PARTICULARIST
COMPOSITIONALITY
Matjaž Potrč and Vojko Strahovnik, University of Ljubljana, Slovenia
Particularist compositionality is offered as a realistic possibility, especially if we take a look at problems encountered in the case of ubiquitously present generalist proposals designed to explain compositionality. It is argued that generalism is not even able to support compositionality in the form of ceteris paribus generalities endowed with exceptions. Thus, no form of normative authority of the general is able to explain compositionality, and a good bet is that particularist patterns do in fact accomplish this job.
Particularism is the view developed in the theory of morals. Its main claim is that moral action is not guided by general principles, but by an appropriate adjustment to the rich and holistic non-repeatable circumstances. According to generalism valence of moral features stays the same over a range of cases, which results in a general pattern. Particularism buys just unique relevant patterns proper to single cases. Particularism can be extended to other areas that bear some substantial relation to normativity, of which language furnishes an example. We propose to look at the question whether particularism-inspired compositionality would be a viable option.
Presuppositions
of the classicist view of compositionality, such as proposed by (Fodor,
Pylyshyn 1988) include:
(i) Generalist and atomist view of rationality, with the following basic constituents:
a. atomistic view of parts
b. simple rule-based and tractable compositional structure
(ii) In order to retain rationality one must pay full respect to productivity (because one plausibly anticipates to find it in rational beings, and actually one does find it there), and productivity can only be achieved by systematicity and compositionality constraints.
(iii) From (ii) it follows that (i) and from – (i) it follows that – (ii); so one must buy (i) and (ii).
There is another alternative though that (with some necessary modifications) buys (ii) but rejects (i) and (iii). This possibility may be labeled particularist compositionality or weak compositionality. The inspiration for the alternative option (ii) is already present in the particularist view on the role of general principles and rules in ethics. It may be discerned in Dancy’s work on rationality, practical judgment and meaning (Dancy 1993, 2000, unpublished).
Classicists try to retain requirement of meaning constancy by arguing for appropriateness of general patterns and by diminishing the importance of contextual variation. They try hard to keep in charge atomist normative authority of the unique lexical meaning. This presents one way of following intuitions as one is confronted with cases of metaphorical meaning. Another way to go is to embrace the fact that metaphorical and other meanings are well understood and used by natural language speakers. The intuition will tell you then that atomistic meaning presuppositions have to be rejected. Classicists see structure to be possible just on the basis of atomistic meaning ingredients undergrid by tractable computational rules leading to general patterns. If you abandon this route, they argue, you stay with just a list of disconnected compositional facts involving meaning, which has to happen in the case of connectionist approach. But just how should we preserve rationality and productivity without the presuppositions of classicist compositionality?
There
is the possibility of an alternative non-classicist structure and of
non-classicist language of thought, as based upon the proposal of dynamical
cognition. "It is entirely possible that normative standards are like
competent human cognition, in this respect – i.e., that normative standards are
too complex, too subtle, and too sophisticated to be formulated as
exceptionless general principles." (Horgan and Tienson, 1996:143 /HT/) The
logical space of possibilities needs to be extended if we should account for
the actual rationality, productivity and compositionality as displayed by
cognizers.
Where classicism goes wrong
There are two main
directions where classicists go wrong: 1. the supposition of atomistic meaning
and not being able to explain cases of
metaphors and of double meaning 2. the supposition that if you do not buy
atomistic meaning and rigid rules, neither compositionality nor systematicity
are possible. As against this, we claim that compositionality does not
necessarily need to be tied to generalism. Particularist non-arbitrariness of
composition is possible.
Ad 1. The classicist compositionality argument for productivity has problems with its requirement about keeping the same meaning of the constituent compositional parts over a range of cases.[1] In the sentence “The man kicked the bucket” expressions “kicked” and “bucket” do not appear with their standard lexical meaning, but with what may be dubbed their metaphorical meaning. The particularist view is that even if we try to save atomism by attributing several meanings to the same expression – so meaning of “good” would differ according to its appearance in “good killing”, “good person” – this would not tell us anything about the meaning of “good” in a certain particular case. Our understanding always succeeds in context, which atomists are unable to explain by just breaking things down into their supposed constituents. Once we grasp this, it can serve as rejection of the classicist argument that context is overestimated. Each breaking down of context into parts makes your position only worse, because you cannot explain why in this or another particular case this meaning is used. Classicist compositionality and systematicity are not viable.
2. Presupposition of atomistic and
tractable compositionality and systematicity conditions
Ad
2. “The man kicked the bucket” case is interpreted by classicists in the way
that understanding of this sentence would depend upon the understanding of
other sentences, such as “The bucket kicked the man”, “The woman kicked the
bucket”, “The man kicked the stone”. But none of these sentences will help us
in understanding the original sentence in its metaphorical meaning, because the
meaning of this sentence is formed entirely under the normative pressures of a
specific context.
Whether we understand an expression
literally or metaphorically depends upon the context where the sentence is
used. If it is used in the context of things going on in your yard, the meaning
will be literal. If the context is that of a violent death, the context will be
metaphorical. Then sentences like "The man died of a violent death."
will become important for the understanding/production of the first sentence.
If we look at these two sentences, they are close indeed. But according to the
classicist view, there is nothing meaning-relevant that would tie them
together. Despite that there is no link here according to the classicist, we
can still introduce systematicity. The understanding of the first sentence
depends upon the understanding of the second. This is one way how to grasp
contextual systematicity.
There is a question
though how you should understand atomistic and literal lexical meaning. We can
help ourselves with Dancy’s thought that we do not need to have an unchangeable
kernel of meaning coming with each expression, even if one allows for some kind
of default meaning. Default meaning is not clearly fixed. It is half-standard
meaning, not the most usual meaning. Even if we understand the default meaning,
we do not necessarily need to understand all the other meanings proper to the
same expression. A generalist introduces the basic meaning and allows for soft
aberrations from it, for whatever needs further contextual explaining. Default
meaning is the enabling condition for meaning.
In the case of classicist organic
unity, such as defended by Moore, all expressions retain their meaning. But
once as they appear in a whole, their composed meaning may become different. So
we get the dominance of organic unity upon standard meanings. According to the
particularist this cannot be a sensible approach. For we can ask how a certain
standard meaning can contribute a meaning that it lacks as a part to the
meaning of the expression as a whole. Particularist talks about the
meaning-in-this-case and not about the standard meaning. There is then the
dominance of the context over all expressions in the sentence:
Meaning-in-this-case-1, Meaning-in-this-case-2, …, Meaning-in-this-case-n. Because
context dominates now, we can explain how the metaphorical meaning comes about:
because we uttered it in this context. Moore and the classicists cannot explain
this.
Contrary to classicists, we do not
propose to treat aberrations from lexical meanings as to be adjusted empirical
departures from general pattern of meaning for the item in question. We rather
propose to generalize irregularity in meanings for a certain lexical item so
that one finishes up with a list of items. Each of the items however will not
come without the structure. The holistic structure determining contextual
meaning of the item at the occasion of each of its appearances upon the list is
the structure of the relevant beautiful pattern specific for the case in
question. Lexical meaning is not rejected, for it plays an important role as
the enabling condition in the background structure determining the particular
meaning. The competent users’ knowledge of meanings is thus not the knowledge
of general rules involving classicist systematicity and compositionality. It is
closer to disposition concerning the proper use – a skill that allows us to
change easily form one context to another.
An outline of particularist
compositionality now starts to unleash itself, which is much closer to our
capabilities for rationality and productivity than either classicists or
connectionists would suggest. Accordingly, the prospective particularist
compositionality dynamical cognition inspired proposal is better off at
explaining all of these than this is the case either for classicist or for
connectionist approaches.
The role of generalities becomes that of the background enabling conditions without the normative authority of general patterns. General patterns give way to the morphological content as the holistic background positioning and relevance providing structure. Productivity depends upon particularistically conceived systematicity and compositionality conditions that do not require parts to make the same contribution or the contribution in the range of the same polarity every time they occur.
Particularist normative authority of meaning is intrinsically involved into a certain context, which makes the varying contribution much more natural and plausible. Particularist view about meaning is not condemned to arbitrariness though. There is an underlying non-classicist holistic structure of beautiful patterns. Such a structure brings relevance with it without the prevalent engagement of generalities. Holism should not be viewed as promoting an arbitrarily composed intermingling of parts. The meaning of concepts or sentences depends upon the meaning of other relevant concepts or sentences. We can retain compositionality and show respect to the productivity presupposition (ii) on the basis of particularist non-arbitrariness.
Potrč (2002) has argued in the following direction: even if there exist some principles, these do not necessarily need to be generalist principles. The principle such as metaphysics related Non Arbitrariness Of Composition or NAOC principle may well be compatible with particularism, as long as the relevant structure is preserved. NAOC principle makes it clear how it is possible for particularism to dismiss the objection that it needs to bring arbitrariness along with it. One way to look at this matter is to distribute the job of generalist principles to the area of ultimate ontology, whereas particularist principles and structure would be recognized in the area of regional ontology. A similar possibility to recognize non-arbitrariness of composition should be recognized in the area of compositionality and meaning, we believe. Although in the following exercise in the area of compositionality and meaning we will not dig into the distinctions comparable to these of ultimate and regional ontology. We will rather look at an approach that recognizes some flexibility in handling compositionality, and also recognizes a particularistically based non-arbitrariness of composition structure there, although it is obstructed in its explicit effort to do this because of its uneasy ways of trying to uphold generalist bonds to which it continues to be committed in the sense of the overall project.
Why ceteris paribus approach looses compositionality from its sight and how to help him find it
The approach of Dynamical Cognition in the area of models of mind has brought to attention the possibility of a relevant structure that is not a classicist structure. According to Dynamical Cognition approach there is a non-classical Language Of Thought that goes a long way towards embracing particularist or weak compositionality for which we argue to be the actual compositionality. The Dynamical Cognition approach however is unable to articulate compositionality under that very name. The reasons for this are in that compositionality and the semantics related to it are of particularist nature, that they follow the path of particular patterns, and not that of generalist patterns. Dynamical Cognition approach however, at least in its HT[2] rendering, still follows general patterns, and thereby it follows the normative authority of the general, even if this means embracing generalities with exceptions or ceteris paribus clauses. So ceteris paribus approach looses compositionality from it sight because this approach is generalist, whereas compositionality as the main structural feature founding the meaning is particularist. Such a position of HT is understandable if we consider that his main task is to argue against exceptionless generalities ruling over cognition, opposing to these the kind of generalities that allow for exceptions or ceteris paribus clauses. But once we recognize the possibility of the existence of particularist beautiful and relevant patterns for which we push here, it becomes quite natural for us to offer help to HT so that he would recognize his own position in semantics and compositionality as being that of particularist compositionality.
We will first take a look at where compositionality gets articulated in HT. The answer is that compositionality is articulated in the Fundamental Principle of Cognitive Design (FPCD). FPCD is indeed a rendering of compositionality, and specifically it is a rendering of particularist compositionality. Then we ask why, if this is the case, FPCD is not explicitly recognized as principle instituting compositionality. The answer is that the issue for HT is to argue for ceteris paribus (CP) generalities as against exceptionless generalities. But CP generalities are still generalities, and so we have to recognize that the normative authority of the general leads the HT project. As FPCD offers a particularist pattern covering compositionality and meaning it becomes understandable that it may not be possible to adequately account for it by the help of the generalist CP patterns. Once we see this, it becomes a natural charitable gesture to offer help to HT so that he would recognize his own compositionality proposal for what it is. HT does indeed articulate well productivity and systematicity. These are compatible with generalist interpretation, and systematicity presents a syntactic requirement for productivity. Compositionality that involves semantic contribution to productivity is a much harder issue. It turns out however that it is a much harder issue for an approach based upon general patterns. As semantics does actually follow particularist patterns, and as FPCD presents such a pattern, this will help HT to recognize FPCD dealing with compositionality and meaning.
1. Where does compositionality get
articulated in HT?
Let us first recapitulate some basic moves here. Productivity is one main issue for HT, for productivity is related to the question of the structure enabling intelligent systems and an effective cognition. Productivity in linguistic matters is the capacity to produce an infinite number of well-formed sentences upon the basis of a finite number of input data. The child gets a limited number of input blocks, such as “mama”, “dad”, “car”, “cat” and “mat”. Then at one moment in time he becomes able to produce a potential infinity of well-formed strings or sentences, such as “The cat is on the mat”.
Productivity is a cognitive ability that needs an explanation though. The explanation comes through a structure that enables production. This structure may be assured through syntactic and semantic preconditions. Syntactic precondition figures systematicity, which may be explained in the following manner: It is impossible for an intelligent cognizer to be able to form the sentence “John loves a girl”, but at the same time being unable to form this other sentence, “A girl loves John”. So the ability to master syntactic frames into which to put the items is one precondition for productivity. Another precondition for productivity is compositionality, which basically requires the items in question to retain the same semantic meaning. So the word “cat” should retain the meaning cat through several occasions of its appearance. It does not seem that productivity would be really possible if the word “cat” would mean cat on one occasion, dog on another occasion, and car on the third occasion, in an arbitrary manner. So a certain semantic constancy of basic features that assures structure seems to be needed for productivity and thus for intelligent behavior of cognizer to stay in place.
The above description of enabling conditions for productivity fits well into the generalist approach. Notice that syntactic structure will be typically guided by propositional logic-like setting of a tractable kind. And constancy of meaning of a feature along several contexts is assured by an atomistic generalist account. The presupposition of the independence of units of meaning from the context is important for the ability of systematicity and compositionality to jointly explain productivity in the classicist view. Fodor and Pylyshyn stress this in a repeated manner, and they even argue that contextuality attributed to language is widely overestimated.
HT does not embrace the straight and unlimited generalist account of cognition. At least he argues for some exceptions customary coming along with generalist principles. And he wishes to articulate a picture of cognition that is quite different from the generalist account. So there seems to be a tension in HT’s approach. On the one hand he still embraces a generalist-based story of cognition, even if this is a generalist story allowing for exceptions. On the other hand he wishes to promote a real wide range of approaches to cognition, some of which, such as Dynamical Cognition itself, do really transcend generalism. But the immense novelty of this direction towards recognizing particularist patterns as being in the basis of structure allowing for productivity cannot perhaps be straightforwardly recognized by an approach that still clings to generalism, even to general patterns allowing for exceptions.
HT in his book does well discuss productivity and systematicity. But although he perhaps entertains the feeling that he did discuss compositionality in there, he certainly did not do it in an explicit manner.[3] Despite this we think that HT does discuss compositionality in his book, although in an implicit manner, without mentioning the expression, in a fundamental principle of cognitive design that he proposes. At least this is what we argue.
Compositionality is articulated by the help of the
Fundamental Principle of Cognitive Design (FPCD). FPCD is indeed a rendering of
compositionality.
There is a real issue where compositionality – in opposition to productivity and systematicity – is articulated in the HT book. Productivity and systematicity are found in the register of the HT book. But not compositionality. Given that compositionality targets the structure that enables the meaning, proper for enabling productivity, we will look where in the HT book meaning is discussed.[4] This happens to be the case with the Fundamental Principle of Cognitive Design (FPCD):
“The Fundamental Principle of Cognitive Design The high-dimensional topography of the activation landscape and the positioning of TCS-realizing points on that landscape are jointly just right to subserve content-appropriate cognitive transitions for the whole vast range of potential TCSs the cognitive system has the capacity to instantiate.” (HT, 154)
The search of HT book is for a structure that is non-classicist, but that nevertheless enables intelligent cognition. HT calls his proposal that of Dynamical Cognition (DC). DC distinguishes itself from classicist cognition model of mind proposal, among other things, by the difference at the middle level of cognitive system’s description, according to Marr’s proposal. Whereas we find an algorithm at the middle level of cognitive system’s description in the case of the classicist proposal, there are mathematical-state transitions to be found at this level according to the generic proposal. (HT 45) The generic proposal subsumes the algorithmic one as its sub-case, a sub-case with the unwanted consequence that it is not rich enough if measured with the actual performance to be found in cognizers, and therefore also that it is not realistic.
FPCD describes the generic picture of what happens at the middle level of cognitive system’s description. HT borrows from connectionism inspired picture of cognition.[5] There are not tractable algorithms that would dominate this picture, but a dynamical high-dimensional landscape. Connectionists usually talk about a big number of dimensions, perhaps one dimension for each neuron, this dimension offering a mathematical projection of this neuron’s attached potential activity space. As brain has many neurons and as the resulting potential rich structure attached to it has many more dimensions than we would be able to cognitively capture, we help ourselves by a useful metaphor of a mathematically described multi-dimensional landscape. Inclines and caption points appear upon this landscape, indicating directions where a representation coming to the cognitive system will have a tendency to settle.
Notice that the mentioned representation comes with the content, and that it thus brings the meaning. Notice also that according to this picture, each representation gets positioned upon a rich background multi-dimensional space, where the tendency that it may bring along is shaped by the forces of the background landscape upon which it gets positioned. The resulting meaning is a real context-related and context-infested outcome. Both the tendency of a certain representation and the underlying landscape upon which the representation gets positioned determine the resulting meaning. This meaning is each time at least slightly different and variable because of the impact of the background landscape upon which the representation gets positioned. But the outcome of meaning is not arbitrary, because it is related to the structure that is offered by the constantly molded rich background landscape.
Actually, FPCD claims that cognitive design is such that the background landscape and the positioning of representations upon this landscape jointly determine the outcome, the Total Cognitive States (TCSs) that the cognitive system displays. Upon the landscape, these TCSs are realized as the TCS-realizing points. Whatever will be displayed at the higher level of cognitive system, the occurrent cognitive states of this system, is thus potentially determined here at the middle level, offering a structure and non-arbitrariness, without the need of any algorithms to intervene in the process and to determine its outcome.
The topography of the activation landscape guides potential cognitive transitions. Notice now that there is a hand-and-glove or joint interaction of “content-appropriate cognitive transitions” here that determine the meaning. So what is described by FPCD is actually a semantic precondition for productivity. It is a description of how representations follow a structure in such a way that they result in the production of relevant meaning.
In
this way compositionality is articulated by the help of the Fundamental
Principle of Cognitive Design (FPCD). FPCD is indeed a rendering of
compositionality. Compositionality is a requirement to explain how the meaning can contribute to productivity. In the above picture meaning does
contribute to productivity in a much more plausible manner as this would be
possible for atomist generalist requirements for compositionality. In fact, the FPCD
picture also brings systematicity or syntactic requirements in a much more
plausible interdependency relation than this is accounted for by the classicist
generalist picture concerning compositionality. It gives the sense to how the
background structure of the cognitive system is related to the issues of
meaning. This is not possible in a real plausible manner by the generalist
requirement to substantially limit the issue of compositionality to atomism.
FPCD is a rendering of
particularist compositionality – a story telling how meaning is possible on the
basis of particularist dynamical patterns. FPCD actually presents a case of
particularist dynamical pattern supporting the production of meaning.
It is not difficult to see not only that FPCD offers a story featuring compositionality. FPCD picture offers a particularist compositionality.
What are the characteristics of particularism? They feature embracing of holism as against atomism. Particularism also offers normative authority of particular patterns and their relevance, as against the normative authority of generalist patterns.
Now the picture of compositionality offered by FPCD certainly does not offer any atomistic proposals about how the meaning enters into cognitive processing. Rather it offers the background multi-dimensional dynamical molding landscape picture, which seems to be the very mark of holistic approach.
The meaning and its role in cognition is certainly also not rendered by FPCD according to generalist patterns, which opt for the projection of the same atomistic not changeable and context independent meaning of a certain feature over a whole range of cases. Rather, the joint molding of “the high-dimensional topography of the activation landscape and the positioning of TCS-realizing points on that landscape” leads us to expect the contribution of meaning of a single feature as a joint result of the tendency that it brings along with the positioning upon the non-classically rendered background. This however means that each single appearing of the TCS as meaning determining feature will be slightly different and thus substantially context dependent. But there will be the contribution of the structure, and nothing will really happen in an arbitrary manner. Just that this structure will not be the one assured by generalist projection, but that of each single case. This all happens in an automatic adjusting manner is the cognitive system.
What is compositionality? A plea
for meaning as production, as creativity, along the poetic lines. The sheer
wrongfulness of atomistic and generalist conceiving of compositionality as a
function of tractable coming together of atomistic parts. One problem for such
a view is delimiting the units of meaning: are these words, phrases, sentences,
narration? The narrative contextual meaning seems to dominate the atomistic
parts such as words, so that these atomistic parts are even not necessary for
the production of a certain narrative meaning.
If meaning and
compositionality are described as based upon particularist patterns by
FPCD, this seems to be quite a long shot away from the usual generalist and
atomistic presenting of compositionality. So at this point it seems to be
sensible to ask in a fresh manner what exactly compositionality should be. If
FPCD is right, then the real compositionality is not assured by an atomistic
fixing of meaning and by its tractable extending over a range of generalist
pattern governed cases. Rather, a new way to look at compositionality offers
itself. Compositionality determines the effectiveness of meaning in a cognitive
system. But this effectiveness just cannot come through a shallow repetition
according to generalist pattern. The meaning is rather a production, a
creativity, something that happens according to poetic lines. The meaning of
the word “fly” in John Donne verse[6]
certainly does not seem to be reducible to the most standard dictionary
meaning. Such examples clearly show the sheer
wrongfulness of atomistic and generalist conceiving of compositionality as a
function of tractable coming together of atomistic parts. And it is clear that
poetry is an effective and relevant production of meaning. One problem for the
generalist view of compositionality is that it atomistically delimits the units
of meaning. But are meanings really limited to single words, to phrases?
Perhaps they are determined by sentences and words that just bring in some
meaning forces as the offered tendencies that may be and in most cases are
overridden by the context in which they appear. But a sentence does not offer a
context enough. Narration seems to be a much better candidate. The narrative
contextual meaning seems to dominate atomistic parts such as words, so that
these atomistic parts are even not necessary for the production of a certain
narrative meaning. Consider that the same story, the same fable, with the same
meaning, may be narrated by different choice of words, but still having then
same meaning. Consider that historical narration actually enriches itself by joining of contradictory narrative explanations.
This
all shows the profoundly wrong proposal of compositionality as meaning of a composite
entity such as a sentence being composed of meaning of its constituent parts
and of the manner of coming together of these parts. According to this simplistic atomistic
generalist approach the meaning of the sentence “The cat is on the mat” would
depend upon the contribution of meanings of semantic atoms such as “cat”,
“mat”, “being on”, and upon the manner of these coming together. But this is
just a parody of meaning which is really a production and thrives upon
unexpected but relevantly structured particularist patterns. Unhappily, such
parody of compositionality was basically accepted as an adequate rendering of
compositionality by most analytic philosophers, from Dummett and Davidson to
Blackburn.
FPCD is a principle.
As a principle, FPCD is not a generalist kind of principle; it is rather a particularist kind of principle, or at least particularism compatible kind of principle. So it is a kind of principle comparable to the Non Arbitrariness Of Composition (NAOC) principle that does not allow for arbitrariness, all in providing the needed structure. One lesson is that there exist particularist principles – the principles that follow particularist and not generalist patterns. The existence of such particularist principles is hard to acknowledge, as we are so much used for the principles to be attached to the generalist patterns. But this is false. One has to acknowledge the particularist principles, one species of which is NAOC. In this last mentioned principle, there is composition, and so there is structure. This is a particularist structure.
FPCD has the dynamics and richness intrinsically built into it. The hand in glove syntax and semantics cooperating feature is the guiding principle. These kind of features point in the direction of a particularist structure. Dynamics is not something for which generalist patterns would easily allow. Neither is richness something that generalist patterns would easily account for.
The bottom line is that HT is still under the spell of generalist patterns that govern the Fodor–Pylyshyn account of compositionality. This is the reason that he cannot even range the FPCD principle into the discussion of productivity. FPCD introduces particularist pattern, related to meaning. So HT elaborates this as it should be, in a particularist way, but is unable to treat it in the area of productivity-systematicity discussion according to the mode of Fodor-Pylyshyn to which he sticks in an explicit manner. Fodor-Pylyshyn offers a generalist approach, and generalist pattern as a way to approach things. But FPCD actually brings in and works out particularist approach. FPCD presents a particularist pattern.
The bottom line is that there are these two kinds of patterns: general patterns and particularist patterns. FPCD presents a case of particularist pattern. But the whole debate about productivity is made according to generalist patterns.
So HT does expose particularist pattern of FPCD. Well in opposition to the generalist presupposition that the meaning should be atomistic and thus generalist, the meaning is rather holistic and particularist. Why? Because the whole of the context determines the meaning. You cannot say that the meaning is that of a word. This would already take you down the atomist and generalist line. The meaning is not (as Dummett presupposes) the meaning of a whole, of a sentence, which is composed out of atomistic elements, say of words. The meaning rather belongs to the whole of the sequence. You cannot identify the meaning atomistic blocks (Word? Phrase? Sentence? Paragraph? Chapter? Do these have vague borders?). Rather, the meaning seems to be much more dynamic. Consider narration. The meaning in narration is not that of a word, of a sentence, of a paragraph, of the whole chapter. Consider that narration may proceed in several ways and that it still may remain the same narration. There are different narrative forms of the same story or of the same fable such as narrated by different storytellers. But it is still the same old story, the same narrative block. This shows that the meaning is really holistic and not atomistic at all. The same narration stays over several quite distinctive variations along several different narrators, say narrators of folk fables, or narrators of historical events, or narrators of historical stories. Consider that in this later case the same event will be even enriched by the variations in narrative approaches. If different political approaches of narration are applied to the same historical event, we can say that these approaches are in a sense not incompatible at all. They rather give a broader and sharper meaning or richer overall meaning to the same event. So this is a proof that inconsistencies enrich the meaning, and in this sense they do create meaning. But if this is the case then all the generalist presuppositions about atomism and tractability of meaning are simply false. There is no real atomist meaning, because there are wide variations possible in narration that contribute to the meaning, that enrich it. Atomism really cannot survive. An important feature that goes well along with dynamicism and narration, but not with atomism and generalism: meaning is created; it is an effect of sense. This cannot at all be captured by presuppositions of atomism and tractability. Each time there is a meaning, a real sense, a novelty should be created. This is not even sufficiently appreciated by Freud who still has these atomistic explanations of puns, related to words, or to parts of words. The meaning goes well over these, it dominates the elements that are involved into it. The dominance and the novelty of meaning are closely related. But neither of these may be adequately rendered by atomistic and generalist tractable patterns. We have effect of meaning, meaning as something creative. This only shows that meaning is really particular and that it is not reducible to atomistic elements and to tractable procedures, in the manner as Dummett supposes, when he determines the meaning to be a function of tractable bringing together of atomistic words meanings into the meaning of a sentence. This generalist picture is completely wrong picture of the meaning, not just slightly misguided. The lesson to be learned here is that meaning is really an effect of surprise, of particular patterns. So the meaning is intrinsically related to the particularist and holistic effects. This is how our language functions and evolves. This is also why the meaning poses such a difficult question. But it is a difficult question for generalist patterns, not for the particularist patterns view. The perspective should be completely changed towards particularist patterns in order to understand meaning and compositionality principles related to it. Holism and dynamics then become important for meaning, because these follow particularist patterns.
Is this holistic approach captured by FPCD? The answer is: yes. Notice that FPCD proposes dynamics; it proposes holistic and constantly changing landscape, that is rich and does not proceed along the atomist and generalist ways. The meaning is not a matter of atomistic elements, according to FPCD. It is matter of hand-and-glove collaboration of syntactic and semantic forces. What does this mean? It means that there is this landscape, not tractable or atomistically capturable, and that the meaning comes forward as an effect upon this landscape, with the change of this multidimensional landscape.
HT talks about the “high-dimensional topography of the activation landscape”, thus about rich and dynamical background upon which semantics has its hold. What can one understand on the basis of this? That the meaning or semantic is really something holistic. There are thus holistic preconditions of the semantic meaning, the background that enables the meaning and that dominates it. We can take an example of poetry, of words used in a poem. It is clear that the whole of holistic background contributes to the meaning there, and certainly meaning is not a function of addition of atomistic meaning elements according to tractable procedures, in a kind of mereological manner. The meaning is rooted in the background, it is based upon the not tractable background. The dictionary meaning of words is nothing but a contributory factor in the holistic background of this multi-dimensional landscape. In a poem, the words do not have their meaning as a function of atomistic meaning coming together in a tractable manner. The background is of the utmost importance for the production of the meaning in a poem. Now we can read a poem. It is just not sensible to presuppose that the meaning given by the poem in question is a function of contribution of atomistic meaning ingredients. A poem makes it very explicit that a meaning of a certain word is given by the background landscape upon which the whole of poetic narration succeeds. Here is a verse from John Donne:
19Call us what you will, we are made such by love;
20 Call her one, me another fly,
It would be strange to claim that the word “fly” means an animal here that we usually designate with this name. You can see that the meaning of the word fly gives just an indication of direction towards the overall meaning of this verse inside the whole of the poem, and this succeeds upon a background landscape [7]
Now background has the most important role in the overall production of meaning.
The idea is that poetry is not an aberration of language, but that it shows the very truth about how language works.
There is this background that structures the meaning, and also the meaning of so called elements that appear in the whole. There is no sense to insist upon an atomistic meaning of a word, such as of a word fly. The meaning is much wider, and it comes from the background. The generalistically accountable dictionary meaning of the word fly may at most figure as contributory indicator of some part of direction for the overall produced meaning.
There is this activation landscape. This means that the background is much richer as that whatever gets activated upon it. Whatever is activated to determine the meaning of a whole is just one possibility upon this rich background. Because the background is holistic (multi-dimensional landscape is in the back of it) it just gives an indeterminate indication how the meaning should get its capture[8]. But once the pattern gets formed, also through whatever comes upon the background landscape, there is determinacy of the meaning, the determinacy that comes from the background particularist pattern. This pattern is not repeatable, it is unique, but it is not arbitrary at all. It gives the right and determinate direction to meaning through its unique particular determinateness.
The activation landscape is the background for possible realizations of meaning. These realizations are unique.
Whatever appears as a meaning here cannot be captured by atomistic means because of its rootedness upon the background of multi-dimensional landscape.
The captions of meaning[9] are determined by the background. They are just slogans or headers upon which the background comes to the fore.
TCS’s are Total Cognitive States. The main mistake is to look at TCS’s in an atomistic manner. What is TCS of a poem, of your remark? Where exactly to delimit it? Such a delimitation is not possible because the meaning that they show is determined in a holistic multi-dimensional background.
There is rich topography of the background landscape. And there is positioning of the so called TCSs at this landscape.
There is the talk about “cognitive transitions” that are “content appropriate”, i.e. the content determines the meaning. But content may best be seen perhaps as being rooted in the background landscape, so it is not some kind of atomistic matter. The meaning is thus as well holistic and not atomistic.
Now hand and glove picture. There is dynamics of the background landscape. By positioning of the captions of meaning upon the landscape, the background landscape gets molded itself. But molding of the rich dynamical landscape, all the time transformed by the positioning of TCS’s (vague indicator meaning forces) produces the effect of meaning, which is itself a dynamical matter, having to do with such things as surprise. The meaning is relevant, becomes constantly the meaning through novelty, through the surprise and through production of meaning on this basis.
Notice also mentioning of a whole range of possible directions in which the meaning might go. The actual meaning is determined through a range of these possibilities.
What is the teaching of the above? It is a view of production of meaning, and actually a view of meaning as production – as production upon a multi-dimensional landscape. This requires richness and holism. So meaning is produced in a hand and glove constantly appearing match of the background and positioning of TCS’s upon this background, which is itself a dynamical process.
Particularist pattern of a relevant kind is substantial for this whole process of meaning. This beautiful pattern is relevant, relevance providing, and it is dynamical.
FPCD is a principle, but not a generalist and rather a particularist principle, better the manner of putting things together, a structure. It is a non-arbitrary structure built upon a particularist pattern, relevant pattern, i.e. beautiful pattern.
The main claim was that FPCD introduces particularist pattern. Now this particularist pattern is actually what the production of meaning is based upon. But because of this an account of meaning is incompatible with generalist Fodor-Pylyshyn account based upon generalist patterns.
HT also says the following about semantics and thus about the basis of meaning:
“Structural encoding of semantic properties and relations Key semantic properties and relations of total cognitive states are encoded within the structure of the mathematical system that is the locus of cognitive design. Semantic properties and relations are encoded by mathematical properties and relations of mathematical states that realize those TCSs.” (HT 155)
This is actually rendering of the encoding of semantic properties. It is description of the cognitive background as the basis of meaning and of semantic properties, as possible activation states, that also exercise their force upon the actual meaning of TCS’s. There is this rich dynamical background that may only be described mathematically, because mathematical description fits to the dynamics of a system. The encoding however does not succeed arbitrarily but rather has a structure as its basis. This can only be a particularist shaped structure, based on holistic dynamical pressures bringing non-arbitrariness along with them.
In FPCD a full-blooded holistic and particularistic approach seems to be present already. The question therefore looms large why in HT we cannot find compositionality in an explicit manner. One of the reasons besides to others that we elaborate is that the whole vocabulary or the conceptual scheme in HT book is attuned to soft generalities or to the ceteris paribus clauses. At the same time there is no direct vocabulary of beautiful patterns present in the HT book. In the sense of elaboration of FPCD picture and proposal, HT entirely appropriates particularism. But he is unable to make this last step of explicitly acknowledging it because the vocabulary that he accepts goes against it.
If, on the other hand, you approach the problem of compositionality from the point of view of beautiful patterns, then it becomes somehow smoothly natural that you deny any kind of normative authority to general principles – both to the general principles in the rigid form without exceptions and also in the form of the soft ceteris paribus clauses. All that is needed in order to accomplish this is just the particularist interpretation of the FPCD principle. We will now argue for this point and for related issues.
2. Why is FPCD not explicitly recognized as principle instituting
compositionality, why does the discussion of FPCD not come under the heading of
compositionality?
The FPCD does not come under the heading
of compositionality in HT, although the meaning is discussed in it, and the
structural contribution of meaning to the cognitive capability of productivity.
The answer: the issue for HT is to argue for ceteris
paribus (CP) generalities as against exceptionless generalities.
The answer to the question why FPCD is not
explicitly recognized as a principle in the register of compositionality comes
from the overall project into which HT is engaged. The overall HT project is to
argue in favor of generalities with exception or for ceteris paribus generalities, as against the
generalities without exception, in the area of philosophy of psychology and
cognition. This is certainly a right track to follow and a valuable exercise.
But CP generalities are still generalities. So the
normative authority of the general leads the HT project.
However, CP generalities are still
generalities. They are effective under the normative authority of the general
that they bring along with them.
But,
can CP generalities contribute to the meaning or to the explanation of meaning
at all? Consider Donne’s usage of the word “fly”. Now try to apply the
following CP procedure to the meaning explanation of this case, something such
as “Fly means fly, unless something in the context comes in and overrides the
original lexical meaning”. We can see that there will be no explanation coming
at all from such a principle. The reason is that it is a general normativity
based principle. On the other hand, it seems that the invoking of particularist
patterns and of their contextually bound structure may show a direction of
non-arbitrary explanation here.
As FPCD offers a particularist
pattern covering compositionality and meaning it becomes understandable that it
may not be possible to adequately account for it by the help of the generalist
CP patterns.
Now FPCD offers a particularist pattern covering compositionality and meaning. Try now to apply CP generalities account on FPCD. It turns out that it will not be able to work. The reason is that CP approach just offers general patterns as a measure. But it is not possible to account with the help of generalist patterns the specificity of particular pattern determining meaning, such as proposed by FPCD.
3. Once we see that the specificity of particularist patterns
regulating meaning cannot be adequately presented by generalist patterns, this
leads to natural charitably inclined gesture to offer help to HT so that he
would recognize his own compositionality proposal for what it is.
It seems to be a good bet that
particularist patterns will not be able to be adequately represented by
generalist patterns, such as by CP generalities. A natural gesture here is to
offer a helping hand to HT and make him realize what he was up to as he
endorsed FPCD. He was up to expose particularist patterns as founding meaning
and compositionality. We may offer such help and give HT a little push in
direction of particularism, for we do explicitly embrace particularist patterns
as founding meaning and compositionality.
HT does indeed articulate well
productivity and systematicity. These are compatible with generalist interpretation,
and systematicity presents a syntactic requirement for productivity.
It is interesting that in his book, HT
articulates well productivity in the general sense and systematicity. But these
are not semantic features. Systematicity, say, is a syntactic feature, and as
such it is much sooner to be ranged under generalist patterns and under
explanation according to these. Systematicity represents syntactic requirement
for productivity.
Compositionality that involves
semantic contribution to productivity is a much harder issue.
Compositionality and its adjoined semantic
presents a much harder issue. It seems that the issue of meaning is hard. Most
people do opt for holistic view of meaning, but find it hard to combine such a
view with the rest of their overall atomisitic and generalist tractable
presuppositions in this area. Devitt opts for meaning molecularism, but again
has similar generalism induced difficulties. Fodor offers atomistic account of
meaning that appears as a kind of reductio ad absurdum of generalist atomism.
But compositionality only
presents a harder issue for an approach based upon general patterns.
However, compositionality is such a hard
issue for the general pattern approaches only. Once as one embraces the
possibility of particular patterns that bring relevance with them for a kind of
free, compositionality is not such a hard issue at all. Why?
As semantics does actually follow
particularist patterns, and as FPCD presents such a pattern, this will help HT
to recognize FPCD dealing with compositionality and meaning.
Semantic actually follows particular patterns. HT has seen this well enough. He has proposed FPCD as a particularist pattern. Once that we have explicitly embraced particularist patterns, we can help HT to recognize his move towards particularist compositionality that he was not able to recognize himself though, because of his commitment to generalist patterns, although generalities with exception form CP patterns.
References
Dancy, J. 1993. Moral Reasons. Oxford: Blackwell.
Dancy, J. 2000. Practical Reality. Oxford: Oxford University Press.
Dancy, J. Unpublished. Ethics Without Principles.
Fodor, J. and Pylyshyn, Z.1988. Connectionism and cognitive architecture: A critical analysis. Cognition 28: 3-71.
Horgan, T. and Tienson, J. 1996. Connectionism and the Philosophy of Psychology. MIT Press. [HT]
Potrč, M. 2002.
Non Arbitrarines of Composition and Particularism. GPS 63: 235-255.
Potrč, M., Unpublished. Particularism and Productivity Argument.
[1] In the theory by Fodor and Pylyshyn this requirement or presupposition is present in their Compositionality Principle: “insofar as language is systematic, a lexical item must make approximately the same semantic contribution to each expression in which it occurs. […] Similarity of constituent structure accounts for the semantic relatedness between systematically related sentences only to the extent that the semantical properties of the shared constituents are context-independent.” (Fodor and Pylyshyn, 42)
[2] HT is used as an abbreviation for the position in the Horgan T. and Tienson, J. 1996 book Connectionism and philosophy of psychology. The position of this book is that of Dynamical Cognition as a model of mind, based upon ceteris paribus clauses or upon generalities with exceptions. HT is used here as if it would designate a certain unique person holding such a view, despite that the actual authorship comes from a joint effort of two persons. HT also captures the individual that it designates in a certain moment of time. It is possible that none of the actual persons entering into the 1996 time slice of HT now still shares exactly the same view.
[3] In a private correspondence HT said that he tends to summarize his approach towards compositionality such as proposed in his book. But there is no explicit discussion of compositionality to be found in the register of HT book at all, despite of his several counts of mentioning productivity and systematicity.
[4] At the very beginning of his book, HT delimits his inquiry in such a manner that it would not include the discussion of contentfull intentional states and of origins of intentionality.
[5] Fodor and Pylyshyn (p. 45, say) in advance condemn the very possibility of the connectionist approach explaining productivity, systematicity and compositionality. HT to the contrary uses connectionism as an inspiration upon which to build an approach of Dynamical Cognition.
[6] This verse is also reproduced in the next section: “Call us what you will, we are made such by love; Call her one, me another fly”.
[7] The landscape that was indicated by Searle as the background while he discussed intentionality.
[8] Lacan mentioned the capturing point of meaning.
[9] Take a look at the preceding footnote.