Introduction
Literature Review
Theoretical Preliminaries
Experiments
Theory and Review
Endnotes


5. Some Observations Regarding the Nature and Structure of Language

5.1 Introduction

The question that led to this research arose as I was developing spelling checker dictionaries. As I typed the dictionaries into the computer, I found, as I moved from the B's to the C's, that the semantics of the words themselves seemed to change. Then same thing happened when I finished C and started D. Again and again I felt this strange effect. Something about the words with different initial letters felt different in some undefinable yet unmistakable way. At some point, I decided to spend an hour or two classifying words to see whether or not this peculiarity might have some basis in semantic predispositions among the phonemes -- semantic predispositions that perhaps could even be quantified. I was sufficiently surprised by the regularity of the results that I found myself drawn to devise the experiments outlined herein.

In expressing what I have observed, I sought a way -- any way -- to articulate what I observed. The means by which an empirical fact gets articulated is really irrelevant so long as what was observed can be expressed clearly enough that others can understand it, and perhaps verify it for themselves. What follows regarding the 'theory' underlying the data just presented is written in that spirit. It is not an attempt to set up a coherent framework that I feel reflects some ultimate truth.

5.1.1 Informal Overview of the Empirical Facts

Before I proceed to offer a more formal description of my findings, let me first itemize in as straightforward terms as possible what I believe these 14 tests to show.

· Phonosemantic Hypothesis: I believe the data in Appendices I-XIV taken as a whole to constitute very strong evidence for the Phonosemantic Hypothesis:

The Phonosemantic Hypothesis:
In every language of the world, every word containing a given phoneme has some specific element of meaning which is lacking in words not containing that phoneme. In this sense, we can say that every phoneme is meaning-bearing. The meaning that the phoneme bears is rooted in its articulation.

· Natural Classification: In order to devise proofs for the Phonosemantic Hypothesis, I define various types of classification systems. I first defined 'Natural Classifications'. One intuits the classifications to be ungrammatical; this suggests that the Natural Classes as psychologically real. I use the following four criteria to discover the Natural Classes of a language:

Natural Classification
Criterion 1. Very nearly every word within the given natural set fits in some semantic class.
Criterion 2. Each semantic class contains a fairly large percentage of the words in that natural set.
Criterion 3. There are relatively few semantic classes in the classification.
Criterion 4. The semantic classes in the classification are distinct

Again, natural classifications are defined over natural sets of words, and semantic classes themselves must be composed of words in a natural set. A natural set must be definable by means of a single non-disjuntive characterization. For example, 'all French monosyllables' form a natural set, as do 'all Chinese words referring to birds' as do 'all words beginning with /p/ and referring to a fruit'. But disjunctive characterizations, like 'Russian verbs of motion or Russian words beginning with /s/' do not describe natural sets. The natural sets over which the natural classifications in this dissertation are defined are those which are characterized in terms of the phonological form of the word, such as 'all English monomorphemes with /r/ in second position'.

We cannot easily abstract away from these Natural Classifications, because they lie at the very heart of what for us distinguishes a word from a mere string of sounds. Sound-meaning actually underlies all word semantics. But because we cannot in general stretch our mind enough to abstract away from the Natural Classes, we must work within them and through them if we wish to catch glimpses of the effects of phonosemantic Iconism.

· Phonosemantic Classification: If one is to prove the Phonosemantic Hypothesis for a language, then the language must be shown to conform to the stricter criteria for a Phonosemantic Classification outlined below. From this definition it is clear that the phonesthemes are subject to the first four requirements of a Natural Classification and then some. All classifications must submit to the limitations imposed by a language's Natural Classes, and this is true also of the phonesthemes, which taken as a whole form what I have called a Phonosemantic Classification:

Phonosemantic Classification
Criterion 1.
Very nearly every word within the given natural set fits in some semantic class.
Criterion 2. Each semantic class contains a fairly large percentage of the words in that natural set.
Criterion 3. There are relatively few semantic classes in the classification.
Criterion 4. The semantic classes in the classification are distinct
Criterion 5. Each word fits into an average of a fairly large number of classes.
Criterion 6. The semantic classes are narrowly defined. By a 'narrowly defined' semantic class, I mean one which encompasses a small percentage of words in the language as a whole.
Criterion 7.. A much smaller percentage of the words which do not match the relevant phonological characterization fit into any class.
Criterion 8. Those words that do not match the relevant phonological characterization but which nevertheless do fit in the classification fit on average in a smaller percentage of classes, than those words which do match the phonological characterization.
Criterion 9. Any class in a Phonosemantic Classification can be defined narrowly enough that words not matching the relevant phonological characterization are excluded from it.

As one develops a Phonosemantic Classification for a set of words bearing a common phonological trait, one finds that although one cannot grammatically violate the Natural Classes, one has a good deal of liberty otherwise to classify things as one wishes. Various Phonosemantic Classifications emphasize or suppress diverse aspects of the phonosemantics. For this reason, some aspects of a Phonosemantic Classification are not psychologically real in the way that the Natural Classes are psychologically real. Phonosemantic classifications are nevertheless important, because it is the possibility of creating them which serves as the primary method I have used for verifying the Phonosemantic Hypothesis. If one is to show that phonemes have meanings, one must show that they are associated with some semantic domain which other phonemes are not associated with. Speaking in information theoretical terms, we must show that phonemes must carry information. This is essentially the requirement that Phonosemantic Classifications impose.

· Phonosemantic Association: So monosyllabic words in English which contain a given consonant fall within much narrower semantic domains than one would expect if the relationship between phonology and semantics were arbitrary. I find this is in large part due to a natural process called 'Clustering' or what I have termed Phonosemantic Association:

Phonosemantic Association
When semantic domain S is associated disproportionately frequently with phoneme X, then people will be inclined to associate semantic domain S with phoneme X productively.

Clustering is effectively the tendency for language to associate identifiable referents with every form. Informally, it is the tendency to try to make everything mean something coherent. Because of Semantic Association, we also find the reverse phenomenon -- that phonemes are disproportionately represented in certain Natural Classes called the phonesthemes.

· Resemblance to Articulation: This semantic domain which becomes associated with each phoneme through Phonosemantic Association resembles the articulation of that phoneme. When the resemblance cannot in principle be literal (because, for example, there is no literal way for the articulation of /gl/ to resemble reflected light), then the resemblance is metaphorical. (In this case, the phoneme /g/ is articulated with a closure deep in the throat, and this translates metaphorically into 'hiddenness'. Thus light in /g/ has a hidden source -- it is reflected. The /l/ is the most 'liquid' of consonants. That is, it's articulation conforms to that of its neighbors. This translates metaphorically into things which have mass but no particular form, like air, water, and also light.) This seems to me one piece of circumstantial evidence that phoneme semantics is not only conventional in nature, but also natural, or Iconic.

· Specificity of Iconic Effects: Further evidence that phoneme semantics is inherently natural or Iconic can be found in the very specific semantic effect that the presence of a given phoneme has when words which fall within the same very narrow Natural Classes are compared. This effect again tends to resemble the articulation of the relevant phoneme, albeit as it is projected metaphorically onto the Natural Class under study. For example, in whatever context, /r/ tends to have an intense and abrasive effect on word semantics. The phoneme /t/ tends to presuppose directedness toward some goal. The phoneme /k/ tends to cut, classify and contain, and so forth. For example, in the Natural Class of light, this 'cutting' manifests as 'color'. In the Natural Class of geological formations, the 'cutting' manifests as 'coves' and 'crevasses'. In the Natural Class of 'music', the cutting manifests as 'chords' and 'keys'.

· Interference of Concrete Reference: The more concrete and unambiguous the referent for the word, the more difficult it is to fit into a Phonosemantic Classification. Throughout our experiments, those words which did not fit in the Phonosemantic Classification without exception fit in one of the following Natural Classes, which we term 'the Concrete Nouns':

Concrete Nouns:
people, titles, body parts, clothing, cloth, periods of time, games, animals, plants, plant parts, food, minerals, containers, vehicles, buildings, rooms, furniture, tools, weapons, musical instruments, colors, symbols, units of measurement

The reason for this pattern apparently concerns what the word refers to. If the referent for a word by its very nature is connotative or interpretive, then the word's phonosemantics can cooperate with its referent. For example, 'slide' is a smooth motion. The smoothness and slipperiness so common in /sl/ shows up in the actual referent for 'slide'. If, however, the word refers to some very concrete and identifiable object in the world, then the phonosemantics of the word seems to impose a connotation or interpretation on the word rather than affecting what the word actually refers to. For example, the referent for 'dog' is an animal. Its referent is not that of a ugly animal. The dreariness which appears disproportionately in words containing /d/ manifests in 'dog' as a connotation superimposed on the word 'dog'.

· Opposites: If a Phonosemantic Classification includes a given semantic domain, it's very common for it also to include the opposite of that semantic domain. For example, words containing /p/ disproportionately often involve putting and picking up, pouring and also sponging, points and planes, problems and their solutions, etc. Words containing /k/ feature kings and commoners, catching and dismissing, combining and cutting, cruelty and kindness, etc. It was pointed out that a word and its opposite are very similar in meaning. This phenomenon would make sense if we thought of Iconic meaning as a substrate which underlay other levels of meaning. What was therefore one thing -- Length -- on the level of Iconic meaning would be viewed as two aspects of that thing -- Long and Short -- through the prism imposed by the Natural Classes and reference.

· Positional Iconism: The position that a consonant occupies in a syllable also affects its meaning. Consonants that appear before the vowel form the backdrop for the action of the word, and consonants that appear after the vowel express the result of the action implicit in the word.

· Cross-Linguistic Iconism: When all words matching a given phonological form are compared cross-linguistically they tend to fall into the same Phonosemantic Classification also in vocabularies that are not cognate. Furthermore, if the relationship between phonology and word semantics were attributable only to Clustering and therefore wholly conventional and not Iconic (as e.g. Sapir believed), then one would not predict that this would be the case. The preponderance of cross-linguistiuc phonesthemes therefore constitutes evidence for the Iconic and productive nature of phonosemantics.

· Productivity: When informants are asked to perform experiments which reveal their linguistic intuitions regarding the semantics of quasi-words, the results consistently show that Phonosemantic Association is productive. Furthermore, the fact that native speakers consistently prefer to associate nonsense words with some phonologically similar words and not others, suggests that factors other than Clustering are also affecting their intuitions.


5.1.2 The Paradox

The first issue that one generally confronts in considering the possibility of a large scale phonosemantic correlation -- and which in my view tends to lead both nautralists and conventionalists to jump prematurely to inaccurate conclusion -- is the obvious paradox implicit in the Phonosemantic Hypothesis. If there is some correlation between phonetic form and word semantics, how can this possibly be reconciled with regular sound change, not to mention the existence of diverse languages, synchronic phonological processes of all sorts and any number of other very obvious counter examples?

As the history of linguistics demonstrates, the tendency has been to jump to the conclusion that they can't be reconciled, and that there therefore can be no phonosemantic correlations. One decides in advance of even investigating the data that there's no point wasting one's time. After having looked for some time at the data and having become convinced that phonosemantics at least deserves further consideration, one is next tempted to jump to another conclusion -- that phonosemantics must be a fossil of earlier etymological processes, and that it therefore must be more pervasive some languages and vocabularies than others.

In this dissertation I have endeavored to make clear to the reader in replicable experiments why I myself felt cornered into accepting the pervasiveness, cross-linguistic, Iconic and productive nature of phonosemantics. Any reader who still finds him- or herself in doubt is challenged to conduct the tests I have outlined for any language. It is my belief that anyone who gives the data the same level of attention that I have, must come to the same general conclusion. And the phonosemantic literature amply supports this position. It was only the process of actually performing the experiments that has made converts of most phonosemanticists, present author included.

What then are we to do about the paradox? The undeniability of both sets of conflicting data forced me to reposition myself relative to this issue. If we can no longer deny that pervasive phonosemantic correlations coexist with equally pervasive and apparently conflicting phonological processes of other types, then it no longer makes sense to ask whether phonosemantic correlations can be reconciled with the overwhelming masses of counterevidence. They have to be reconcilable, because they both exist. The question therefore only remains how they can be reconciled.

In the remainder of this discussion, I will assume that the Phonosemantic Hypothesis is valid. If I have failed through these 14 experiments in raising any doubts in the reader regarding arbitrariness of the sign, then I shall have nothing more to say on the subject. Rather I will proceed from asking whether it holds to inquiring how it can hold.

Generally when one encounters an apparent paradox, it is the result of failure to draw fine enough distinctions. One is treating two things which really are different as if they were the same. In this case, I believe the resolution to our paradox lies in the recognition that word semantics is not one unified thing, but that it is structured -- some parts of a word are affected by phonological change and some parts are not. Jakobson found his resolution to the paradox by distinguishing three levels of word semantics in a manner outlined by C. S. Peirce, and my approach is very similar. I do not take issue with Jakobson and Waugh. The reason I do not use their terminology is that I have approached the subject over time in my own way in isolation from their work, and consequently have found my own way of expressing what I believe to be a similar insight.


5.2 The Structure of a Word

Let me begin by itemizing the various pieces I find in the puzzle and then proceed to attempt to interrelate them. The reader will please forgive the extent to which I simply reiterate fundamentals for the sake of exposition.

5.2.1 Structural Levels

I assume word semantics to consist in at least the following components:Structural Levels

Phoneme -> Phonetic Feature
Morpheme
Word

All the data provided in these 14 experiments is confined to monomorphemes, and most of it is confined to monosyllables. So I will be dealing in the following discussion only with phonemes as analyzed into phonetic features and with words. Affixation will not play a role in my discussion.


5.2.2 Semantic Levels

I recognize at least these three levels of semantics:Semantic Levels

1. Iconism
2. Classification
3. Reference

No word exists that does not have semantics on all three levels. Put another way, a word that does not have these three levels of semantics is not a word. The referent is the most accessible semantic level viewed from the perspective of parole and the least fundamental viewed from the perspective of langue. Iconic meaning is the most basic semantic level from the perspective of langue and the most obscure from the perspective of parole. The classificatory level lies in between these two.


5.2.2.1 Iconism

I have argued on several grounds that from the viewpoint of langue, the most basic semantic level is the Iconic. Our description of word semantics is simplified if we think of the synesthetic effect of the sound as underlying other processes which are superimposed on it. The most important argument for this analysis is that in many ways -- some of which I have tested for -- words can be shown to have a meaning which is broader than the combined semantics of all of its dictionary senses. Furthermore, it can be shown that this broader semantics is related to its phonological structure in a regular way. I think we therefore most effectively view this Iconic meaning as the substrate underlying and uniting the semantics of a word, and we view other aspects of semantics as superimposed upon it.

I have shown this to be the case :

· when analyzing invented definitions for nonsense words. (Experiment 11)
· when comparing phonesthemes cross-linguistically. (Experiment 10)
· when analyzing sets of words whose meaning is so similar that the only differences between them appears to pattern regularly with the phonological structure of the word. (Experiments 2, 5, 6, 9)
· in the fact that the phoneme profiles that result from Phonosemantic Association resemble the articulation of the phoneme. (Experiments 1, 6)
· in the fact that phonetic features seem to have a coherent meaning. (Experiment 2, 6)
· in the fact that a phoneme's position in a word seems to affect how it influences the word's semantics. (Experiments 8, 9, 10)
· in the fact that words invented even for semi-abstract images resemble each other much more than one would anticipate if there were no Iconism active in language. (Experiment 14)

The Iconic effect can also be seen in areas I have not considered, such as in idioms, metaphorical usage, poetic and productive speech.

I think of Iconism much in the way I understand Peirce. The Iconic dimension abstracted away from the levels of classification and reference must be one unified thing which refers to nothing other than itself. The Iconic is the level on which form is content, on which the word means what it is (structurally). I have observed that when Phonosemantic Classifications are found for monosyllabic words containing some arbitrary consonant, the words tend to fall into classes which reflect the phoneme's articulation. This is, however, only indirect evidence for Iconism; it is not Iconism as I understand it. The Iconic meaning that lies at the root of word semantics as I understand it can only be experienced as a feeling-tone, which becomes articulated as something more concrete when it is viewed through the prisms of the Natural Classes.


5.2.2.2 Classification

As mentioned in the text, languages divide words into Natural Classes. English, for example, divides words into 'clothing', verbs of 'striking', 'musical instruments', and so forth. It does not classify words according to 'rectangular things' or 'objects which can break'. These facts are functionally determined and are part of the grammar of English. As speakers of English, we do not have a choice relative to this classificational system. As mentioned on several occasions, the criteria we use for determining a Natural Class are the four criteria which define a Natural Classification.

More intuitively, this can be observed in the fact that Natural Classes are not fuzzy sets. 'Washing' is a form of cleaning -- no two ways about it. Dogs are animals -- no ambiguity there either. Anger is an emotion and rage is anger, etc. Furthermore, sorrow and befuddlement are not anger and televisions are not animals. Of course, one can come up with examples of words that fall in the grey zone. Is plankton a plant or not? Tests could be devised to show that in the mind of a given English speaker 'plankton' either falls in a Natural Class of 'microorganisms' or it doesn't. It either is a 'plant', or it isn't. I suspect, people sometimes think they are classifying a word one way, when they in fact aren't. For example, if I come home from work and tell my daughter, ÒI saw a beautiful plant this afternoon,Ó and she feels later deceived when she learns that what I saw was plankton, then she does not classify 'plankton' as a plant whether she consciously thinks it's a plant or not. The English word 'plant' in her vocabulary does not include 'plankton'. In fact, I find once one starts testing in this way for how people actually use a word rather than how they think it should be used, confusion or fuzziness regarding Natural Class adherence is remarkably rare.

The Natural Classification is hierarchical. The Natural Class of mammals falls in the Natural Class of animals, as does the Natural Class of domesticated vs. wild animals. There is a fair amount of overlap among the Natural Classes as well. For example, many plants are also food, so plants fall into one Natural Classification based on their biological structure and some of them also fall in another classification based on how we prepare them for food.

True Iconism is inherently blind to the Functional Classes. It can be shown that True Iconism affects what the word is like rather than what the word refers to -- it affects its connotation as opposed to its denotation. The inherent blindness of Natural Classes to phonological form can be observed in several ways. First, phonemes are distributed fairly evenly in the Functional Classes and the Concrete Noun classes in particular. Second, if one chooses some less concrete class, such as 'verbs of washing' one generally finds almost all the phonemes represented, though usually not with the same frequency as one finds them in the language overall -- Clustering gives rise to disproportions. Third, one tends to find that each phoneme gravitates toward words concerning certain types of washing. Fourth, it is exceptionally rare that I have found a Natural Class for which every word contains a given phoneme, and in the cases where this seems to be the case, the Natural Class is very narrowly defined indeed.

However, there is another phonosemantic process which I have called Phonosemantic Association or Clustering, which does give rise to disproportions between certain kinds of Natural Classes, called phonesthemes, and phonological forms. Phonosemantic Association does not conspire to make each semantic domain have a certain phonological form, but the inverse. Phonosemantic Association in general conspires to associate with each phonological form a unified conceptual semantic domain. Phonosemantic Association is a special case of a more general tendency in human psychology which I call Semantic Association which seeks to attribute a unified meaning to any form whatsoever. The phonesthemes or phonological disproportions in the Natural Classes are a side effect of Phonosemantic Association.

In this context, experiment 7 (see Appendix 7 and 4.7ff) is particularly instructive. There I divided Location words into two types of Natural Classifications. One of these was a Phonosemantic Classification which only accommodates words with the appropriate phonological form. The other was a Functional Classification for the same set of words, which is by definition not phonosemantic and which accommodates with equal ease words with any phonological form so long as they belong in the Natural Class concerned. Hence there seem to be two really distinct components of language operating simultaneously. The Functional Classification of Location words is the one we all think of when we think of locations: geographical vs. political divisions, buildings, streets, rooms, etc.. The Phonosemantic Classification is a mix-match of various kinds of classes -- whatever the given phoneme seems to prefer. /b/ in English happens to have a lot of bogs, beds, bottoms of things and borders. This comes out as a sort of haphazard intersection between the classes of things that gravitate toward /b/ and the types of things we classify as 'places'. And there's no one right phonesthemic classification the way there are right and wrong (grammatical and ungrammatical) Functional Classifications.

It's therefore a curious thing, and a pretty strong indication that Iconism is indeed universally active, that Russian works about the same relative to both types of English Location classifications. One does find Clusterings that are specific to English. The fact that 'bunk' and 'berth' exist in English is probably in part responsible by Clustering for the fact that in English 'bed' also starts with /b/. But the fact that /b/ has a lot of words of blockage, barriers, binding and bases or foundations appears to be pervasive throughout languages and arguably also reflects its articulation.


5.2.2.3 Reference

5.2.2.3.1 Reference in General

I'm not prepared to discuss the nature of reference here -- whether it be a mapping onto a class of objects in the phenomenal world or more functionally defined. But whatever the nature of reference, we come back to the fact that the phoneme strings /kæt/, /dag/ and /hors/ all belong in a single class. The classificational level of semantics does not inherently express what they all refer to, only that they belong together. However, each member in a given Natural Class has a common element of reference. Although on the level of Classification, 'cat', 'dog' and 'horse' are just members of a class, on the level of Reference, it is clear that they all refer to mammals.

I think we do well to think of each unique referent as governing its own Natural Class. For example, 'dove' and 'pigeon' form a Natural Class because they have the same referent, but in my English, 'chipmunk' forms a Natural Class all its own. I know no co-referent for 'chipmunk'. I find most of the Concrete Nouns, therefore, to fall at the lowest level into very small Natural Classes -- classes often consisting of only one member.

I think of many of the less concrete Natural Classes, therefore, as containing many words for the same referent. For example, probably it's easiest to think of all the English monosyllables for forms of 'jumping' as having the same referent, namely jumping. I think the differences between 'hop' and 'bound' and 'jump' are due not to differences of reference, but primarily due to phonosemantic differences which define the inflections of jumping. There are other non-Iconic (arbitrary) differences between words which have the same referent. Their argument structures can vary. In this case their semantic selectional restrictions vary somewhat -- the prototypical subject for 'jump' is a person, whereas for 'hop', it tends to be an animal. Each of these words has senses apart from the sense in which they share a common referent ('jump' a car, bell 'hop', outer 'bound'), but these senses will be interrelated metaphorically -- each in their own way -- with the motional referent that they have in common. An excellent demonstration of deep reaching correlations between Natural Classes and argument structure is found in Levin (1993).


5.2.2.3.2 Concrete Reference

I have defined the Concrete Nouns as those nouns which are members of a particular set of Natural Classes. I also think of Concrete Nouns as those words whose referents people agree upon. That is, people in general differ as to what constitutes 'intelligence' or 'politics' or 'beauty', but they all pretty much agree on what objects in the world constitute the sets of 'goats' or 'hammers'. In addition, we have just observed that Concrete Nouns are those which on the lowest level of the hierarchy rarely share their referents with other words in the language. That is, many other animals share the 'animal' aspect of the referent for the word 'fox', but very few (if any) share all the aspects of the referent of the word 'fox'. This is not in general the case with Natural Classes other than the Concrete Nouns.

I have pointed out that whenever I have formed a Phonosemantic Classification, all of the words which don't conform to the classification end up being Concrete Nouns. In general, the Concrete Nouns are less susceptible to Clustering. The phonemes are more evenly distributed in the Concrete Nouns than in other words. I've also observed that whereas it's often possible to correlate the specific inflections of light or motion in very narrowly defined Natural Classes with their phonological form, this is not in general possible with the Concrete Nouns. If I take a very narrow class of Concrete Nouns, such as words for spices, it's very hard to see any correlation between the types of spices and the phonological forms of the words used to refer to them. This would make sense if we thought of each Concrete Noun as governing its own semantic domain or Natural Class, because it has a unique referent. If there are 30 words which share a referent with 'shine', then one can compare them phonosemantically and find that the inflections of light are related to the inflections of phonological form in the words. But if there is only one word in English for 'coriander', then there's nothing to compare phonosemantically. Seen another way, in the Concrete Nouns, the effect of the articulation of a word on its meaning cannot be observed through the medium of phonesthemes, because there are no other words which fall in a common Natural Class with it. The concreteness of the reference has isolated the word, and left it with nothing to be compared to.

As I've mentioned, it seems that certain senses of certain words must at all cost have very clear referents. That is, I think it's important that for a few words, like 'goat' and 'hammer' we all agree which objects in the world they refer to. I think the entire infrastructure of a language depends on these words with very unambiguous referents. If these words loosened their grip so that we could argue as much about what things are 'goats' as we can argue about what things are 'inane' or 'restorative' or 'musical', then language would cease to be usable as a tool to talk about anything. If it so lost its grip on the phenomenal world that no one agreed on what any word referred to, then it would become a completely self-referential mush -- all sound and fury quite literally signifying nothing. So these Concrete Nouns cannot loosen their grip to allow the freer dimension of their inherent semantics to become too salient or we lose the ability to convey information. It's so important that certain words remain anchored in what they refer to, in other words, that there's little leeway for the form to give expression to what they are like. If we get too carried away with what certain words are like, we might start arguing about what they refer to, and that would be disastrous.

Other words are then linked up to the Concrete Nouns by means of Semantic Relations, like hyponymy, meronymy, etc. We might say that language is anchored in the phenomenal world by means of concrete reference and these concrete words are in turn anchored to other words by means of Semantic Relations. For example, 'house' is a concrete noun. Pretty much every native speaker of English agrees on which objects in the world are 'houses'. They also agree that a mansion is a house -- that is, there is a relationship of metonymy/hyponymy between the words 'house' and 'mansion'. Indeed, we agree that a 'mansion' is a big house. We may, however, disagree about which big houses are mansions. The word 'mansion', in other words, has a looser grip on its referents than does the word 'house' and it is anchored to 'house' by a Semantic Relation.


5.2.3 Semantic Association

Reference really applies at the level of the word, not the phoneme, morpheme, or sentence. When we label something, we give it a word. Like reference, classification applies on the syntagmatic level of the word -- it is words which are classified. And as mentioned, all words in a given Natural Class bear a common element of reference. For this reason, we can also think of Natural Classes as having a referent. Iconism, on the other hand, is not applicable to any linguistic level (phoneme, morpheme, word). And Semantic Association applies on all linguistic levels (phoneme, morpheme, word).

Semantic Association seeks to imbue linguistic structures -- phonemes, words, morphemes, etc. -- with informational content. On the level of the word, this semantic disproportion is glaringly obvious, because that is the level at which we name things. The disproportionately small semantic domain within which a word is used is introduced into language consciously. We consciously decide that the string 'enchilada' will be used in the context of only one very specific type of thing in the phenomenal world, and that 'swift' will only be used in a different context. But this reference that is consciously assigned on the level of the word trickles down to the levels of the morpheme -- where the nature of the semantic disproportions is semi-conscious, and the phoneme and the phonetic feature where we are largely unconscious of the semantic disproportions. Nonetheless, all these lower levels, as I have tried to show, have informational content, because they are associated with semantic sub-domains, rather than with the entire semantic range of the language.

The result is that the phoneme, like the word, also governs a unified semantic space different from the semantic space of all other phonemes. And like a word, a phoneme has something akin to interrelated senses, such as we have seen many times throughout the dissertation in lists like these:

/d/

A1 End, Death, Sleep, Drug 50 10.4%
A2 Diminishment, Smallness 83 17.3%
A3 Breadth, Dragging On 41 8.5%
A4 Scarcity, Danger 60 12.5%
A5 Confusion, Discord and Barriers 69 14.4%
A6 Dark, Dirty and Dreary 97 20.2%
B1 Divisions, Groups, Amounts 130 27.1%
C1 Execution of Pending Process 91 19.0%
C2 Motion 31 6.5%
D1 Down 154 32.1%
E1 Good, Dear 33 6.9%
F1 Water 62 12.9%
G1 Light and Color 11 2.3%

In phoneme semantics, one can see the purely Iconic level, which affects the feeling-tone of words like 'glimmer', 'gleam' and 'glitter'. There the various inflections of light are reflected synesthetically in the sound in a manner which can't be articulated in words, because the effect of the sound is on a level that lies lower than the classificatory or referential levels of meaning. And one can also see a phoneme semantics operative on higher levels in schema such as that outlined above for /d/ in which phonosemantic classes are not purely musical, so to speak, but can to some degree be conceptualized and analyzed in concrete terms. It is only this latter type of semantics which I associate with Clustering.


5.2.4 Semantic Relations and Subcategorization

I think of Semantic Relations as functions which relate words and Natural Classes of words to one another. The classes of Mammals and Animals are interrelated by metonymy/hyponymy. The word 'long' is related to the word 'short' by antonymy, and so forth. In addition to Semantic Relations, we have in syntagmatics, of course, subcategorization and semantic selectional restrictions. I've found in the course of developing a lexical database that verbs select semantically for Natural Classes. And there is, of course, a universe of complexity in these domains, with which the reader may be familiar and which lie outside the purview of this discussion.

Concrete nouns in general enter into less complex networks both syntagmatically and paradigmatically than do other words. If we look at a Natural Class whose reference is not wholly concrete, but at least much more concrete than that of some Natural Classes -- the class of light -- we find the situation already considerably more complicated than what we find in most Concrete Nouns. We find, for example, several types of verbs involving light. In the Natural Class of 'looking' verbs, light serves as an instrument. There is a class of verbs for avoiding light: blink, flinch, blur, squint, etc. There is another class of transitive verbs in which light illuminates something else: flood, light up, illuminate, etc. And there is of course a class of intransitive verbs which simply express different inflections of light: glow, glare, shimmer, shine, etc. In addition to these, there are nouns for light. There are simply nouns for various inflections of light: beam, bolt, sheen, ray, etc. There are nouns for objects whose purpose is to illuminate: lamp, bulb, globe, light, chandelier, etc. There is a class of words for times when it is relatively light: day, dawn, dusk, etc. There is a class of nouns for which light plays a sort of instrumental role, such as reflecting surfaces (glass, glaze, gloss, mirror, etc. ) and a class of nouns referring to things through which light prototypically shines: (film, lens, glass, window,...). There is a class of gleaming celestial bodies (sun, moon, star, planet, comet...). And then there are the adjectives... The Natural Class of 'light' is related to large Natural Classes of words, such as words for color, for darkness and for fire. Fire in turn is related to classes for heat and dryness. In Natural Classes predominated by verbs, like 'Motion', the situation is more complex still.

All evidence that I have encountered suggests that phonosemantic Iconism is fundamentally blind to all of this. It sees no parts of speech and no Natural Classes and no phonemes or morphemes or Semantic Relations or paradigms. All these conceptual distinctions lie on higher levels of semantics. What we see in the phonesthemes and in the effect of phonological structure on the semantics of very narrowly defined Natural Classes seems to me the reflection of Iconism through the prism of these other overlying factors. And it is for this reason probably, that Socrates thought of Iconic meaning as being the essence of the word and as relating synesthetically to the 'essence' of the thing to which it referred. In the case of light, as we have seen, the sounds in 'glitter' or 'gleam' do not relate synesthetically to the referent associated with the Natural Class 'light', but to what the light is like -- to the various specific inflections of light implied by these words.


5.3 How the Proposed Word Structure Accounts for the Empirical Facts

Now that we have considered the various basic components of words semantics, let us consider how they interact to produce various phenomena related to those we have observed in these 14 experiments.

5.3.1 Phoneme Physics and Classification

If what I say is correct, then on the most fundamental level, a word is a reflection of its articulation, and that aspect of its semantics can be thought of as reflecting the physics of the mouth during articulation. The articulation of /r/ has enough kinetic force inherent in it to break the barrier of /b/, and the complex /br/ therefore appears in words in which the barrier is broken. The phoneme sequence /br/ appears disproportionately frequently in the Natural Class of breakage, and in the Natural Class of geometric form, this /br/ interaction manifests as 'brushiness'. The phoneme /l/, on the other hand, only has sufficient energy to make /b/ 'bulge' into a 'ball' shape. It does not break the /b/. The energies implicit in the two phonemes more or less cancel each other out, and /b/ and /l/ therefore appear in many words of literal and metaphorical 'balance' when they appear on opposite sides of an intervening vowel.

The physical forces and their reflection in word semantics can, in other words, be abstracted away from the Natural Class to which the word is assigned. It is the dynamic substrate which underlies the word. When the dynamic implicit in /br/ is revealed through the prism of the Natural Class of geometric form, it manifests as 'brushiness', whereas the dynamic force implicit in /bl/ manifests through the Natural Class of 'geometric form' as a 'bulge'. When /br/ manifests through the Natural Class of 'groups' of things, it is a type or division (brand, branch, breed), whereas /bl/ manifests as a collective (bloc, block, blend). When /br/ manifests through the Natural Class of 'growth', it manifests as reproduction (breed), whereas /bl/ manifests as blossoming and blooming. Through the Natural Class of impediments, /bl/ manifests as a blockage or emptiness (blur, blind, blot, bluff, blunt, blear, etc.) behind which one thing is retained, whereas /br/ manifests as a boundary between two things (brim, bridge, brink). Through the Natural Class of 'fire', /bl/ manifests as light, and /br/ as heat.


5.3.2 Phonosemantic Association and Iconism

Phonosemantic Association cannot violate the physics of Iconic semantics, but within the confines imposed by it, Clustering seeks to associate with each phoneme a unified conceptual whole. This is especially clear in semantic domains which cannot be reflected in articulation. For example, the phoneme /k/ forms a container of the mouth. So, for that matter, does /g/. The containers that cluster into /k/ tend to have a cover and/or fastener and they tend to be for human use. Those that cluster into /g/ tend to be natural formations and they tend to have no cover. Quite generally, voiced consonants tend to cluster toward the natural world -- physical processes, natural phenomena, etc., and unvoiced consonants tend to cluster toward human purposes and designs. As far as I can see, these distinctions have no basis in articulation. The mouth is no more closed and locked during the articulation of /k/ than the articulation of /g/. Surely natural phenomena cannot be associated with voicing on any other basis than a conceptual one. This cannot be an essentially Iconic association. But it might be a metaphoric connection. I can imagine that people subjectively feel that voiced sounds are somehow more earthy, forceful and natural than unvoiced sounds.

That all of these Clusterings are still subject to the limitations imposed by Iconism can also be seen in the fact that the Clustered words also fall into other phonesthemes. The 'plant babies' which Cluster into /b/ are not little powdery spores or spiky pine cones, but those that bulge just as the phoneme /b/ itself does: bulb, blossom, bloom, peach blow, etc. And the /b/ paraphernalia associated with babies works this way also. A 'bib' is a barrier. A 'burp' is an explosion. A 'bubble' is a bulge. And so forth. These are domains which are all Iconically related to /b/ as well.


5.3.3 Phonosemantic Association and Natural Classes

Consider the semantics of two similar phonemes. The pronunciation of the phoneme /p/ differs from that of /b/ only in voicing; yet their semantic worlds are quite distinct. An unvoiced sound has more precision than its voiced counterpart. It is not as heavy. Similar to what we saw for /l/ and /r/ in Experiment 8, most phonesthemes for /b/ have a corresponding phonestheme for /p/, but the corresponding phonesthemes also differ in some respect. The result in /p/ is not an explosion, but a more precise 'placement'. The barrier in /b/ manifests as parting, separation and selection in /p/. The bulge in /b/ has a corresponding precise 'point' in /p/, which can be spread into a 'plane' when followed by /l/ or /æ/. Like /b/, /p/ has a labial 'bias' (see experiment 6 -- the Bias in the Labials), although it takes a different form.

When we compare the two very closely related phonemes /b/ and /p/, a number of patterns emerge which are difficult to correlate only with their respective pronunciations. Where /b/ has balloons and bulges, /p/ has pebbles and peaks? Where /b/ has boards and branches, /p/ has pikes and prongs. Where /b/ has beating, /p/ has mostly pricking.

/b/ phonesthemes include:

/b/ Phonesthemes
Round Things -- bagel, ball, balloon, bangle, bead, bell, belly, berry, bladder, blimp, blip, blister, blob, blotch, bobbin, bowl, bracelet, bulb, bulge, button
Bumps -- balls, barrow, bay, bead, belly, blip, blister, bloat, blob, boil, boll, bolster, boob, booby, bosom, boss, breast, bubble, buckle, bud, bug(eye), bulge, bum, bump, bun, bunch, bunion, buns, burl, bust, butt, butte
Boards -- balk, bar, batten, bead(window), beam, billet, bloom, board, boom, brace
Hit -- bandy, barge, bash, baste, bat, bate, batter, bean, beat, belt, biff, blow, bludgeon, bob, bolt, bomb, bombard, bonk, boot, bop, bounce, brain, brake, bray, breeze, bruise, brush, buck, budge, buff, buffet, bump, bung, bunker(golf), bunt, bust, butcher, butt

/p/ phonesthemes include:

/p/ Phonesthemes
Point -- pastille, pea, pearl, pebble, pecan, pellet, penny, pence, period, pill, pimento, pimple, pip, pit, pixel, pock, point, pore, port, pox, prick, puck, pupil
Peak -- pass, peak, pedestal, perch, pike, pile, pinnacle, point, pole, pyramid, pyre
Prong -- paddle, pale, pawl, peg, perch, pick, picket, pike, pile, pin, piton, pivot, pock, poker, pole, post, probe, prod, prong, prow
Pierce -- peck, peg, pick, pierce, pike, pin, pinch, pink, pitch, plant, plug, poke, prick, prickle, probe, prod, prong, prop, punch, put

The /p/ words in each case tend to be sharper, and often harder. The phoneme /p/ tends to be smaller than /b/, less violent and more precise. And on the whole, the phonesthemes can be seen to reflect the articulation of the labial stop. The /b/ and /p/ classes fall into common Natural Classes like this:

Round Things: blip, point
Elevated: bump, peak
Sticks: board/branch, prong
Violent Physical Contact: beat, pierce

Within each individual phoneme, however, the corresponding phonesthemes cluster together differently. That is, points in /p/ are in a common semantic class with blips and bubbles in /b/, and prongs in /p/ are in a common semantic class with boards and branches in /b/. But prongs and points in /p/ are also in a semantic class together, because prongs have points. Boards and branches, on the other hand, in general share no class with the bulges and bubbles of /b/.

When one looks at how the more narrowly defined phonesthemes fit into superclasses, they cluster like this:

/p/ Clustering: {points, prong}, {prong, pierce}, {point, pierce}, {peak, prong}
/b/ Clustering: {bubble, bump}, {beat, branch}

It seems to me that this curious array of facts can be described fairly well in terms of the notions presented in this dissertation. Because /b/ and /p/ have similar articulations, we would expect them to have similar semantics, assuming that Iconism is active in language. If it is not, then this array of facts is indeed mysterious. The fact that these /b/ and /p/ words fall into identifiable classes which can be compared in this manner is evidence for the psychological reality of Natural Classes. But the differences in the Clustering dynamic as described in this last little table cannot, it seems to me, be accounted for by Iconism and Natural Classes alone -- we need the additional notion of Clustering to explain this... the tendency of the mind to try to get each phoneme to signify a unified conceptual space. The Clustering dynamic such as that described in this little table is very similar to the athematic metaphoric interrelationships between words sentences described by Rhodes and Lawler(1981). The difference is that these are partial lexical entries for phonemes rather than for words.


5.3.4 Iconic Meaning and Syntagmatic Context

Inclusion of a linguistic form in the context of other similar forms limits the semantic range of that form. For example, when one puts the word 'take' in the context 'take up', only one part of its semantic potential is being made use of. If one puts it in the context 'take over', then a different part of its semantic potential is being highlighted.

This happens also on the phoneme level. For example, 'drown' and 'drip' emphasize the downwardness and wateriness in /d/, whereas 'dim' and 'daunt' emphasizes its 'diminishing'. Since not all aspects of a phoneme meaning are equally salient in every word, we have to look at all the words to become familiar with the meaning of the phoneme. Similarly, we have to look at all the possible contexts (senses) of a word to get a feel for its Iconic meaning.

This tendency for a higher level to fracture a single thing into many can be seen especially in the many opposites that one perceives among the phonesthemes. As we have mentioned, it is extremely common to find a concept and its direct opposite heavily represented within the phonesthemes for a single phoneme. (I'm being careful to use the word 'opposite' rather than 'antonym', which has a narrower definition than I intend.) Examples of opposite phonesthemes in /h/ include Heaven/Hell, high/hole, help/harm,... In /n/ we find now/never, nice/nasty, noon/night, etc. We also pointed out that the opposite of a word is very similar to it semantically. When one looks at phonemes, one looks through the perspective of morphemes or in this case words. The higher level is like a prism that fractures the underlying unified semantics. What was one thing, like 'length' at the phoneme level, looks like two opposite things 'long' and 'short' from the perspective of the morpheme.

Notice that placing a word in a context imposes on it a limited function. A dictionary sense is nothing more than a heuristic description of a range of related functions that this word is commonly used for. In fact, every novel context (phrase or sentence) which a word appears in defines for it a new sense. Every context is a function and every function is a sense. Reference is closely related to function. What determines what a word refers to is how the word must be used.

Very informally, I'd like to draw an analogy between a word and a person, because I think it clarifies how I think about this. On one level, a person's body just is what it is -- tall or short, fat or thin, strong or weak. On another level, a person can give her body a function as a musician or linguist or mother or basketball player. These functions, like the functions of a word are not intrinsic to the body type. They are arbitrary, taken up for a time and perhaps laid aside for a time. The person may have several functions, and her tendency will be to Cluster -- to draw on her music when she plays basketball or does linguistics. These professions are what this person's body is used for at a given place or time. They define how she is interconnected in the larger scheme of things. It's the same with a word. The phonological structure of a word are analogous to the body of a person. It has certain predispositions already built in. It is strong or weak, fat or thin. Those predispositions in part determine what the person or word will do well. Once a person takes on a job, that natural predisposition will flavor the way that the person does the job -- whether she tends to be quick and effective or slow and thoughtful, outgoing or withdrawn. Similarly, when a word 'glimmer' takes on a job as a 'light' word, it does its job a certain way. If the job that a person does is very limited, like flipping burgers, then the person's individuality will not be as obvious in performing that role. In the same way, concrete reference blocks the salience of Iconic meaning. The fact that a person flips burgers, however, does not in any way diminish their inherent nature or their potential to express themselves creatively in other realms. Similarly, it's not the case that some words are inherently more Iconic than others. The extent to which their Iconism is visible is simply dependent on the function they are fulfilling.

So a person's job expresses only one facet of their personality, which doesn't cease to exist just because they have this job. Similarly, when one puts a word in a sentence -- one gives it a job -- only one facet of its meaning gets illuminated, but that doesn't mean its native form ceases to exist or loses its potential to affect things Iconically. And similarly a phoneme when used within the context of one Natural Class exhibits only one facet of its potential, but doesn't thereby lose the potential.


5.3.5 Senses and Phonesthemes

The result of this interplay between Iconic meaning and the other aspects of semantics, including classification and reference is to fracture what was one thing -- the Iconic meaning -- into lots of separate things -- the word senses. The word's various senses arise as a result of this process, and that's why we have referred to them as epiphenomena throughout this dissertation. As a result of this fracturing, it becomes difficult to perceive the original whole which holds everything together. Reference does this by taking a possible context for a word very seriously, and reifying a contextual meaning. It makes static what was fluid by limiting it to a context.

Phonesthemes are very similar to word senses. Phonesthemes are not generally defined as a set of words which have a common element of phonological form and semantics. Rather they are defined as a phonological form and its corresponding semantics. And just as a single word has multiple senses, so a single consonant has multiple phonesthemes. Just as senses are the result of classifying the one underlying meaning of the word into the various contexts in which it can appear, so phonesthemes are the result of classifying the one underlying phoneme semantics into the various contexts within which it can appear. This is why I speak of phonesthemes also as epiphenomena.


5.3.6 Basic Words and Senses

Obviously on the level of the word, Semantic Association seeks to associate with a string of sounds a unified semantic whole. With few exceptions, the various senses of a word are interrelated by metaphoric extension, hyponymy and other semantic processes. There is an analogous process on the level of the phoneme. Semantic Association is sensitive to the most basic words of a language, as it is sensitive to the most basic senses of a word, and it conspires to expand on the semantics of that more basic word by developing words which both sound similar and have a similar meaning.

For example, /b/ by its Iconic nature implies pressure built up behind a barrier and then a rupture of that barrier. In English, that is metaphorically connected with birth, and the basic English words 'birth', 'bear' and 'baby' all begin with /b/. Around these words we find Clustering in the plant world of buds, blooms and blossoms, and in the animals world, we find disproportionately many words like 'cub', 'breed' and 'brood'. In addition to this, we find disproportionately many words associated idiomatically and peripherally with birth, babies and children: big with child, bib, burp, bubble, breaking water, not to mention a spectrum of words associated with beginnings: booting a computer, boarding a ship, breaking new ground, the brink of disaster, broaching a subject, etc. Many of these words also contain an /r/. In Russian, birth tends to be predominated more by /r/ and less by /b/ than in English, and the basic words for birth and child begin with /r/ in Russian, and often contain a /b/ as well.

We have observed that, on the level of what the word is (i.e. its form, not what it refers to), Iconism is fundamental, and everything else is built up on it. But on the level of how the word is used, the most basic sense of a word is fundamental, and all the other senses of the word are related back to it. For that reason, studies like McCune(1983) and Rhodes and Lawler(1981) place such emphasis on metaphor and the other semantic processes which interrelate word senses. From one perspective (that of langue), we might say that all word senses are equally important and all words are equally basic. A word either exists or it doesn't. A word either is used within the context of some Natural Class or it isn't. But from the perspective of a language user (parole), some words and some senses certainly are more basic than others. The most basic words and the most basic senses for words as we recognize them tend to have concrete reference.


5.4 Ramifications of Phonosemantics for Issues in Linguistic Theory

5.5.1 The Function of Language and Abstract Semantic Representations

The view of language I propose here in which only a part of semantics can be reduced to reference predisposes one to view language relativistically. From a non-relativistic perspective, the function of language is viewed in truth-theoretical or informational terms. From the relativistic perspective, the other functions of language are emphasized, beyond that of simply imparting information. To the extent that semantics is Iconic, it must be viewed not so much as a tool for stating facts but more as a medium within which speakers of a given language simply operate and interrelate without any particular purpose. To the extent that language is Iconic, semantics cannot be abstracted away from language itself and language as we know it cannot be abstracted away from man any more than music can be abstracted away from the notes which make it up. Therefore anyone who accepts that there is some element of language which is Iconic must accept that abstract semantic representations will never fully represent the meaning of a word. Some aspect of a language's meaning cannot be conceptualized or translated or abstracted away from.


5.5.2 Semantic Primitives

One of the interesting consequences of the Phonosemantic Hypothesis is that it provides us with one obvious set of what seem to me very logical semantic primitives -- the phonemes themselves. At least one set of primitives, in other words, seems to me rooted in the very form of language. But the Natural Classes also serve as primitives in a way which is much less tied into the form of language itself. The class 'animal' seems to me in English a primitive. There are many indications of this. For example, if a child asks me what a 'badger' is, I must answer, ÒIt is an animal.Ó I cannot answer, ÒIt's a brown object,Ó without being considered deceptive. The fact that the four criteria I have defined for a Natural Class do hold of some sets of words is another indication. Furthermore, it seems that at least some of these Natural Classes are part of a universal menu from which languages select. 'Animal' is a Natural Class in all the languages I know anything about, and probably is a Natural Class in most languages.

Much of the difficulty we have had in identifying semantic primitives, it seems to me, has lain in the fact that we have tried to analyze the entire semantics of a word like 'badger' in terms of Natural Class affiliation. Once we can think of phonemes as semantic primitives as well as Natural Classes, then we can analyze much more of the semantics of words satisfactorily.


5.5.3 Universals

Jakobson pointed out that most of Greenberg's universals have an Iconic quality about them. Iconism by its very nature must be universal. If two phonemes are pronounced identically in two different languages (which they of course rarely are), then they must have the same Iconic meaning. But to what degree does Phonosemantic Association also follow universal patterns? Several of the experiments presented herein suggest that the Clustering dynamic is at least in part universal.

In experiment 10, the string /s//t//r/ was shown to have very similar semantics cross-linguistically. It was, however, also true that the emphasis on the various phonesthemes varied from language to language. The word distribution in Experiment 10 came out as follows:

Albanian -- Struggle -- 5, Stretch/Spread -- 5, Strong -- 3, Stop -- 3, Straight -- 2, Start -- 1, Strange/Distant -- 1, Stroll -- 0, Strike -- 0
Catalan -- Stretch/Spread -- 16, Straight -- 14, Struggle -- 14, Strange/Distant -- 14, Stop -- 14, Strong -- 11, Strike/Tear -- 8, Stroll -- 6, Start -- 3
English -- Strong/Stern -- 18, Straight -- 17 , Stretch/Spread -- 13, Struggle -- 13, Stop -- 10, Strange/Distant -- 9, Stroll -- 7, Strike -- 5, Start -- 2
German -- Straight -- 33, Strong/Strict -- 15, Stop -- 14, Stretch/Spread -- 7, Struggle -- 5, Strike -- 5, Stroll -- 4, Strange/Distant -- 3, Start -- 3
Greek -- Strong -- 16, Stop -- 10, Struggle -- 9, Straight -- 7, Strike/Tear -- 7, Strange/Distant -- 5, Stroll -- 3, Stretch/Spread -- 3, Start -- 0
Hindi -- Stretch/Spread -- 5, Struggle -- 3, Straight -- 3, Strong -- 3, Start -- 1, Stop -- 1, Strange/Distant -- 1, Strike/Tear -- 1, Stroll -- 0
Indonesian -- Strong -- 10, Straight -- 4, Struggle -- 4, Stretch/Spread -- 4, Stop -- 2, Start -- 1, Strange/Distant -- 1, Strike/Tear -- 1, Stroll -- 0
Irish -- Struggle -- 30, Straight -- 27, Stop -- 23, Strong -- 23, Strange/Distant -- 17, Stretch/Spread -- 15, Strike/Tear -- 14, Stroll -- 5, Start -- 0
Lithuanian -- Straight -- 8, Struggle -- 2, Stop -- 2, Strong -- 2, Strike/Tear -- 1, Strange/Distant -- 1, Start -- 0, Stretch/Spread -- 0
Norwegian -- Straight -- 36 , Strong -- 26, Struggle -- 13, Stretch/Spread -- 10, Strike -- 8, Stop -- 7, Start -- 6, Strange -- 6, Stroll -- 2
Russian -- Straight -- 23, Strange/Distant -- 11, Strong/Strict -- 8, Struggle -- 5, Stop -- 4, Strike -- 3, Start -- 3, Stroll -- 2, Stretch/Spread -- 2
Welsh -- Straight -- 8, Stop -- 5, Strange/Distant -- 4, Struggle -- 3, Stretch/Spread -- 3, Strike/Tear -- 1, Strong -- 2, Start -- 0


5.5.4 A Possible Mechanism by which Sound Shifts Interact with Phonosemantics

Let us now return to our paradox. How might a quite general productive and synchronic correlation between phonological form and semantics be reconciled with Grimm's Law, or for that matter assimilation, or for that matter, the existence of different languages? We might begin to uncover this by looking once again at data such as that presented in Experiment 10. Celtic has diverged so much from Germanic that it's probably safe to assume that most of the /s//t//r/ words in Irish are not cognate with those of English, and yet the fit in the same Phonosemantic Classification:

English
Straight
-- stair, steer, straight, strait, strand, strap, straw, streak, stream, street, stretch, string, strip, stripe, strobe, stroke

Irish
Straight
-- starr (tooth, tusk, jut, rough pull, fit of anger, round of boxing, sturdy), starran (projection), steotar (sugar stick), storn (straddle pin), straibeir (lash), straic (strip of cloth, stroke of a cane, state, level, pride), straille (tall, lazy aimless person), straimead (tape, streamer, heavy stroke), straip (strap), stran (prominent tooth), strapa (strap), strat (stay between masts), streaclan (band, gaiter), strearac (tree creeper), strileaman (long, nervous person), strioc (stripe, repentance), striocail (making tracks, striving), striolla (girth, girdle), strior (impulse, gust, enthusiasm, stripe), strioradan (anything hanging, limp), striopan (strip, streamer), striopar (strip, tatter), stroc (iron keel band), stropa (strope), struic (crest, ridge), strup (curved spout), sutrog (candle)

I am suggesting that the English and corresponding Irish /s//t//r/ words have a common element of Iconic meaning, but that the English and Irish words differ on the level of Classification and Reference. One of the English words 'street' falls in the class of 'roads and paths'. There are no similar Irish words in this class. One Irish word, however, fits in the class of 'people' and no corresponding English words fit in this class or have anything remotely resembling this referent. All these words share is a similar tension and stretchiness, which I have done my best to show probably arises from the internal tension implicit in the articulation of the phoneme combination /s//t//r/.

The phonosemantic evidence suggests that when a word changes its pronunciation during diachronic sound shifts, its Iconic meaning also must change. This might be analogous to replacing all the employees in a company with others. The company has the same structure, but the people are different, and two things happen. The new employees gradually take on slightly altered assignments which are related to their original assignment, but which are more to their taste, and the company reorganizes a bit in deference to the new blood. The whole system, in other words, takes a little time to settle in. I expect something very similar happens after a language undergoes a major sound shift.

The Concrete Nouns would remain relatively unaffected -- and it may be observed that almost all examples of regular sound change provided to beginning linguistic students are Concrete Nouns. The reason for this is that of the most basic roots in a language, only they have really identifiable correlates in other languages. One can fairly unequivocally translate 'goat' or 'candle' into French and Russian and Thai. But what is the one right translation of 'mad' or 'swift' or 'twist' or 'wring' or indeed 'have' and 'to' and 'up' into these same languages? That cannot be answered so readily. That is because the Iconic dimension of their meaning holds much greater sway in these words. Their function within the language is determined to a much greater extent by their phonological form. Because these large scale phonological changes can only be demonstrated by comparing words in one language with their equivalents in another, the words compared are Concrete Nouns. The remaining words undergo a semantic shift as well as a phonological one. That shift, however, probably rarely involves assignment to a new referent and therefore reclassification in a new Natural Class. I would predict that it affects not what the word is, but what it is like.

If this shift is only allophonic, as is the case with Mid-Western /str/ vs. Texan /Str/, then I would predict that the sound change would only affect the feeling-tone of the word, but not particularly its Clustering dynamic. But if the shift involved an actual reorganization of the phonemes -- some former allophones which gain phonemic status, for example -- then I would anticipate much more visible semantic effects. In either case, I would anticipate as always relatively little change in the Concrete Nouns.

To test this hypothesis, let's take as an example again /str/ and /Str/ in Texan. I might try to test whether there is a tendency in dialects that used /Str/ rather than /str/ to tend to prefer metaphorical and idiomatic extensions of the word which by Semantic Association favored the semantics of /S/ rather than /s/. For example, the flatness in /S/ might start to predominate over the linearity in /s/. We might roughly divide the linear /str/ words as follows:

Thin: straight, strand(hair), straw, string, strobe
Flat: strait, strand(shore), strap, stream, street, strip, stripe
Either: streak, stretch, stroke

To verify this hypothesis, we might look in dialects like Texan, that pronounce all these words /Str/ whether they prefer the flat words to the 'stringy' ones. Might they be more inclined to say 'go right home' much more frequently than the 'go straight home' since /S/ is flatter and 'straight' is not? Perhaps one would find that 'twine' was preferred over 'string' and 'laser' over 'strobe'?

The Concrete Nouns are anchored by their referents in any case, and those are the words which can most reliably be compared. It seems to me therefore possible that when a language undergoes a diachronic sound change, the words other than the Concrete Nouns would gradually be replaced by synonyms which are more suited to the phonosemantics of the language and that the usage of those words that remained would also shift to accommodate the language's phonosemantics. The fact of a large scale sound change in a language seems to me a strange fact in itself. It's hard to imagine that one's mother tongue could undergo such major changes in a relatively short period of time, and yet it happens. Is it that much stranger to imagine that the semantics of words other than the Concrete Nouns shifts with it?


5.5.5 Resolution to the Cratelean Paradox

Let me now return to the original Cratylean dilemma and state explicitly what the data presented here has to say about it. Very briefly, Cratylus' position is that a word cannot in principle be unfitting to its function. The original name-maker could not have made a mistake. However, in the dialogue it is pointed out that Hermogenes himself is a poor man -- something analogous to a poor man being named 'Mr. Rich'. Furthermore, after some discussion, /r/ is determined to mean motion, yet the Greek root for 'motion' (kinesis) does not contain an /r/. Hermogenes' position is that the sign is arbitrary and should be arbitrary, for only then can it truly represent things rightly. And Socrates concludes that there is, after all, an underlying mimetic principle, but that the original 'name-giver' does indeed at times sadly make mistakes.

I think the data arrived at by the methods presented in this dissertation shows that all three men are correct -- they are only holding different parts of the elephant. Cratylus is right that the original name-giver cannot make a mistake, if we think of the name-giver not as a person who consciously invented language in the past, but as a natural Iconic force active in the present. Hermogenes is correct that reference is and must be arbitrary or conventional. If it were not, there would be no concreteness of reference, language would lose its grip on the world, and we would no longer be able to talk about anything. And Socrates is right that words are not always fitting... if we accept that the fittingness of words is a function of parole (speaking) rather than of langue (the grammar itself). Analogously we might say that no species is inherently superior to any other in any absolute way, but some species do fly better than others and some species do talk better than others. So some words are better suited to certain contexts than others. However, the iconic effect cannot in and of itself be wrong or right, because it is simply a natural force. It cannot make a mistake.

Put another way, there is no right or wrong referent for a particular string of sounds. Reference is arbitrary and all choices of referent are equally good, equally true. However, having first chosen a referent for a word, the sound will then necessarily affect the connotation of the word, and it will affect the clustering dynamic of the language. These processes are determined by natural law and are subconscious. They cannot in principle be wrong, any more than an object can wrongly reflect the natural law F=ma. Force just does equal mass times acceleration. Similarly, there's no point discussing whether it's right or wrong that words for light containing an /r/ be harsh, because it simply is that way whether we like it or not. These are the facts as I see them from the perspective of langue. There is no right or wrong about the arbitrary choices of referent, because they are truly arbitrary, and there is no right or wrong about the natural processes of clustering and true iconism, because there couldn't in principle be a choice. Therefore although sound does affect meaning, it does not in my view follow that there could be better or worse languages.

However, on the level of 'parole', there is such a thing as 'the right word' for a context. For one thing, there is such a thing as lying. Lying is a consciously chosen inappropriate referent made with the intention to deceive. But there are also other forms of failure to choose 'the right word'. One can, for example, choose a word whose referent is fitting, but which is phonosemantically not appropriate to the context. Then one is not lying. One is not stating something counterfactual. The facts are perhaps true. The referents are perhaps all viable, and the sentences all have positive truth value. Butthe hearer's subconscious is manipulated into into buying a product or accepting a particular point of view by skewing the phonosemantics. This is done in propaganda and advertising all the time.

Notice that my proposed resolution to the Cratylean paradox could not be arrived at without having established certain theoretical preliminaries. For example, Socrates, Hermogenes and Cratylus all assume that the way we name things is by consciously constructing a grammar. To some extent it is true that we consciously choose the referents for words. But part of the process is also unconscious. For the most part, the grammar of a language is built up through using it, through parole. Specifically, reference is largely conscious; classification is semi-conscious; clustering is subconscious, but can be brought to consciousness with some effort; and true iconism is subconscious and takes some real work to see clearly. So as prerequisite to my solution to the Cratylean paradoxes, I had to have the notion, most fully expounded upon in the generative tradition, of language as a natural process which is largely unconscious and whose structures have to be brought to consciousness by empirical methods.

Furthermore, the above resolution to the paradoxes could not have been formulated without reference to the Saussurian notion of langue vs. parole. Hermogenes and Cratylus are both correct from the perspective of langue, and Socrates is correct from the perspective of parole.

Is it coincidence that perhaps the two greatest promoters of the notion of 'arbitrariness of the sign' among linguists also provided fundamental theoretical constructs required to formulate my proposed resolution to the Cratylean paradox? I think not. It's very common in the history of science that both extremes of an issue have to be taken on fully and explored on their own terms before they can be resolved into a sensible whole.


5.6 Future Research

There are, of course, many other experiments one could devise to test for the Phonosemantic Hypothesis, Clustering, Iconism and the tendency of reference to obscure the phonosemantic effects. Hopefully future research will lead us to conduct similar experiments for an ever wider variety of languages and to compare the results cross-linguistically. This study has also concerned itself primarily consonants rather than vowels, monosyllables rather than polysyllables and with English much more than with other languages. Much research remains to be conducted on the relationship between morphology or syntax and phonosemantics. English idioms are a very fruitful domain for phonosemantic research. The treatment of the semantics of phonetic features in this work could be expanded on greatly. Furthermore, this field bears an obvious relationship to the fields of etymology and language origins, not to mention lexicography, cybernetic processing of language, language teaching and any number of other practical applications.

I find that phonosemantics in no way differs from other deep areas of human inquiry -- the more deeply one investigates it, the wider its horizons prove to be. I have come out of these investigations with a firm conviction that investigations into phonosemantics have been given very short shrift over the centuries for reasons that have nothing to do with the field's importance to our understanding of basic human concerns.

There is also a vast amount of research yet to be done into the nature and structure of Natural Classes. 'Grammaticality' can clearly be applied to classification schemes, since some classifications are grammatical and others are not. Reference is also related to at least Functional Natural Classes in the sense that all elements in a Functional Class have a common element of reference. But the members of the phonesthemes are not always related by a common referent. If one does define a class such that all the elements have a common referent, then one can see Iconism functioning. But there is much more to this structure than I have discerned. There is clearly a complex relationship between the Semantic Relations (antonymy, meronymy, hyponymy, etc.) and the Natural Classes, but much of that remains fuzzy. The nature and structure of Natural Classes and their relationship with the Semantic Relations serves as the primary subject of my current research.

In order to understand many of the phenomena brought up by these phonosemantic experiments, one has to distinguish Phonosemantic Association from Iconism. Some aspects of Iconism are clearly blind to reference and some aspects of it are not. In this I disagree with Von Humboldt who saw these two aspects of Iconism as completely unrelated. If they were unrelated, then a Phonosemantic Classification for given phoneme would, for example. not resemble its articulation. But the exact nature of the relationship between Phonosemantic Association and Iconism still requires further work.


5.7 Concluding Remarks

There has been a rising tide of interest in linguistic Iconism in recent years. The first sound symbolism conference ever was held in 1993. In 1998, the Linguistic Iconism Association was formed, and it now has nearly 300 members, many of whom have become interested in the phenomenon only recently. The Internet has also made sound symbolism much more visible. Many articles which could not previously find publication are now generally available.

This dissertation is a contribution to this dialogue, my attempt to help give voice to a perspective on linguistic science has not held sway in the mainstream for many years. I anticipate that the reader may not agree with all the thoughts I offer, but whatever position the reader may hold, I believe I have presented quite strong evidence that the phonosemantic perspective on language deserves continued consideration.