less-than. Alternative explanations for the transformation test outcomes (e.g., based on equivalence) are considered and found wanting.

Thanks, that looks like what I was looking for

stevenchayes wrote:If you want to access RFT studies easily, the cheap way is to join ACBS (costs a buck) at http://www.contextualpsychology.org and download the articles. Also you can then join a list serve with several hundred RFT interested folks on it ... any criticism etc posted there will generate intelligent and thoughtful responses.

I'll check it out.

stevenchayes wrote:Hope I have not intruded.

Of course not, I always appreciate having my misinformed ramblings destroyed by an expert. Sorry if you took anything I've said personally, it's easy to be dismissive of ideas and research from a position of anonymity, but I didn't mean any disrespect.

Thanks again for all your comments, I hope you are able to find the time to stick around and add more comments to this and other discussions.

my_wan wrote:Very excellent 1st post Steve and to the intrusion

I've played with self organizing mechanistic systems in attempts to improve neural net technologies beyond the limits they now have. Especially with respect to predefined input/output directionality, not the path which the standard approach doesn't specify. Though I think I have some good clues on the basics, more complex AI intelligences will require a far more open ended approach to the self organization of intelligences. My limited knowledge and sense of RFT is that it holds some promising ideals that might be formulated in algorithms. Self organized systems remains a nascent math and science, but I like this field. Biological and psychological sciences are a nice source of raw data.

Your rebuttal to the framing of RFT as an equivalence theory appears reasonably convincing to me. What studies and/or experiments do you think would be the most useful for someone wanting to consider it in terms of algorithm development?

Here´s a neural network model of RFT, but it lacks biological plausibility (I don´t know of any more recent and plausible model of RFT):
Dermot Barnes & Peter Hampson. (1997). Connectionist models of arbitrarily applicable relational responding: A possible role for the hippocampal system. Advances in Psychology. Neural-network models of cognition: biobehavioral foundations. Ed. Donahoe & Dorsel. http://ifile.it/2nf1xm6/BZ5Qp7b9.7z (password: ebooksclub.org)

There´re also some models of stimulus equivalence:
Fiona Lyddy & Dermot Barnes-Holmes. Stimulus Equivalence as a Function of Training Protocol in a Connectionist Network. Journal of Speech and Language Pathology and Applied Behavior Analysis.
Fiona Lyddy & Dermot Barnes-Holmes. (2001). A transfer of sequence function via equivalence in a connectionist network. The Psychological Record.
Okada, Sakagami, & Yamakawa. (2005). Modeling stimulus equivalence with multi layered neural networks. Computational intelligence and bioinspired systems. (Google Books)
Dermot Barnes. (1993). Stimulus equivalence and connectionism: implications for behavior analysis and cognitive science. The Psychological Record.

Mr.Samsa wrote:Of course it hasn't been controlled, it's impossible to control for it... If that's what you need in order to disprove RFT then you're essentially claiming that it is unfalsifiable (since animals cannot do stimulus equivalence, and arguably never will be able to, so we can't control the reinforcement histories of human subjects).

No, I don´t think so. The control of reinforcement history is difficult, but not impossible, and there´s good research about that issue. Also, I don´t think we can be so sure that “nonhuman animals cannot do stimulus equivalence”. See:
Lionello-DeNolf, The Search for Symmetry: 25 Years in Review, http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2685068/

Mr.Samsa wrote:It doesn't have to be negative evidence, it just needs to be evidence that better fits one theory over another. One theory predicts this result, and the other can potentially deal with it by referring to hypothetical unknown reinforcement histories. It's not negative evidence exactly, but it's a black mark against RFT.

It´s neither “a better fitness” nor “a black mark”, because those studies were not comparing different predictions (RFT doesn´t predict that explicit reinforcement is necessary during each experiment). I don´t think that positive evidence for pavlovian conditioning implies negative evidence for RFT. Anyway, I do think that there´s positive evidence for pavlovian conditioning that could be incorporated in RFT. For example, it seems that statistical learning of stimulus regularities is relevant for grammar, and stimulus pairings without reinforcement are relevant for semantics. (If Steven is reading this, I´d like to ask him if RFT researchers are incorporating such non-operant processes in their accounts, or if they are proposing operant accounts of the same evidence).

Mr.Samsa wrote:Those are just specific relations which are a product of the mechanism that RFT proposes, they aren't the subject matter that is under question though. It's the mechanism that the RFT proponents are suggesting controls stimulus equivalence that is important. I can't find how RFT differs from Sidman's theory in this regard, and as such his theory would make the same predictions about the relations you've described above (assuming they have evidence backing them up). Can you link me to some studies that demonstrate the prediction and control of these relations? (Or refer me to which papers you've linked to might be relevant). I'm trying to figure out how exactly they study them.

I guess Steven´s post has answered these issues.

When you have some time, I´d be interested in your thoughts about our “molecular-molar” debate: What´s your evidence for saying that “single reinforcing event is irrelevant and it doesn't matter what behavior it follows” (when there´s neural and behavioral research that supports the opposite claim, i.e. that single reinforcing events can increase some behaviors by triggering synaptic changes in neural networks)?

stevenchayes wrote:Hope I have not intruded.

Steve Hayes

Welcome, and thanks for the references. I´d like to ask you if RFT researchers are incorporating non-operant processes in their accounts of language (for example, it seems that statistical learning of stimulus regularities is relevant for grammar, and stimulus pairings without reinforcement are relevant for semantics), or if they´re proposing operant accounts of that evidence.

seeker wrote:Here´s a neural network model of RFT, but it lacks biological plausibility (I don´t know of any more recent and plausible model of RFT):
Dermot Barnes & Peter Hampson. (1997). Connectionist models of arbitrarily applicable relational responding: A possible role for the hippocampal system. Advances in Psychology. Neural-network models of cognition: biobehavioral foundations. Ed. Donahoe & Dorsel. http://ifile.it/2nf1xm6/BZ5Qp7b9.7z (password: ebooksclub.org)

There´re also some models of stimulus equivalence:
Fiona Lyddy & Dermot Barnes-Holmes. Stimulus Equivalence as a Function of Training Protocol in a Connectionist Network. Journal of Speech and Language Pathology and Applied Behavior Analysis.
Fiona Lyddy & Dermot Barnes-Holmes. (2001). A transfer of sequence function via equivalence in a connectionist network. The Psychological Record.
Okada, Sakagami, & Yamakawa. (2005). Modeling stimulus equivalence with multi layered neural networks. Computational intelligence and bioinspired systems. (Google Books)
Dermot Barnes. (1993). Stimulus equivalence and connectionism: implications for behavior analysis and cognitive science. The Psychological Record.

Nice. I'm still reading the e-book, and haven't got to the references, but the e-book leans into the direction that interest me quiet heavily. Thanks.

An approach I'm attempting is based on adding a logical substrate not contained in the feed-forward or other standard artificial neural net approaches. It's predicated on slime mold intelligence and the induced growth patterns of ganglia in response to use or experience. I can't mimic these growth rules directly in a solid state neural net, so I'm using what amounts to a 2 channel neural design in an attempt to mimic the functionality, at least as I postulate that functionality to work. Thus the learning is built into the structural rules, and there is no separate training and prediction modes. There is not even any predefined distinction between inputs and outputs. Rather the two channels have opposing Hebbian and anti-Hebbian potentials, in which the learning rules are a structural balance in these propensities.

The point is that, in principle, this should allow us to dial up or down the propensity for certain forms of logical abstractions. The anti-Hebbian propensity of the logic (mechanistically the Hebbian channel) is an organizational principle analogous to abstractions like harmony, beauty, knowledge unification, or universal oneness for the fundie types. The predicates of which are formed to avoid cognitive dissonance which then form the foundations of a given world view. This is most easily intuitively seen in the self syncing behavior of metronomes on a movable base, where the elasticity of that base varies with experiences imposed as sensory disturbances within the set of metronomes. Much like the propensity of biological neurons to self sync.

This abstract way of attempting to embed and tune propensities for certain abstract logical constructs is why I'm interested in these theoretical psychological constructs.

my_wan wrote:An approach I'm attempting is based on adding a logical substrate not contained in the feed-forward or other standard artificial neural net approaches. It's predicated on slime mold intelligence and the induced growth patterns of ganglia in response to use or experience...

Have you got any text that I can read to see your neural network approach?

seeker wrote:

my_wan wrote:An approach I'm attempting is based on adding a logical substrate not contained in the feed-forward or other standard artificial neural net approaches. It's predicated on slime mold intelligence and the induced growth patterns of ganglia in response to use or experience...

Have you got any text that I can read to see your neural network approach?

Nothing worth publishing. I'm searching for learning functions for it. Basically all it does, in principle at this time, is store experience. It is merely an attempt at some new neural net approaches, making educated guesses about biological functions. It's highly theoretical at present.

Except for training approaches the neural channel is basically a standard neural net. Like the metronomes, they merely self sync in accordance with the elasticity of the base, and take sensory data as excited states of of individual neurons, like pixel data for sight. The connection weights here are merely short term. The presumption is, based on slime mold intelligence and ganglia growth, that this substrate structure stores the long term data. Which looks, at the neural level, much like the short term connection weight variances. I can't mimic biological ganglia growth directly, so to add this subchannel I've added a separate neural subchannel, like a pair of neurons acting in tandem. Now both the subchannel and neural can in principle have separate learning functions, but long term learning is stored in the subchannel, the primary neural connection weights (base elasticity for metronomes). While the short term connection variances are defined by variances in the primary neurons.

The primary neurons, if lacking any distinct connection weights of their own, will simply force the information stored in the subchannel into a compact form, such that any pattern similarities will tend to get stored in the same sunchannel connection sets. The two channels act as a Hebbian and anti-Hebbian channel acting in tandem. Globally, this leads to a propensity to store all data as part of a whole interconnected set.

My problem is in how to define the training function more explicitly. Basically it explicitly makes use of resonance phenomena, like what took down the Tacoma Narrows Bridge. Epilepsy cab even be viewed as a similar type event. Providing an operational meaning to such resonances is not easy, but must be tied to the predictability of upcoming sensory data relative to sensory data. Given a subchannel channel pair [S:C], the future expectation value is defined in [S], where the present experience is defined in [C] at the sensory input point. Other [S:C] are merely connecting neurons, some of which are expected future sensory input points. Hence when an expectation matches reality, the bit state of the channels are [0:0] or [1:1], which reinforces that set of subchannel connections, and destructively interferes when they don't match.

This still presumes some expectation of upcoming sensory data, which like a genetic algorithm can begin from a random state. Ideally the two channels takes the place of separate training and prediction modes in the standard nets. Yet to what degree this correspondence between channels can both train and define the weights between each channel separately is problematic. I'm primarily concerned with subchannel training at this point, with the primary channel merely imposing organizational structure. Higher level reasoning presumably requires manipulating the primary channel, likely with a separate network using the first network as sensory input, in place external sensory input. It likely takes a form we perceive as qualia.

The idea is to derive as many training functions as possible from the dynamics of the system itself, with the minimum of preselected inputs or external error feedback. The patterns inherent in the foundational framework for linguistics and other psychological constructs are important for identifying potential learning functions inherent in mechanistic resonant patterns of neural constructs. Thus my search is presently very open ended, and predicated on a lot of qualitative correspondence with empirical data. Regardless of volume, such qualitative data is highly suspect in specifics. Some incremental improvements in neural nets would be a nice consequence though.

Back to the RFT: I think that it´s a good research program for studying language. Semantics can be understood as the learning of a frame of coordination between words and their referents (objects, behaviors, stimular characteristics, etc.), and the accumulation of relational frames between words (creating "relational nets"). Grammar and generativity can be understood as the sequential recombination of frames, after having learned the semantic coordination frames of the basic components. The informative function of sentences can be understood as a transformation of function of words and referents according to a relational frame. The production and comprehension of rules can be understood as the combination of several relational frames (e.g. coordination, time, cause). Other interesting issues (e.g. so-called "theory of mind", "self", humour, perspective taking, theorization, mentalistic concepts, etc.) can also be studied with this proposal.

seeker wrote:

Mr.Samsa wrote:I agree that one advantage of RFT is that it does explicitly investigate more areas of language and cognition, but I think this is mostly to do with the way they've proposed their arguments where they've phrased the concepts of stimulus equivalence in a more "cognitive" language that allows for more overlap in the fields. I don't think there is anything special about RFT in that sense though apart from terminology - that is, most of that research you're discussing is made possible through stimulus equivalence, and not RFT.

I don´t think so. RFT has studied many relations that are different to stimulus equivalence. I´ve mentioned some of them: “A is opposite to B” (frame of opposition), “A is different from B” (frame of distinction), “A is bigger than B” (frame of comparison), “A is better than B” (frame of comparison), “A is part of B” (hierarchical frame), “A occurs after B” (temporal frame), “A is condition for B” (conditional frame), “A is mine, B is yours” (deictic frame). It´s clear that none of these relations are “stimulus equivalence” (they don´t comply the criteria of simmetry and transitivity). So you have the burden of proof: how would you explain all those relations just through stimulus equivalence?
RFT has its own proposal to explain all those relations (“each relation is learned through multiple exemplar training, and it becomes a relational operant controlled by discriminative stimuli”). This proposal is not reductible to stimulus equivalence.

Ok, I've finally got around to reading that paper on metaphor (direct link to pdf here). I've also found an article-length introduction to RTF which has helped me with the terminology (pdf here, the article starts on p. 61) so I think I'm getting a grip on this. I've found the treatment of metaphor and of its relation to analogy intriguing. In comparison to Skinner's view which is outlined at the beginning, I think this one can account for a wider range of metaphors, and I agree with seeker that its advantage is that it deals with a number of different relations (not just equivalence), incorporating them into a single model. (Obviously, I don't know contemporary rival accounts in psychology).

I'd quite like to have a closer look at this some time and relate it to some work on metaphor that has been done in literary theory. I have some ideas on how this could provide an interesting angle on some issues in literary history that might make for an interesting paper (or alternatively, might prove ridiculous, but I won't know unless I try ). Either way, publishing this here would probably convince all the literary folks at my university that I'd finally gone completely bonkers.

katja z wrote:I'd quite like to have a closer look at this some time and relate it to some work on metaphor that has been done in literary theory.

I´d be interested in knowing your ideas about RFT and metaphors.

seeker wrote:

katja z wrote:I'd quite like to have a closer look at this some time and relate it to some work on metaphor that has been done in literary theory.

I´d be interested in knowing your ideas about RFT and metaphors.

Me too It would require a bit more of my time than I can give it just now, I'd like to read up on it before commenting in more detail, so it won't be tomorrow, but it's picked my curiosity enough that I'll certainly get back to it.

My apology for bumping this thread, but I am currently doing some (background) research about RFT.

A quick glance of the output of various theories that RFT has produced in various studies, I am a bit skeptic to see how RFT brings something new to the whole discussion about behavioral sciences when it deals with acquisition. Think Siegwart Lindenbergs's goal framing theory (GFT) in sociology and prospect theory, as another prime example that is quite elegant, and does seem to relate to RFT in many ways, and perhaps might even build upon or correlate to such existent theories. Maybe someone can enlighten me on the subject of RFT and how exactly was developed? Currently I cannot find any background on which theories RFT builds.

I also wrestle with the following proposition I noticed in RFT:

By means of RFT, it seems to claim that by constructing an instruction such as below, a vague linguistic process acquires a new feature into an existing "frame" and remains in that frame indefinitely, and as a result it will be chained together. (In this example, leading to a prejudice against a certain group of people):



I find that somewhat simplistic. Let for the sake of the argument try this instruction:



Does this instruction mean that the color pink will be acquired for all elephants permanently? I hardly think so. The color pink might hold a temporary relation to the already grounded response to the stimulus "elephant" which is gray. We might correlate certain features that could be acquired into a new frame, but that does not mean that the acquired feature is therefore permanent. I can't see how this sentence can taint the perception of elephants as a whole consistently, unless someone never saw an elephant and assumes it is pink and not "grayish".

The creation of stereotypes seem to involve a lot more background motivations than purely acquired features. Background goals seem not to be taken into account, which in term can influence the outcome of the discrimination. Background motivations might be: group behavior, self-preservation, catharsis and "built-in" constraints (genetics). Thus a cumulated and interactive influence of heterogeneous motives rather than a pure construct of linguistics.

Please correct me if I am wrong.

EDIT TO ADD: I created pictures.

Hello Sasha. I don't think your example constitutes a problem for RFT. You're isolating two sentences from their contexts, and you're isolating two words from each sentence. So you're ignoring both the context of the sentences and the other frames within the sentences, and after doing that, you complain that your analysis was "somewhat simplistic". Of course it was, but ignoring the context and the other frames within the sentences was your decision, and not RFT's proposal.
What happens if you include the context and the other frames? That's the area of research of RFT, and you'll find lots of experimental studies exploring different aspects of this kind of questions. If you want to explore them, you can read this book (http://www.amazon.com/Relational-Frame-Theory-Post-Skinnerian-Cognition/dp/0306466007) or see this page (http://contextualpsychology.org/rft) or participate in this group (http://tech.groups.yahoo.com/group/rft/).

Seeker, lack of contextual background motives was exactly my point and the main reason I brought it up. The first example I gave was taken from a scientific study that used RFT as a tool to understand and predict reactions people give on acts of Terrorism, and the causal response of discrimination mainly due to a simple linguistic construct, which I find too simplistic. I might try to join that group, it would be interesting as I'm really keen to learn more about it.