C:\B\CONFER~1\Bledanal\compatib 4 11 00.wpd

Matjaž Potrč


Paradox of analysis envisions two possibilities. Analysis either succeeds or it doesn't. If it succeeds though, it is trivial. But if it doesn't, it is simply false. So, in neither case is analysis feasible.

This result of the paradox of analysis is in patent discord with success of analysis in our everyday practices and in philosophy. The secret of this success is that those kinds of analysis do not apply high standard criteria which are employed in the formulation of the paradox.

Explanation of this puzzle argues for compatibilism of high grade and of lower grade requirements put on analysis. While high grade criteria come naturally for a detached philosophical approach, they do not work well with everyday practices.

Some explanation of cognitive roots is tempted, leading towards the tendency of embracing higher criteria, with their decisive role in establishing of the paradox.

We should first take a look at what the paradox of analysis claims.

Here is one example. You take a concept that you analyze, such as cat. Now, what you analize is the analysandum, and that by the use of which you analyze it is the analyzans. You are aiming at the following appropriately fulfilled form:

analysandum = analysans.

For the case of the concept cat (analysandum) your task is to provide some appropriate analysans, i.e. some informative explanation of the concept, aiming at something such as

cat = an animal with characteristics XYZ

where XYZ is a slot filled with an appropriate description. What exactly this description amounts to has to be furnished by whoever has the task to provide analysis.

Now there are two possibilities for the outcome of the analysis.

First, there is possibility that analysis succeeds or again there is possibility that it does not succeed. In the case where analysis is successful, there is an exact match of content on the side of the analyzandum and on the side of the analysans. But it is hard to imagine some other kind of match or more exact match in this case as the result

(T) cat = cat.

We dubbed this case (T) because such an exact match produces triviality. In other words, if analysis succeeds, there is an exact match of the side of content proper to analyzadum and of the side of analyzans. And thus analysis is trivial.

Another possibility is that the analysis is unsuccessful. But if the analysis is unsuccessful, we have the following situation

(F) cat = XYZ

where the side of analyzans furnishes some different content as there is on the side of analyzandum. It may be some complicated kind of content or some tricky description, which we simply and for convenience abbreviated by XYZ. The difference of "cat" and "XYZ" clearly indicates that there is a mismatch between contents of analyzandum and of analyzans side. For otherwise we would have the situation (T), where the left and the right hand side of equation are identical. Because this is not the case here, we simply dubbed the situation (F) for falsity. (One may of course consider the value XYZ to be cat, but this is just a limit case, which is left out here, and besides to this it is covered by (T) already. Also, (F) would then turn into (T).)

In other words: if we engage into an analysis of a concept, two possibilities are offered to us. Either the analysis is trivial or it is false: either there is a perfect content match between the sides of analyzandum and of analyzans, or there isn't. In either case analysis will not be achieved. But there are no other possibilities left to us. Thus, analysis is impossible. This is the paradox of the analysis.

We now have seriously to ask the question whether analysis is possible. If such turns out to be the case, then something is wrong with the reasoning promoted by the paradox of analysis (PA). For PA concludes that analysis is not possible.

I will argue that analysis is possible indeed, which is shown by the practice of analysis and also by the success of our use of concepts for explanatory purposes. I will then provide a diagnosis about why PA comes to the conclusion that it does, despite our successful practices in explanation of concepts, namely to the conclusion that analysis is impossible. I will argue that this is due to the too high requirements put on the concept of analysis in PA itself, thus that there is changing of score in the language game, in the standards put on the understanding of and of the use of concepts, a change that proceeds unnoticed here. Finally, I will try to explain the shape of cognition, and particularly the contribution of morphological content, which is responsible for such shift in normative criteria for the use of concepts, according to which an appropriate analysis should occur.

The first task is thus to find out whether analysis is possible. I think that there are several important considerations showing that analysis is possible indeed. Here are some facts that may be given in order to support my conviction.

The first of considerations that I will engage in now is important, because it has to do with the everyday reasoning. It is not difficult to see that analysis is a powerful tool. It is a powerful explanatory tool. Every time we reach for a dictionary to clarify a word the exact meaning of whish escapes us, we are looking for an analysis of the concept in question. And most of the time we easily find synonyms that help us to grasp the desired meaning. Every time we explain some concept to a child we make use of the analysis as well. Just another area that may be given here is teaching: a successful teaching will clarify concepts to students, and gradually by that means it will build a wider understanding of the area being thought.

The mentioned practices of analysis are clearly successful, which is proved in our lives repeatedly and every day. Thus the practices of analysis that we use are neither trivial nor false. Triviality is precluded by a strong push in direction of informativeness. We are inclined to expect and to get some new information, so that we easily feel and are uncomfortable with the requirement of informativeness being breached even in the case of circular definitions. These are definitions where the analysans contains a part of concept of the analysandum, like in the obviously wrong and not informative analysis

(*) cat = a catly creature

We strongly feel that there is something wrong here. And a person who does not know what a cat is will not be able to learn something from (*). Even more then this goes for the straight case of triviality, such as identified in the PA analysis above, namely in the case of (T).

It is clear that the main goal of analysis is its informativeness, but this informativeness in none of the cases cited (dictionary, child, teaching) is guided by the PA requirements (T) or (F). So this is a first hint that PA is not an adequate description of our everyday practices of analysis. It is not guided by (T), because this would go straightly against the requirement of informativeness. To some extent it complies with (F), but again (F) is not understood in an exceptionless manner as it is in PA. Some falsity is thus allowed and even welcome in analysis and we may say that a benign kind of falsity, or viable falsity, in any case should dominate triviality. This is what we get if we take a look at PA from the side of requirements of analysis as performed in our everyday practices, particularly considering the requirement of informativeness. It also seems to be a general truth that informativeness requires some failure of exact matching, and that it even requires equivocation, to some viable extent.

It is also far from just a negligible achievement what the analysis is able to perform as a tool in philosophy. Here it goes under the name of philosophical analysis. Perhaps philosophical analysis is not the only tool available to philosophers. Empirical considerations involving concepts may be of importance here, just to give one example. And many valuable insights may come for the philosophical work through the metaphorical use of concepts, which clearly is not the philosophical analysis of those concepts.

But taken all this into account, we may say that it is difficult to see what other kind of tool would be so important in philosophy as the analysis of concepts. Big works in tradition and our everyday practices of writing, reading and lecturing philosophy are all based on the tools of conceptual analysis. What would any kind of philosophy look like without the analysis of concepts as its powerful tool? Heidegger, in the end, is in the business of clarifying concepts, and many times he is quite successful. He provides insights into some dimension of concepts that other approaches have left out.

The point is again here that all these practices succeed without ever embracing the alternatives (T) or (F) as proposed by the analysis of PA. Whole of our philosophical tradition is neither trivial, nor can it be completely false. Perhaps again this gives us a hint into direction that too high requirements were put on analysis by PA for what may practically count as an appropriate analysis of concepts. Without ever embracing (T) or (F), in its non-benign sense, we do a lot indeed for clarification of concepts.

There seems to be a similarity between mechanisms engendering PA and between those engendering the causal exclusion problem, as presented by Horgan (to be published). The compatibilist treatment of the causal exclusion problem is proposed to serve as a model for identifying mechanisms engendering PA.

What is the causal exclusion problem? Suppose that the mental is supervenient on the physical subvenient basis. While it seems a fair bet to acknowledge causal efficacy of the physical basis on the mental, it is much harder to see causal efficiency of the mental upon the physical to come about. Particularly it might be difficult to embrace this opposite direction of causation if one appropriates such plausible principle as the causal closure of physics. But if there is no possibility of this kind of backward causation, then the status of the mental can just be epiphenomenal. So we end up with exclusion of the causal efficacy of the mental and with acknowledging causal incompatibilism between the areas of the physical and of the mental.

The problem is approached from a compatibilist angle by Horgan (to be published). How is compatibilism possible in the area of causality? The main claim is about changing scores in the language game. There are metaphysical questions and there are questions concerning concepts. In the case of concepts, one uses these as tied to various contextual parameters, either according to the higher requirements, or again according to the less stringent requirements.

There are several interpretations of causality. Most of them suggest incompatibility of closure of physics and of the causal efficacy of the mental. Compatibilist interpretation has it the other way round. It claims that the causal closure of physics and the causal efficacy of the mental are compatible. This can be done - compatibilism argues - by looking at the contextual variation of (semantic) standards according to various contextual parameters.

Impression of incompatibility comes with the fact that one shifts - without noticing it - standards in the language game, by shifting - in principle - standards to the higher grade of requirement, to the more demanding principles.

According to the highest possible standards taken, one supposes that there is no possible compatibility between the causal closure of physics and between the causal efficacy of the mental.

One has used the highest standards appropriate for description of the causal and one transfers these standards to all possible cases, and to all possible levels of descriptions.

But taking lower standards persuades one soon that different levels of descriptions are really compatible. The idea is that you can describe a system at various levels, and that these do not exclude each other at all, that they are thoroughly compatible. But somehow there is a drive not to embrace the lower level. Why?

Morphological content, according to the understanding of dynamical cognition approach in modeling of mind, is opposed to the occurrent content. While the occurrent is whatever explicitly appears in a cognitive system, the morphological is in principle not accessible to it. Dynamical cognition is inspired by connectionist approach in the modeling of mind, particularly by many dimensional virtual landscape which is the product of assigning a dimension in the virtual space to each neuron. The result is an intractably rich area which, for ease of presentation, may be rendered as a mountainous landscape, with peeks and valleys between them. This landscape may now serve as a metaphor for morphological content - being identical to the shape of the landscape - which with its forces pushes occurrent items to appear. Morphological content is thus a dynamical and intractable precondition of the occurrent cognition.

The idea here is that changing parameters in the score of the language game - or perhaps better, in the process of determining which content will appear at the occurrent level - does not happen according to the requirements featuring tractable rules and the kind of cognition that naturally goes with these, namely according to the occurrent cognition.

Changing of parameters is happening according to forces in the background of the occurrent cognition, namely according to the forces at work in the morphological content.

Psychological explanation: the naturalness of shifting scores in the language game to higher standards (but we are much slower in being prepared to lower standards) accords to the somehow natural push for the system to achieve local minima, to settle down at the nearest appropriate incline at the landscape. Local minima are searched any time the system has to automatically accommodate the incoming conceptual (or even subconceptual) information. Acknowledging inclines at the landscape and their pull is quite standard for connectionist inspired models of mind, and such framework is also compatible with interpretations in physics.

Two things seem to be happening at the background morphological landscape. The first is the already mentioned push towards the local minima at the landscape, the need for semantic information to settle down at some point. This is the pointing push, the push towards settlement at some nearest local semantic point. On the other hand we have automatic search going on over larger parts on the cognitive landscape, of the morphological content. These pushes become displayed particularly in cases where there is some natural semantic indecision, i.e. where there is shifting going on between two or more senses - where the system just does not know immediately where exactly to settle. These are cases where the morphological content - all the large underlying cognitive background - comes to the fore, where it becomes visible and occurrent. Typically those will be examples of jokes, of understanding someone despite their heavy foreign accent. This is where morphological content, the background, becomes visible and occurrent, or to express it better, where its effect become occurrent at the semantic level. So it is not really that morphological content itself would become occurrent. Morphological content is just the background landscape, it is in the weights of the cognitive system - to use some connectionist expressions.

The tendency of the system to pointing, to settle down, naturally leads to higher normative requirements. This is explanation of "some reason" indicated by Lewis why cognitive system has the tendency to go into higher normative state, the tendency to sharpen requirements in the language game. This also explains why once the local semantic minima are reached, it is fairly difficult to lower them.

The tendency, on the other hand, that shows itself in cases of semantic indecision (jokes, foreign accent) reveals the existence of morphological cognitive background.

It is now on time to give a diagnosis how it came to such high requirements put on analysis as promoted by the alternatives (T) and (F) as proposed by PA.

The answer will be that scores in the language game have been shifted to a dramatic extent, in respect to the question what is an appropriate task of analysis, without that we would be able to notice this.

There are various uses of concepts such as "trivial" and "false". Sometimes, in exceptional circumstances, under very high standards applied to the term, we mean triviality to be all the way down triviality, such as forthcoming in the example

(T) cat = cat.

But other times, we may accuse some person for triviality if she gives us a long explanation that we just feel does not bring into discussion the novelty we have expected it to bring. In the first kind of cases triviality is really not interesting. But it seems that the second kind of triviality is able to bring us sometimes quite a lot of useful information, despite that in the end we are not completely satisfied with it.

Similarly it goes for falsity. Sometimes, under high standards of evaluation, we take falsity really to mean something completely false, such as

cat = dog.

But more often we are accusing someone of falsity if she has furnished us some information, but we do not think that this information is adequate enough. By falsity we then mean something less stringent than we did in the first case. Fuzzy logic seems to have embraced such idea as it claims that some proposition may be false, say, to 78% and not just being a case of all or nothing. (One difficulty for fuzzy approach is that it is practically impossible to assign the exact quantitative parameters, i.e. exact number of percentage.)

Here is an example of shifting scores in the language game, provided for the case of causal compatibilism. (Horgan, To be published. The first set of seven points is thus a quote.) The case shows how scores for evaluation of some concepts have been shifted to higher requirements without that we would notice it.

A parallel case as was proposed for this analysis of causal incompatibilism may be now applied to the paradox of analysis.

The important move is step 6., i.e. failure to accommodate to the lower standards in the analysis of concepts as effectuated in everyday practices, as compared to the high standards put on analysis under the high formal requirements.

How does such a failure in accommodation come about? In what manner is it supported by the structure of our cognition?

We have seen that taking connectionist network as an approximation to how our cognition works, we must acknowledge local minima into which the semantic information tends to settle. The pull into the direction of local minima is quite strong, and it succeeds automatically. This is understandable, for our cognitive system needs to accommodate fast the semantic information that is at its disposition, coming and streaming to it incessantly from the environment.

It is understandable now that any concept, including concepts of triviality and falsity, but also that of analysis, will tend to get as solid settlement in the local minimum as possible. Thus, it will be then natural that the requirements for triviality or for falsity will be as high as possible, following this inclination of the system to settle in local minima. On the other hand, once such a stance is taken, it will be fairly difficult for the cognitive system to abandon local minima. This is comparable to the overall case which we notice every day (or better: actually we do not notice it in our case, but just for the case of others) how difficult it is to change one's opinions and beliefs, i.e. how difficult and uncommon it is to change the settled local minima. A good philosopher may come to be able to effectuate this at least to some extent.

In the case of PA, the scores for concepts of triviality and falsity are also set as high as possible, quasi automatically. They are then not lowered as the thinking concentrates to all the rest of the cases, to the everyday cases. But they should have been lowered. Because they are not, PA's conclusion ensues.

What should be the lesson of all this, namely first identifying a diagnosis of behavior for concepts in PA and later linking this behavior to the structure of our cognition, notably to the morphological content? The PA relative concepts in this case are triviality and falsity.

The proposal is to go compatibilist on the issues involved in PA. We must find some way to allow both high grade requirements and lower grade everyday requirements for application of concepts to be compatible. Of course, they will be compatible if considered each at their own level. The high requirements should hold for highly formal considerations concerning analysis. The lower requirements, with some elbow room, should be in value for all of the rest of cases, notably for the everyday explanatory practices linked to the analysis.


- Graham, George and Horgan, Terence. (1993). "Southern Fundamentalism and the End of Philosophy." Philosophical issues 5. 219-247.

- Henderson, David and Horgan Terence. (To be published). "What is A Priori, and What Is It Good For?"

- Horgan, Terence. (To be published). "Causal Compatibilism and the Exclusion Problem."

- Horgan, Terence and Timmons, Mark. (1993). "Metaphysical Naturalism, Semantic Normativity, and Meta-Semantic Irrealism." Philosophical issues 4. 180-204.

- Horgan, Terence and Tienson, John. (1996), Connectionism and the Philosophy of Psychology. M.I.T.

- Kim, Jaegwon. (1998). Mind in a Physical World. M.I.T.

- Lewis, David. "Scorekeeping in Language Game." Journal of Philosophical Logic 8, 339-359. Reprinted in Lewis (1983).

- Lewis, David. (1983). Philosophical Papers. Volume 1. Oxford.

- Potrč, Matjaž, ed. (1999). Connectionism and the Philosophy of Psychology. Acta analytica 22, Roell: Dettelbach.

- Potrč, Matjaž (1999). "Morphological Content." in Potrč (1999), 133-149.

- Potrč, Matjaž (To be published). "The Desequilibrated Dominance."