REFERENCE AND REPRESENTATION

Think of the following assortment of activities:

What do these activities have in common, and how do they differ?

Each activity points something out; it refers— that is, it separates part of one’s environment or experience from its background. But here the similarity ends, for the activities we have listed exemplify different types of referring.

The most fundamental (though not always the most practical) means of referring is to point up or present the object or experience itself. I hold up a ball for a child or point to the monkey at the zoo. Presentation of objects themselves would seem an efficient and unambiguous way to single something out. The efficacy of such a scheme was appreciated by the savants in Jonathan Swift’s Academy of Lagado, where Lemuel Gulliver encountered a unique linguistic experiment, a “Scheme for entirely abolishing all Words whatsoever” (1934: 182). Conversations were to be conducted not by using words, but by carrying about the objects one wished to signify:

Unfortunately, the seeming simplicity of communication through presentation collapses under the slightest scrutiny. To begin with, presentation is highly inefficient. As Gulliver discovered, the need to carry about one’s referents is cumbersome at best:

In other cases, communication by presentation becomes impossible. How does one fit New York City or the Atlantic Ocean into one’s sack? There are even some topics of reference for which no object corresponds, including movements (running), abstractions (happiness), and objects of our imagination (unicorns).

And the problem with presentation goes deeper. Mere presentation —holding the ball, pointing to the monkey, or even saying “What’s that?” in the presence of an object of reference—may still yield extreme ambiguity. Nida reminds us that when a missionary trying to learn a native language points to a tree with his finger and says, ‘What is that?’, “the reply is more often than not, Tour finger” (1954:223). An example of such misunderstanding of intended referent involved Captain Cook:

Tangible contact with the thing itself does not ensure transparency of semantic intent. As foreign travelers may learn, standing in a moving bus, tapping one’s watch, and pointing backwards may not convey the intended question, “What time does this bus return to the city?” Physically extracting an item from its environment (e.g., holding up a ball) may still leave us with the problem of distinguishing between a class and an individual, since we do not know whether the thing being presented is to be understood as the thing itself or as an example of the class to which it belongs.

We avoid some of these problems in referring when we replace presentation with representation— that is, when we use something else to stand in place of that to which we wish to refer. We imitate the behaviors of others (Miss California’s walk), we draw pictures of things (a cat), we tie strings around our fingers—and we use language.

Taken literally, a representation is a “re-presenting” or a “giving again.” In actual use, though, that re-presenting cannot be of the thing itself. That which is represented “must be made present indirectly, through an intermediary; it must be made present in some sense, while nevertheless remaining literally absent (Pitkin, 1969:16). Picking up a stone when one wishes to refer to that particular stone is an act of presentation, but picking up the same stone and talking about it directly or using it to refer to the general class of stones or to the Rock of Gibraltar is an act of representation.

The advantages of representation over presentation are obvious. Representations enable us to refer to things (events, properties, objects) which are not present—either because they are somewhere else (“the conductor in the next train compartment”), do not now exist (“the first bicycle I ever owned”), or never did exist (“the ghost I imagined”). Of course, representations can be used in the presence of the things they represent. I can hold a photograph of my dog next to his head to compare the likeness, and I can ask you to “bring me the chair that’s in the corner,” using language to save myself the trouble of getting up and pointing to the object physically (presentation). Language used in the presence of its referent I shall call direct reference, while language used to refer to that which is absent I shall call indirect reference. The use of language for indirect reference (what Charles Hockett calls displacement—see chapter 3) has long been recognized as one of the properties distinguishing human speech from nearly all other communication schemes in animals. We have tended to overlook the fact that the entire modality of written language is predicated upon the use of language to refer indirectly (see chapter 6).

LEVELS AND FORMS OF REPRESENTATION: A MODEL

However one wishes to characterize the essential properties of human beings (e.g., man the speaker, man the tool maker, man the thinker), a fundamental part of being human is the ability to represent one’s environment, be it real or imagined.¹ Therefore our primary “something” to be represented will be human experience. The next question is, of course, How is experience to be represented? Which pieces of it are to be singled out and through which medium? The answers are potentially infinite, but I shall consider only five illustrative domains of representation: language, science, cartography, art, and politics.²

The domains of experience represented in these five areas are neither coterminous nor predetermined. Language has, in principle, the entire range of experience as its domain, although, as anthropological linguists have shown, not all societies choose to encode the same distinctions of experience (the standard example here being the Eskimos’ many terms referring to snow). Similarly, science—itself a specialized linguistic representation of reality—has the whole physical universe as its potential domain but has only exploited restricted portions of it during different periods. At the other end of the spectrum, maps, which were originally used for charting physical contours and man-made constructs (e.g., roads and cities) on the earth, have been extended to represent the heavens, the structure of the human brain, and many other realms (Robinson and Petchenik, 1976).

The range of phenomena in these kinds of representations and the means chosen for making such renditions are determined by the function the representations are meant to serve. If, for example, I am in Athens and wish to get from Syntagma Square to the Parthenon (but understand no Greek), the most helpful service a friendly bystander can perform is to draw me a map (this assumes that I understand how to read a map). The map will not need to include landmarks I shall not be passing, although a limited degree of redundancy may be useful (e.g., indicating both street names and major business establishments along the route). Similar examples which predict mode and range of representation can be given for each of the other domains as well.

I shall refer to these direct representations of human experience (by whatever mode) as “first-order representations.” Each of these first order representations can itself be represented by second-order representations. The second-order representations in these domains are: linguistic theories, models of scientific theories, theories of maps, theories of art, and theories of political representation (we shall come to the choice of terminology shortly). And again, through third-order representation, many—but significantly not all—of these second-order representations can themselves be represented, yielding models of linguistic theories, models of theories of maps, models of theories of art, and models of theories of political representation. As with first-order representation, the range of phenomena represented in second- and third- order representation is determined by the function of the representation of the higher order. Figure 2.1 summarizes the levels and forms of representation we have discussed so far.

The model illustrates that the process and products of representation are not restricted to languages and linguistics. Rather, the more general act of representing is something in which languages and linguistics participate. This not to say that the kinds of representation used in different domains (e.g., language and map making) or on different levels (e.g., theory making and model making) are wholly parallel. In fact, it has been argued that analogies between, for example, maps and languages, or maps and scientific theories may be highly misleading (Robinson and Petchenik, 1976).

Fig. 2.1 Levels and Sample Forms of Representation

Moreover, the model itself generates a number of questions about items it contains. To begin with, I have used the terms theory and model without having defined them. Chao identifies thirty-nine different ways in which the term model has been used, “some of which have exactly opposite meanings” (1968:202). Similar counts can surely be made for the term theory. Both theories and models are representations, but how do they differ?

For purposes of discussion,³ let us define a theory as a statement or set of statements that makes a generalization about a set of data such that all the current data are accounted for and future data can be predicted accurately. A critical feature of a theory is that it can be expressed verbally. Let us then define a model as a visual representation of a theory that encodes the significant generalizations made by the theory, but whose comprehension may depend on a prior understanding of the theory.⁴

In figure 2.1, I have encoded these definitions (of theories representing reality and models representing theories) by considering theories to be second-order representations and models to be third-order representations; and yet in a distinct sense the model itself is a representation either of the first order representation or of reality. For, if the theory has made an accurate representation of X, and the model is an accurate representation of the theory, then, by transitivity, the model should be an accurate representation—however abstract or distant—of X.

The one enigma in our otherwise symmetrical characterization of representation, however, is the status of scientific theories and models of scientific theories. In each of the other four examples, the first-order representation is a phenomenon in which ordinary members of a society might participate—a language, a map, a work of art, or a political system. In the case of science, however, the first-order representation is a theory—that is, a theoretical statement or statements used to talk about reality, but generally of interest to only a highly restricted portion of the population. The asymmetry continues in the second-order representation, which, in the case of science, is model making, while in the other instances, it is theory making. Finally, there is no third- order representation for science, while for the other four types of representation, models are the third-order of representation.

Is our characterization correct, and if so, does the asymmetry matter? Linguistic theorists, from the Neogrammarians (see Jankowsky, 1972), through the American structuralists (see Bloomfield, 1939), down to transformational grammarians (see Chomsky, 1968) have repeatedly stressed the parallels between physical science and linguistic science, and, especially in the case of Chomsky, the parallel ways in which theories and models are constructed and tested. If representation in language and linguistics is not comparable to representation in science, then the entire question of goals, methodology, and evaluation in linguistics will need to be rethought.

The crux of the problem is whether theories are actually direct representations of reality, or whether, as with language and maps, there is another first-order representation which mediates between reality and theory. One might argue, for example, that just as linguistic theory is a theory about language, scientific theories are dependent upon linguistic representation as a way of characterizing data. This observation is true—but it overlooks a critical distinction. While the task of linguistic theory is to characterize human language as it exists, the goal of scientific theory is to see beyond language to the phenomena themselves, and where necessary, to change language to suit its needs.⁵ Except where new phenomena are introduced into their environments (lunar space probes, hula hoops), natural language users rarely need to alter their language to better characterize reality. Attempts are sometimes made by children and foreigners, but typically it is the neophyte who adapts to the existing language, rather than vice versa.⁶

Of what significance is this discrepancy in orders of representation between science on the one hand and language and linguistics on the other? The answer is not clear, given our very incomplete understanding of representation, theories, and models in a single modality. However, before we assume comparability of terminology across domains (e.g., model in science and model in linguistics), we should be aware that the terms may not be comparable at all.

PROBLEMS OF LINGUISTIC REPRESENTATION

In constructing a representation, we use one thing to stand in place of another. In the case of human language, the type of representation most interesting to us, that place holder is typically a word (be it written, spoken, or signed). As stated in chapter 1, Saussure spoke of the representation as the signifiant (the “signifying”). That which was being represented was the signifié (the “signified”). Taken together, the signifiant and signifié form the linguistic sign.

Less clear, though, is what Saussure’s signifié actually refers to. According to Saussure, the signifié belongs to the realm of ideas, not to the realm of the real world: “The linguistic sign unites, not a thing and a name, but a concept and a sound-image” (1959:66). This decision to make the referent an idea rather than a thing⁷ has clear pragmatic advantage. Since much of what we use our language to talk about either never existed (unicorns, Martians) or would be impossible to identify in the objective world (dreams, constitutionality, or the prototype which the class term chair labels), we would not be able to use language to talk about intangibles if the referents of our linguistic signs were necessarily items of experience.

Yet the retreat from reality to ideation has its problems as well; we still have to figure out what our thoughts are representing. Unless we take all of our thoughts to be innate, we still must find a source for our mental referents in order to understand what we are talking about when we use language. The question is important to the linguist, but may also be vividly real to the average language user. R. W. Apple, Jr., described why the military in Iran, once loyal to the Shah and then to Prime Minister Shahpur Bakhtiar, switched their loyalty to the Ayatollah Khomeini: “most of the soldiers and many of their officers . . . ultimately refused to kill and be killed for abstractions called ‘constitutionalism’ and ‘legitimacy’” (New York Times, 12 February 1979, p. Ai).

In talking about what is being represented (or signified), I shall speak of experience rather than ideas—with full recognition that I shall have little to say about the referents of words such as truth, lopsided, or confusion. Placing the study of referents in the realm of ideas may not, however, help here. Besides, since thinking is grounded in experience, it seems sensible to get to the heart of the matter straightaway.

THE CHOICE OF A SIGNIFIÉ

We all know that societies differ in the ways they divide continuous phenomena in nature (and in their linguistic labels for these divisions). A favorite example here is the color spectrum. The Shona (in Rhodesia) do not distinguish linguistically between what speakers of English called red, orange, and purple. On the other hand, the Shona linguistically divide up the spectral band which English speakers generally call green (Krauss, 1968). The same kinds of discrepancies in labeling occur when naming parts of the human body. Old Norse hond can refer to the entire arm (from fingers to shoulder), while English hand refers only to the area from fingers to wrist.

The choice of which span of experience is to be represented by a linguistic sign may vary not only from language to language but also from modality to modality. In English, if I want to convey the message that I am looking at you, I need to say something like “I am looking at you”—a multiword utterance which distinctly specifies an agent, an activity, and an object. In visual representation systems, though, such individuated, linear representation may not be necessary. American Sign Language, for example, would encode the message with a single sign: a V hand shape (formed with the index and middle fingers) moving outward from the addressers eyes to the individual being seen. Or consider the fact that some verbs incorporate locative markers within the verbs themselves, while others must be paired with lexically distinct prepositions (Old English onwacnian vs. Modern English wake up; Modern English pare vs. file down).

The American descriptivist tradition of linguistics has prepared linguists to expect stark variation among linguistic communities in the ways in which language will be used to label the world. Recall Joos’s fabled pronouncement that “languages can differ from each other without limit and in unpredictable ways” (1957:96). There have, however, been contrary tendencies, especially in the semantic theories growing out of Chomsky’s rationalist position (e.g., Katz and Fodor, 1963; Fillmore, 1968a; Schank, 1972), but also deriving from psycholinguists studies (e.g., Rosch, 1973). The increasing reversion to an innatist stance with respect to the individual’s classification of experience is strongly reminiscent of the universal language movements that flourished in the seventeenth and early eighteenth centuries. Universal language schemes and their implications for the study of linguistic representation will be examined in more detail later on in this chapter.

THE CHOICE OF A SIGNIFIANT: ARBITRARINESS AND ICONICITY

A different set of problems arises when we look at what kinds of linguistic “place holders” are available to represent our experience. The central issue appears to be the existence of any recognizable likeness between that which is being represented (the signifié) and that which is doing the representing (the signifiant). Although the question of the relationship between the signifié and the signifiant traces at least back to Plato’s Cratylus, it was Saussure who enunciated the modern dictum that “the bond between the signifier and the signified is arbitrary”:

Saussure is careful to consider possible objections to his principle, namely, onomatopoeia and interjections. He concludes, however, that such words are few, that they often differ from language to language (dogs go bow-wow in English, but ouaoua in French), and that their symbolic origin is often in doubt (ibid., pp.69-70).

The dictum of the arbitrariness of the linguistic sign has so pervaded linguistic analysis that (by unwritten statute) any linguistic discussion must reaffirm this dictum whenever the data themselves are at all suggestive of iconic sources of signs (i.e., signs looking like that which they signify). The reductio ad absurdum of this position is a comment by John Waterman that Plato’s discussion of the physis position in the nomos (“convention”)/physis (“nature”) controversy over the origin of language was “incredibly naïve”:

Anyone who believes that Plato was lacking in humor, especially in the Cratylus, either has not read the text, or is already so committed to the principle of the arbitrariness of the linguistic sign that any attempt at counterargument is automatically interpreted as “naïve.”

Less dramatic but equally strained are discussions of the linguistic bona fides of visual forms of communication. As shown in chapter 6, the insistence upon a phonetic base for writing leads to an arbitrary definition of language, which produces some undesirable consequences. Similarly, attempts to exaggerate the arbitrary pairings of signifier and signified in sign language leads to an incomplete understanding of ephemeral visual language (see chapter 7).

FILTERS BETWEEN EXPERIENCE AND LANGUAGE

I have been considering problems which may arise in using language to represent our experiences. In chapter 1, in a different context (“What Do Users of a ‘Common Language’ Hold in Common?”), we observed that linguistic signs must be shared if we are to speak of a “language” or of a “linguistic community.” Unfortunately, agreement between individuals is often lacking. Moreover, entire language communities may differ markedly in the ways they use language to divide their experience (consider Shona color terms). And yet there are those who maintain the extreme position that reality can be divided into a universal set of experiential—and linguistic—categories.

How might we hope to make sense out of these various perspectives? Recall the delineation, in chapter 1, of four variables of human linguistic communication: the individual, linguistic structure, experience, and society (Figure 1.1). By reorganizing and expanding our diagram, the relationship between experience and language can be viewed as potentially mediated by “filters” which regulate what portions of the objective world are linguistically encoded by the individual and the language community. Figure 2.2 summarizes the possible filters between language and experience.

Fig. 2.2 Filters between Experience and Language

I shall examine each variable in turn, illustrating the effects each filter might have upon the formulation of a linguistic representation (either lexical or syntactic). In doing so, though, there is the risk of opening a Pandora’s Box of problems which have troubled philosophers and psychologists for centuries. The limits of the present volume will not permit a thorough examination of the traditional arguments concerning the relationship between perception, cognition, and language. I can, however, indicate where the problem of linguistic representation intersects with inquiries made by students of perception and cognition.

No Filters

The simplest view of linguistic representation assumes that the human mind directly perceives what is “out there.” This form of realism holds that sense data are received and processed with no interference from biological, cultural, or experiental presuppositions. Contemporary studies in the philosophy of science have rejected this position, arguing instead that any observation (be it the everyday occurrence of looking at a tree or an observation in a scientific experiment) is theory-laden. That is, it is impossible for a person to be objective about what he sees. Rather, our past history (both individual and social) shapes our perceptions, which, in turn, may condition our linguistic representations of those perceptions.

Norwood Russell Hanson has us

In our own use of ordinary language, we reflect persisting dissonance between our common-sense perceptions and Keplerian beliefs. We are intellectually capable of referring to the earth revolving around the sun, yet in our everyday speech, we talk of the sun rising and setting.

Somewhat closer to home, we can ask whether humans must “learn” to order their perceptions, or whether our sensory apparatus is fully functioning at birth. If the former is the case, we must inquire further whether all humans develop their sensory abilities in the same way, regardless of individual or cultural differences (much as the human skull hardens and the vocal tract changes shape during the early months of life)—or, alternatively, whether environmental differences can produce differential perceptual abilities?

There is now a vast amount of data on human perceptual abilities, especially on the development of visual perception in young infants. Many of these studies suggest that people must “learn” to see, although how much is learned (and when) is still a matter of intense debate.⁸ There are, nevertheless, clear examples which demonstrate the role of experience in directing—and selecting—what we are able to see.

One such case is described by R.L. Gregory (1966; Gregory and Wallace, 1963). A man (called S.B.) who was blind from birth regained his sight at the age of 52. Gregory wondered what effect the lack of visual experience would have on S.B.’s visual perceptual abilities. His study of S.B. indicates that experience (or lack thereof) played a significant role in S.B.’s perceptual judgments, as illustrated by the following three examples.

Just after regaining sight, S.B. could not make out the world of objects with the same degree of clarity that was later to develop:

S.B.’s initial problem is reminiscent of the reports of anthropologists that populations which are not accustomed to visual representations have difficulty recognizing objects—including themselves—in photographs (see Segall, Campbell, and Herskovits, 1966:32; Deregowski, 1973). Recent studies have shown that even when peoples not accustomed to visual representations are able to make out figures, the process may be laborious and tiresome. In a study of the Me’en, a remote Ethiopian tribe, Deregowski, Muldrow, and Muldrow (1973) note the following observation from a thirty-five-year-old man:

But, for S.B., experience did not bring the entire world of objects into focus. Particularly telling were S.B.’s attempts to draw a bus, which are illustrated in figure 2.3:

Fig. 2.3 S.B.’s Drawings of a Bus after Regaining Sight Age 52 (from Gregory, 1966:198-199)

Similarly, although S.B. was able to recognize most objects at sight, “On one real object he did show real surprise, and this was an object he could not have known by touch—the moon” (ibid., p. 195).

A third example suggests that experience is a necessary ingredient not only for perceiving specific items of our world, but also, more generally, for aiding us in discerning what kind of sensory input is significant for our purposes. William James’s description of the infant’s world as “a blooming, buzzing confusion” seems to cast its shadow in a world initially unordered by sight. Thus,

It would seem that S.B. had learned to determine which tactile (or auditory) sensations were significant and which could be ignored, but not so for visual stimuli; hence, the visual stimuli were overwhelming.

Biological Filters

By “biological filters” I mean genetic predispositions to perceive one’s environment in preestablished ways. Commonly known as innatism, this position maintains that our predispositions to certain sorts of perception are part of what it means for us to be human. The linguistic concomitant to perceptual innatism is either that certain “ideas” are innate and that language represents these ideas directly (like the word two representing the number 2), or that the linguistic categories themselves are innate.

The Cartesian theory of innate ideas presupposes that all humans come to know the world through the same blueprint. Unlike the “no filter” approach, innatists do admit that our perceptions are “filtered” before being understood (or linguistically encoded), but they suggest that the blueprint is the same for all people.

The debate over innatism, from Plato’s Meno, through Descartes and Locke, up to Chomsky’s neo-Cartesianism, has a long and confused history.⁹ It is not clear, for example, how much of human knowledge Plato believed to be innate, whether an ability that only manifests itself in the course of human development (rather than at birth) can be said to be innate, or precisely which aspects of language might be called innate. Further, there has been much discussion of the difference between “universals” and “innatism,” a distinction that Hilary Putnam (1967) has argued for but that Chomsky has tended to minimize. In the domain of language proper, we do not really know what data for or against the linguistic hypothesis would look like. As long as we admit a distinction between surface and underlying levels of language analysis without placing any bounds on how abstract our underlying analysis may be, then it is possible to recode any apparent violations of language universals (read “innate ideas”) into “underlying” commonalities with surface differences.

Perception is one domain in which it is possible to gather objective data on the issue of innate “filters.” If humans are indeed “predisposed” to perceive their environments in the same way, then perceptual tests should yield similar results across culturally distinct populations. If we do not find such commonalities we have not disproved the notion of innatism in perception or language (properties other than the ones we have examined might still be innate), but we have at least cast doubt on the innatist enterprise.

One area for most clearly testing the innatism of perception is that of optical illusions. If perceptual abilities are innate, then we would expect all people (and all peoples) to make similar perceptual judgments on similar phenomena. Homogeneous responses in perception experiments would also support the “no filters” hypothesis—i.e., that the world is perceived (and labeled) directly as it really is.

In using optical illusions to test for homogeneity of perception, we make use of the fact that, by definition, more than one interpretation of the visual array is possible. In figure 2.4, we may “see” the single drawing as both an Eskimo and an Indian. In figure 2.5, our visual cues give us a different interpretation of the relative lengths of the two horizontal lines than would measurement. In the real world, the same object cannot be both an Eskimo and an Indian, and two lines cannot be both equal and unequal. We must therefore posit some mental shaping (or filtering) of the data to account for these illusions. If we assume the ability to perceive optical illusions to be consistent among individuals (and across peoples), then we must also assume some common mental faculty which accounts for this ability.

Fig. 2.4 Eskimo or Indian?—The Winson Figure (from Gombrich, 1973:239)

Within the past two decades, an increasing amount of data has been gathered which suggests that the ability to perceive optical illusions is a product of sociocultural conditioning rather than biological programming. Rivers (1901, 1905) had observed in his early tests with optical illusions that “the non-Western peoples seem to be less subject to one illusion while more subject to another (cited in Segall et al. 1966:64). Gregory and Wallace, in their study of recovery from blindness, observed that “during the second postoperative month, the patient displayed either no illusion susceptibility at all or a degree of susceptibility considerably less marked than that typical of normal observers” (ibid., p.81). From this, the authors infer “that the patient may have lacked the opportunity to learn the perceptual habits that underlie the illusions” (ibid.).

Several contemporary studies have examined whether the ability to perceive optical illusions is a product of biological predisposition or a case of cultural conditioning (see Segall et al. 1966; Deregowski, 1973). As one of their tests, Segall and his colleagues showed the Müller-Lyer illusion (figure 2.6) to large numbers of European and non-European populations. Only the Europeans were highly susceptible to this illusion—they saw the vertical line to the left as being longer than the one to the right. One explanation for the Europeans’ interpretation is “the carpentered world hypothesis”:

Fig. 2.5 Optical Illusion Resulting from Perspectival Drawing

Fig. 2.6 The Müller-Lyer Illusion

Idiosyncratic Filters

A third possible relationship between experience and our perception (and linguistic encoding) of it involves idiosyncratic filters that result from the unique histories of individuals. In some instances, our experiences become increasingly common (for example, as we become older, we read more of the same books and visit more of the same places). At other times, however, we persist in using similar language to encode very different sets of experience.

Let us consider metaphor and ambiguity as two prototypical examples of how our individual experiences—including both our interactions with our physical environment and our linguistic interactions with one another—can affect the links we make between language and experience. Take first the case of metaphor. If I tell you that my understanding of mathematics grows with time, neither of us conceives of a biological organism which increases in mass the way that a “growing” chicken or a “growing” petunia might. We understand the term grow to be used metaphorically, that is, sharing some but not all of the more literal (or more usual) senses of the verb grow. But metaphoric language must be learned. Werner Leopold illustrates the point in a study of how his daughter Hildegard learned to talk. When Hildegard was not quite two years old, Leopold was working on a picture frame and complained about the weather with the words Es ist heiss (“It is hot”). Leopold reports that Hildegard, who was raised to be bilingual in German and English, responded by feeling “different parts of the picture before me to see if they were hot. The range of the adjective was restricted to tangible objects” (1949:108-109).

As with metaphor, our experience helps determine how we interpret an ambiguous sentence. The two cases are not wholly analogous, though. While metaphor involves extending the meaning of words to new semantic domains—where the pairing needs to be learned (or created) by the language user—ambiguity is more akin to optical illusions in which we possess all the cultural information necessary to “see” both interpretations. At issue in the case of both ambiguity and optical illusions is whether there is one interpretation we are more likely to “see” first.

Transformational grammar has emphasized the fact that native speakers of a language have, as part of their linguistic skills, the ability to disambiguate sentences whose surface structures are ambiguous. Hence, presented with the sentence “He decided on the boat,” speakers of English are able to derive two interpretations: “He chose the boat,” or “He made a decision while physically located on the boat.” And yet the likelihood of one interpretation coming to mind first is likely to be influenced by the context in which the sentence is uttered. Thus, in the sentence we have been using, the first interpretation (“He chose the boat”) is more likely at a boat showroom, while the second is more likely when talking about a person who has weighed anchor.

Another important issue in linguistic ambiguity is whether speaker and hearer share the same experiential history or socio-cultural presuppositions. Sometimes people speak at cross-purposes—having different referents for the same linguistic symbols—without ever realizing their communicative failures. On one occasion I assumed I was commiserating with a colleague over a particularly boring business luncheon meeting we had recently attended. I commented, “Wasn’t that business luncheon ghastly?” He concurred, and I was about to begin citing some of the prime examples of academic rhetoric to which we had both been subject for two grueling hours. Before I could speak, though, he continued, “Considering how much money we paid, you would have thought the hotel could have provided better food.” Had the conversation ended with my initial comment and my colleague’s assent, neither of us would have known that our individual tastes (I thought the food was tolerable and he found the meeting interesting) were leading us to provide radically different interpretations for the same linguistic expression.

Sometimes necessity makes us recognize how ambiguity arises from different presuppositions. A student of mine was studying language in blind children, comparing it with normal language development in their sighted counterparts. In her search of the literature, she encountered an article titled “Oral Problems of Blind Children.” Only when a copy of the article arrived through the interlibrary loan office did she recognize that the subject of the article was not human language, but dentistry.

Sociocultural Filters

The discussions of the “no filters” and the “biological filters” positions have already illustrated the effect of social or cultural experience on what we perceive or what we encode in our language. We have seen that the ability to perceive a visual representation (a picture of a face or of a water buck) may be culturally conditioned, as may the ability to perceive certain optical illusions. An even more dramatic case of the influence of culture on perception and representation is that of traditional notions of navigation in the South Seas.

When we look at a map of Polynesia, Micronesia, and Melanesia (see figure 2.7), we are struck by the vast amount of water which separates comparatively minute bits of land. In Polynesia and Micronesia, for instance, there are roughly “two units of land for every thousand of water” (Lewis, 1972:15). Europeans were, for quite some time, puzzled as to how the population of these isolated islands could share such a high degree of uniformity in language and customs. Strong similarities in linguistic and cultural patterns presuppose regular contact between peoples. What has been unclear to western observers is how regular contact was possible in the South Seas without reasonably sophisticated navigational equipment or, at the very least, the aid of maps—and the indigenous populations had neither. Only careful study, typified by the work of Thomas Gladwin (1970) and David Lewis (1972), revealed a highly sophisticated system of indigenous navigational skills, including means of representing the relationship between sea, land, and sky which are unknown in the west.

The most vivid of these cases is probably the use of stick charting in the Marshall Islands,¹⁰ or stone representations of waves reported by Gilbert Island navigators (such as figure 2.8).¹¹ These are not navigational charts in the western sense, but rather instructional and mnemonic devices concerned with swell patterns (Winkler, 1901:490).

Fig. 2.7 Map of Polynesia, Micronesia, and Melanesia (from Lewis, 1972:14, reprinted by permission of Dr. David Lewis, c/o Curtis Brown (Aust.) Pty. Ltd., Sydney.

Fig. 2.8 Schematic Representation of Gilbert Islands Stone Chart (from Lewis, 1972:186, reprinted by permission of Dr. David Lewis, c/o Curtis Brown (Aust.) Pty. Ltd., Sydney.

Figure 2.8 is a schematic adaptation of a stone “chart” made by Temi Rewi, a Gilbert Islands navigator. The center stone can stand for either an island or a canoe, depending upon the pedagogical role of the “chart.” Thus, for instance,

Novices being trained in the art of navigation learn to recognize islands not by their proximity to other islands (as they would in western navigation), but by the wave patterns characteristic of that island.

Somewhat harder to comprehend—but directly relevant to the issue of how cultural practices can mediate between experience and linguistic representation—is the concept of etak or “reference island” used in the Caroline Islands in Micronesia. Imagine yourself sitting in a moving train. You “know” as you “see the countryside flash by” that it is really you, not the countryside, that is moving. However, with some effort, you can make a gestalt shift and imagine the reverse to be true.¹²

The notion of etak not only makes this gestalt shift, but also creates a vocabulary which speaks of external reference points as actually moving. Instead of sitting in a train, one sits in a canoe, and rather than the scenery “moving,” what “moves” is the islands one passes. Like distant mountains seen on a train ride, the stars one sees in voyaging appear to remain fixed.

Figure 2.9 illustrates how the concept of etak works for the sailor

Fig. 2.9 The Concept of Etak or “Reference Island” in the Caroline Islands, Micronesia (from Lewis, 1972:133, reprinted by permission of Dr. David Lewis, c/o Curtis Brown (Aust.) Pty. Ltd., Sydney.

who is navigating between islands A and B. Points X, Y, and Z are stars which the navigator takes as the third point in a straight line of which his canoe is the first, and the etak point, the middle:

The Carolinian concept of etak strongly resembles the contemporary western understanding of the relationship between the earth and the sun. We “know” that it is the earth, not the sun, which moves, although in ordinary language we continue to talk of the sun rising and setting. Similarly, “the Carolinians are perfectly well aware that the islands do not literally move,” while at the same time, native navigators were “at very great pains” to make sure that Lewis “realized that all the islands concerned with a voyage ‘moved’ under the stars” (1972:134). In the case of the earth and the sun, our ordinary language has not kept pace with our theories, while among the Carolinians, the language of navigation is recognized as being at odds with empirical “knowledge.” Yet in both instances, we cannot help wondering how much influence our linguistic patterns have upon our beliefs about what the physical world is actually like.

Modality Filters

So far, I have been talking about filters between experience and its representation which derive from individual or social factors. We can also consider whether the way in which the linguistic representation itself is formulated has any influence upon what sorts of linguistic representation are possible.

Until we have considered spoken, written, and signed languages with a comparative eye, we cannot assess the strengths or weaknesses each of the three modalities might have as a means of representing experience. The following are some of the beliefs that commonly circulate about the relative degree of representation possible in these three modalities and their implications for a study of linguistic representation:

Sign language is more “universal” than spoken or written language.

Implication: sign language is more “representative” physically of experience than is speech or writing.

Visual languages do not have the same expressive range as spoken languages—for example, visual languages cannot express abstractions, nor can they consistently distinguish between unique individuals; and class terms.

Implication: visual linguistic schemes are either inferior to verbal schémas in an absolute sense or in ease of representation.

I shall consider hypotheses of this sort in chapters 5, 6, and 7.

UNIVERSAL LANGUAGES: UNWITTING LINGUISTIC THEORIES OF REPRESENTATION

During its modern history, linguistics has, as a discipline, paid reasonably little attention to the issue of representation. Instead, it has focused its attention on constructing good second order representations (linguistic theories) of languages (first order representations). The issues of what linguistic signs refer to (that is, the study of semantics) or of how one determines what those referents are (the problems of representation we have just been talking about) have been far less important than the enterprise of puzzling out how words and sentences are constructed out of smaller units.

There has been one important exception to this trend. It has counted relatively few professional linguists among its ranks (Otto Jespersen and Edward Sapir being major exceptions). Nonetheless, it has, perhaps unwittingly, laid the groundwork for the theory of semantics which was to become the theoretical companion of Chomsky’s transformational model in syntax, namely the Katz-Fodor model of semantics (e.g., Katz and Fodor, 1963; Katz, 1966).

The exception to the prevailing linguistic tradition that has ignored problems of representation is the universal language movements. Because these movements have been largely ignored in the linguistic literature, we need to get some feeling for the contexts in which they arose if we are to understand why they have, so far, not succeeded. Some of these reasons are directly tied to their presuppositions about representation, while others are social or political. By highlighting the problems of representation in conscious universal language schemes, we can better understand the similarities between these problems and the ones confronted by contemporary semanticists who share some of their unacknowledged predecessors’ presuppositions.

Interest in universal languages began to develop in western Europe in the early seventeenth century, becoming an issue of intense debate throughout the seventeenth and eighteenth centuries. The topic became less prominent during much of the nineteenth century, but reemerged in a somewhat different guise during the last decades of the 1800s and into the present century. We must investigate why the construction of a universal language was felt to be important, what kinds of languages have been proposed, and what success or failure they have met with—and why.

What is a universal language? Most simply, a universal language is a linguistic system that can potentially be used by speakers of all the world’s languages. We need not restrict our definition to a particular mode of linguistic expression—spoken, written, or gestural. In reality, most proposed languages have entailed both a spoken and a written form, although in a few cases universal writing systems (without spoken counterparts) have been suggested, as have sign languages (for which durable visual representation is awkward).

Furthermore, we need not imply that a universal language would be the sole or even the primary language of speech communities throughout the world. In fact, in most nineteenth- and twentieth-century schemes, the originators explicitly speak of international auxiliary languages (rather than universal languages) to indicate their intention not to replace the native languages of various countries but rather to have speakers learn an additional language to be used only when needed to communicate with linguistic outsiders.

In a letter to Mersenne in 1629, René Descartes distinguished between two types of universal languages: a priori languages (later referred to as “philosophical languages”) and a posteriori languages. A priori languages, which were particularly popular in the seventeenth and eighteenth centuries, were composed of “arbitrarily selected letters, syllables or words indicating an idea, or a group of ideas in accordance with a determined classification and based in no way upon natural languages” (Beaufront, 1919:vii). For the most part, the creation of these languages was motivated by the desire for a clarity of thinking not possible with existing natural languages. The most famous of these schemes, which we shall discuss shortly, is Bishop Wilkins’s “Real Character.” A posteriori languages, on the other hand, have been most prominent within the last hundred years. These languages are “based on roots already existing in the natural languages with a grammar reduced to its most simple expression and comprising neither irregularity nor exception” (ibid., p. viii). A posteriori schemes have, probably without exception, been devised primarily to foster international communication. In practice, though, they have often contained, as an underlying current, the goal of philosophical languages: the creation of a language more logical than natural languages. Esperanto is the best known of the a posteriori languages, although dozens of such languages have been proposed over the last century.

MOTIVATIONS BEHIND THE CREATION OF UNIVERSAL LANGUAGES

Universal language schemes have been motivated by two different classes of factors. The first, concerned with clarifying human thinking, directly confronts the question of how language should represent experience. A second group of forces, dealing with increased cross-national communication, is more social and economic in character. In most instances, several individual factors have simultaneously made the development of such languages desirable.

Universal Language as a Boon to Clear Thinking

The largest single impetus for developing a universal language, both in the seventeenth and again in the twentieth century, was the development of empirical science. Seventeenth-century England experienced a revolution in its philosophy and methodology of science. Francis Bacon, declaring all knowledge to be his province, devised an elaborate inductive procedure by which to know the world (his Novum Organum was intended to replace Aristotle’s Organon). According to Bacon, physicial nature is relatively easily comprehended, given the proper methodology. In The Great Instauration, he reveals his plans for a new empirical science, and many of his contemporaries and heirs shared his optimism. Typical of this enthusiasm was Thomas Sprat’s declaration in his History of the Royal Society that “we may ghess that the absolute perfection of the True Philosophy is not farr off” (1667:29).

Richard Jones (esp. 1932, 1961) argues that this new scientific attitude towards man’s relationship with nature was the primary motivation for seventeenth-century linguistic reform. Baconians became suspicious of ordinary language not only because of its association with the old science, which seemed to depend upon words more than upon nature, but also from a belief that current languages served to obscure rather than reflect reality:

In the Novum Organum, Bacon assails the Idols of the Marketplacevulgar communication—as the most troublesome of all false notions affecting human understanding. These idols

In Of the Advancement of Learning, Bacon suggests that the first distemper of learning occurs “when men study words and not matter.” He asks,

Matter, not form, must be the object of our attention. Otherwise, our capacity for expressing knowledge will be severely limited. For

Underlying Baconian science is the assumption that nature is uniquely describable. While aware of observational contradictions, the Baconian attributed such inconsistencies to insufficient observation or a faulty linguistic system which obscured the true nature of the event. Adherence to the uniqueness principle is a prerequisite for writing a universal philosophical language. Wilkins clearly believed that a unique analysis of nature was possible. In his Essay he writes, “As men do generally agree in the same Principle of Reason, so do they likewise agree in the same Internal Notion or Apprehension of things” (1668: 20).

Seventeenth-century arguments by Bacon and members of the Royal Society that a new language was needed to clarify the findings of scientific explorations were echoed more than two centuries later in the work of early twentieth-century language reformers. In the former case, the solutions proposed were largely a priori languages (i.e., wholly artificial) and in the latter, a posteriori (i.e., based on natural languages), but in both instances the goal was strikingly similar.

The modern complaint has been that existing natural languages are too imprecise and ambiguous to formulate the exact statements we wish our scientific theories to make:

Similarly, Ostwald has argued:

Sapir, one of the few linguists (along with Jespersen) to lend strong support to the international language movement, advocates more generally a logically structured international language:

Twentieth-century proponents of a universal language have, in fact, looked upon science as a prototype of modern thinking and as a model for a widespread move toward internationalism:

Thus we might recognize the special plight of professional scientists, for whom the learning of a single universal language is advocated as an alternative to having to learn a number of already existing languages because

Universal Language as a Means of Cross-National Understanding

Economics also contributed to universal language movements, the focus here being on international understanding more than on improving the “match” between language and experience. In the seventeenth century new trade routes were being charted and speakers of western European languages were unable to make themselves understood in the ports and crossroads where they wished to do business. A fairly well established trading language known as Lingua Franca had developed around the Mediterranean during the Crusades, which allowed for some shared communication between merchants. But as the trade routes extended to Africa and the Far East, where Lingua Franca was not known, and the number of merchants increased, a common language was needed once more.

Because the expansion of trade in early modern Europe coincided with interest in philosophical languages which aided clear thinking and expression, it was not surprising that these same universal language schemes should have been thought useful in facilitating international trade. Historically, however, the linguistic solution which trading groups adopted was not the imposition of a new language on both parties, but rather the gradual development of trade jargons.

Calls for an international language for trading purposes subsided during much of the next two centuries. As trading relationships gave way to empires, and individual Western powers began staking their claims to different sectors of the non-European world, individual solutions to the linguistic problem were devised. In the case of a highly civilized (though non-Western) country such as India, British officials were required to learn Sanskrit before taking up residence in India. In most instances, however, a pidgin language evolved which sufficed for the amount of communication which was necessary between colonizer and native (see chapter 5).

An impetus for universal language development was also provided by religion, especially in the early modern era. To the extent that peoples of differing religious ideologies are also speakers of different languages (which is not always the case—as in the English Civil War), a universal language is potentially useful in resolving ideological differences. Seventeenth-century language planners spoke of the religious value of a universal language on two planes. The first bordered on the mystical, the second was wholly pragmatic.

Comenius, a Czech educational reformer who derived much of his ideology from the mystics Johann Valentin Andreae and Jakob Böhme, believed in a mystical harmony of nations which needed to be restored to its original balance. Andreae had hoped to organize a universal Christian intelligentsia which would pursue the knowledge needed to unite mankind. This wish is manifested in Comenius’s envisioned philosophical college of Pansophia, which he charged with the task of fitting exact expressions in language to exact concepts.

A more mundane concern for religion was the conversion of the heathen. Concomitant with the rise of exploration and foreign trade in early modern Europe was the proliferation of Christian missionaries. Their problem was the same as that of merchants and later colonizers: how to communicate with those who do not speak the same language. The solutions, as it turned out, were also largely the same. Rather than appealing to theoretically useful but practically implausible language schemes, missionaries often forged pidgin languages out of their own languages and those of the indigenous populations. In many cases, missionaries learned the native languages of the people among whom they were proselytizing. In fact, missionaries have produced much of the grammatical information linguists now have on non-Indo-European languages.

BISHOP WILKINS’S ESSAY TOWARDS A REAL CHARACTER, AND A PHILOSOPHICAL LANGUAGE

What did these universal language schemes look like? Since there have been dozens—perhaps hundreds—of schemes proposed, it is not possible to characterize them all. Instead, let us consider the scheme which is perhaps best known, that of Bishop John Wilkins. Wilkins’s scheme is instructive not only because it is more complete than any other but also because it provides striking parallels with linguistic proposals that emerged almost precisely three hundred years later.

Wilkins’s Essay (1668) was written in response to a request by members of the Royal Society of London to create a language which clearly expressed the composition of the natural world and our place in it. The language was a priori in design, not relying upon any existing language for its model. It does seem likely, though, that the idea of creating a character that mirrored reality was influenced by growing knowledge of Chinese characters, which were seen as directly reflecting the world of experience.

In the Essay, Wilkins undertakes to divide all of nature into a series of genera (he calls them genuses and so shall I), predicaments, and species. Having thus divided nature, he then assigns both a phonetic representation and a written representation to each subdivision, so that it can be referred to in spoken and written language. He lists a total of forty genuses, grouped into predicaments (e.g., substance, quantity, quality). Each genus is then subdivided into differences, and each difference is divided into as many as nine species. Figure 2.10 shows Wilkins’s forty genuses (grouped into predicaments), along with the phonological and written character used to express each genus.¹³

The next step is to break down each genus into between six and nine differences. Figure 2.11 shows the spoken and written indicators of each of these differences. The “sound” is placed on the right side of the genus symbol, while the written indicator for the difference is placed on the left. The same symbols are used in marking the subdivisions of each genus, although the referents being marked within the genus are obviously different. The final stage is to divide the differences into species, again using a standard set of sounds and characters to indicate all species differentiations (see figure 2.12). Both the phonological and written indicators of species are placed on the right hand of the difference marker.

Figure 2.13 presents a typical example of difference and species subdivisions of a genus. All items of experience are assumed to be identifiable by genus, difference, and species. To indicate any item linguistically, one strings together the sounds or written symbols for the appropriate genus, difference, and species. Thus if one wanted to refer to “sand,” one would proceed as follows:

Note that the availability of forty genuses, nine differences, and nine species only yields a potential 3,240 unique identifications. We know that there are far more than 3,240 items of experience to identify, and far more than that number of words in the English language. Wilkins obviously recognized the problem; he allowed for additional differences when necessary, and did not always distinguish between some related objects and words. In our example, both “sand” and “gravel” are given the same ontological and linguistic identification. Moreover, Wilkins created grammatical means for deriving adjectives and adverbs, as well as separate symbols for the copula, pronouns, interjections, prepositions, and conjunctions.¹⁴

Fig. 2.10 Bishop Wilkins's Forty Genuses, Grouped into Predicaments

Fig. 2.11 Bishop Wilkins's Nine Indicators of Difference

Fig. 2.12 Bishop Wilkins’s Nine Indicators of Species

PROBLEMS WITH UNIVERSAL LANGUAGE SCHEMES

In the 1920s, Sylvia Pankhurst envisioned a golden future for the idea of a universal language, a prototype of which she calls Interlanguage:

Fig. 2.13 Example of Difference and Species Subdivisions of a Genus

Three hundred years earlier, Robert Boyle had expressed a similar sentiment:

Today we are no closer to using an international language than we were in Boyle’s time. Fifty years have passed since Pankhurst’s prophecy, and we still lack a common language. What has gone wrong? We can identify two classes of problems: technical and theoretical. In many cases, the technical problems could be overcome, while the chances of solving the theoretical problems are far less certain.

Technical Problems

The first two technical problems apply largely (although not exclusively) to a priori languages. In creating a new language, one of the first problems is getting people to learn it. The more difficult the language, the less one’s chance of success. Aware of this problem, language planners from the seventeenth century to the present have sometimes stressed the ease with which their schemes can be learned. Dalgarno, inventor of one early philosophical language,

And Wilkins claimed at the end of his Essay (p. 453) that his new language was forty times easier to learn than Latin. In practice, however, the a priori schemes have proven difficult to learn and use, although one might argue that they have been insufficiently tested.

A second technical problem with a priori languages in particular is their zero level of redundancy; every sound (or every mark) has a distinct meaning which is represented nowhere else in the linguistic message. As Pankhurst observed,

In Wilkins’s scheme, for example, Z a denotes the class of fish, Ze, the class of birds, and Zi the class of beasts; a single mispronunciation might yield a totally unintended genus or species. We would expect a language with no redundancy to be very difficult to learn or to use.

There have been other problems as well. Most a posteriori languages that have actually been worked out have been demonstrably drawn from one of the European languages. Speakers of different European languages (or non-European tongues) have been loath to learn as an “international” language a scheme which is biased in this way. Moreover, no universal language system has received sufficient popular or governmental support to make the language’s use widespread. The International Esperanto Society continues to meet, but neither boards of education nor university departments of modern languages actively support the teaching of Esperanto—or any of the other auxiliary languages. We cannot fairly judge the potential for success or failure of languages which have not been adequately tested.

Theoretical Problems

Even if the technical problems discussed above can be solved, a number of deeper problems concerning the whole conception of a universal language would seem to cast doubt upon the eventual success of the enterprise. All of these stem ultimately from the underlying assumption that a universal language—either a priori or a posteriori—is unchanging and eternally appropriate for all language users:

This belief in the eternal nature of a universal language has been manifested in many ways—and is susceptible to a large number of counterarguments. The most blatant of these assumptions about linguistic stasis was the belief that

Whatever one’s views on the Sapir-Whorf hypothesis (which suggests that languages influence the way in which speakers think about and perceive the world), it is generally recognized that different historical and material conditions may make the linguistic expression of certain topics more critical to some peoples than others. Trobriand Islanders may not need a language which distinguishes types of stone used as permanent building materials (stones, Wilkins’s eighth genus, are finely divided), while inhabitants of New York City may have no use for detailed distinctions of flora and fauna.

The failure of universal language schemes to recognize the lack of uniformity of linguistic needs is complemented by the presupposition made by language architects that their categories for describing “things themselves” are necessarily—and eternally—correct. The problem with this assumption is particularly clear in the case of a priori schemes which explicitly attempt to categorize all knowledge, although the problem is also seen with a posteriori languages to the extent that they claim to mirror reality directly. Pankhurst observes that a priori languages are necessarily doomed to failure as eternal systems since “ideas, and views as to their classification, are constantly changing” (1927:22). As an example, she points out that the Aristotelian classification of elements into earth, air, fire, and water, is no longer accepted. A near contemporary of Wilkins, the mathematician Thomas Baker, made a similar point:

The extent to which an a priori language could handle such changes in scientific theory and empirical knowledge would depend on the rigidity of the formulation of the scheme and on the extent of change in categorization implied by the new knowledge or theory. If, for example, one discovers (as zoologists did) that the whale is not a fish but a mammal, it would be rather simple to change the sound or symbol used to represent whales in the language. However, if the whole system of classification is challenged (as Baker suggested might be necessary in the case of Wilkins’s scheme), then it is not clear what would remain of the original a priori language after the required changes. Moreover, by definition, it was assumed that the classifications made were correct, now and forever, since they were intended to reflect the real structure of things as directly perceived by all peoples.¹⁵

FROM ARISTOTLE TO (ANIMATE)

Since its inception, transformational grammar was intended as a theory of grammatical collocation, not as a theory of meaning. Chomsky explains in Syntactic Structures (1957: chap. 9) that he sees no obvious way in which semantic information is necessary or even useful for grammatical analysis. Largely as a result of this position, the study of semantics was ignored by most transformationalists into the 1960s.

One exception was the work of Jerrold Katz and Jerry Fodor. Their paper “The Structure of a Semantic Theory” (1963) laid the groundwork for the semantic component of grammar that would interpret the syntactic constructs generated by the syntactic component. The semantic theory was based on a set of semantic features which, in combination, were seen as specifying the meanings of all words and sentences in the language.

In the earlier version of the semantic theory, semantic features came in two varieties:semantic markers and distinguishes. Semantic markers (enclosed by parentheses in writing) were “intended to reflect whatever systematic semantic relations hold between that item and the rest of the vocabulary of the language” (Katz and Fodor, 1963:187). Distinguishers (enclosed by brackets) reflected “what is idiosyncratic” about the meaning of a word. Thus, the semantic reading of the word colorful would include the semantic marker (Color) and the distinguisher [Abounding in contrast of variety of bright colors] (ibid., p. 202). Later versions of the theory (Katz, 1966) eliminated the category of distinguisher, leaving only semantic markers.

Where do these features come from? How did Katz and Fodor select them? Are there any criteria for determining the right set? Before exploring these questions, it is only fair to point out that subsequent semantic theories (e.g., Chafe, 1970; Lakoff, 1970b; McCawley, 1968; Schank, 1972) have also proposed the equivalent of semantic features, though sometimes by other names. The question of how these subsequent theorists have selected their semantic primes is just as important —and its answer just as mysterious—as in the case of Katz and Fodor.

The Problem of Categorization

In his attempts to order or analyze his environment, man almost invariably begins to categorize. Our recorded heritage abounds with treatises on types of heavenly bodies, types of angels, type of angles, types of fish. Another favorite typology is that of language. How many parts of speech are there? How many declensions? How many types of sentences?

The precise relationship between ontological and linguistic categories has puzzled both philosophers and linguists. At one extreme we find linguistic relativists, according to whom a unique analysis of nature is impossible or else exceedingly difficult because of diverse human linguistic-perceptual patterns. At the opposite pole are univer-salists who, while recognizing a dependence between language and perception, imply that unique analyses in science are possible because of the uniformity of human thought and language. Between these extremes are linguists with no particular interest in nature—such as Katz and Fodor, and scientists with no particular interest in language—whom we may disregard here.

The best known of all categorizers is Aristotle; his biological treatises (e.g., Historici Animalium, De Partibus Animalium, De Generatione Anirrialium) directly confront the problem of classifying nature. One product of this endeavor is a “scale of nature” in which animals are grouped according to their mode of biological generation. Aristotle’s Categories is, however, a troublesome book. To paraphrase Julius Moravcsik (1967), what is the list of ten categories a list of? Are the predicaments substance, quantity, quality, relation, place, date, position, state, action, and passivity meant to carve up language or nature? Aristotle’s commentators have not been unanimous in their answer. One commentator observes that although most of these categories are mentioned in nearly all of Aristotle’s works, “the doctrine is everywhere treated as something already established” (Ross, 1923:22). According to Antoine Arnauld, “The ten Aristotelian categories are the different classes into which Aristotle divided objects of thought (1662: 1964 edition, p.42). While the Categories is largely a linguistic treatise, it does not assume a one-to-one correspondence between the ten divisions and grammatical distinctions. Moravcsik resolves the dilemma through compromise:

Aristotle’s stature throughout the Middle Ages is exemplified by one of the statutes of Oxford University under which

The Baconian Renaissance, however, revolted against its scholastic heritage. Thomas Vaughan reflects the new empiricism when he urges Aristotelians not to let their

Wilkins himself explicitly rejected scholastic philosophy in his Discourse Concerning a New Planet (1640):

Nevertheless, Wilkins distinguished between Aristotle the philosopher and scientist, and Aristotle as interpreted by the Schoolmen. For the former, Wilkins had much respect. The Bishop’s debt to Aristotle the classifier of nature is obvious. In his Essay, Wilkins defines forty genuses which are subclassified into differences, the difference-genus distinction stemming from Aristotle; and all five of the predicaments under which Wilkins distributes most of his genuses are Aristotelian in origin:substance, quantity, quality, action, and relation.

Katz and Categorization

Katz¹⁶ approaches the language-nature duality in reverse. For Katz, the purpose of semantics is to

This explanation entails defining and distinguishing anomalous, unambiguous, ambiguous, and synonymous sentences. Nowhere does Katz explicitly relate his lexicon to the natural environment. A universalist in spirit, he implies adherence to some version of a Wilkinsian unique description of nature. Without claiming that all languages have identical subclassification systems, he suggests that at least the basic categories of every language are identical. It takes but one more step to identify this common core with a unique perception of nature. But, unlike Wilkins, Katz never commits himself. In fact, references to scientific classification are entirely absent from his semantics.

Despite his indifference to natural description, Katz, like Wilkins, returns to Aristotle’s Categories. He is intensely aware of the severe criticism the Categories has received in modern philosophy, summarizing the arguments as follows:

Three of these points assume that Aristotle’s theory of categories is strictly a theory of linguistic categories, an assumption Moravcsik would deny.

Katz proceeds to offer linguistic cures for philosophical ills by embedding Aristotelian categories within the now familiar framework of Katzian semantics. Semantic categories are defined as “a subset of those semantic markers which semantic theory specifies as substantive universals” (ibid, p.228). Semantic markers, in turn, are

But where does Katz derive his suggestions for labels? He implies that semantic markers are basically formalizations of conventional dictionary entries:

Katz captures this empirical fact by including the semantic marker (Male) in the lexical readings for all words in the first group but not for those in the second. Our problem, however, still remains: How did he know to choose the marker (Male)?

While Katz does not allude to a source, there would seem to be several candidates: first, empirical knowledge of his natural environment and of the English language. As Narveson (1968) implies, (Male) is an “empirical generalization” only in that it suggests Katz’s recognition that many words in the English language connote maleness. A second possible source is componential analysis. Although he never refers to componential analysis, it may be instructive to remember that the 1950s witnessed the growth of a theory of meaning which analyzed groups of words into shared parts. Most componential analyses have focused on kinship terms—where (Male) is a crucial category, though the technique is quite general (cf. Goodenough, 1968; J. Lyons, 1968). Note that some of Wilkins’s discussion neatly dovetails with twentieth-century componential analysis. In a passage arguing against the suitability of Latin as a philosophical language, he observes:

The third potential source is Aristotle himself. An analysis of commentaries on Aristotle reveals some striking precursors to the concept of semantic markers. Consider the introduction to Aristotle’s Categories by the third-century A.D. commentator, Porphyry:

Porphyry’s analysis persisted during the Middle Ages, as evidenced by the adaptation of a diagram found in Thomas Aquinas shown in figure 2.14.

Fig. 2.14 Tree of Porphyry (from Thomas Aquinas, Opusculum XLIV, Tractatus II, Caput IV)

Fig. 2.15 Adaptation of the Tree of Porphyry to Katz’s Semantic Features

The tree is largely translatable into Katzian semantic markers. Divisions in the center correspond to the semantic markers themselves, while the branches on either side represent positive or negative values (figure 2.15). Aristotle’s (Corpus) is interpreted by Katz as (Physical Object), and the next three categories—(Vivens), (Animal), (Homo) —have their direct counterparts in Katz’s model. The most general Aristotelian category, (Substantia), however, has no equivalent in the Katzian scheme. One might argue that the category of substance is redundantly presupposed by Katz. Such redundancy is apparently intended in the Categories:

Fig. 2.16 Katz’s Semantic Entry for Bachelor (from p. 155 of The Philosophy of Language by Jerrold J. Katz. Copyright © 1966 by Jerrold J. Katz. Reprinted by permission of Harper & Row, Publishers, Inc.)

That (Substantia) is, in fact, the most general category on the tree is made explicit by the term (Generalissimum) in the upper right hand corner of Aquinas’s diagram.

Having translated Porphyry’s Tree into Katzian markers, it remains for us to translate Katz into Aristotelian terminology. Katz’s citation for bachelor in his book The Philosophy of Language (1966) is shown in figure 2.16. Much of this lexical item is directly convertible into the Tree of Porphyry (figure 2.17).

Fig. 2.17 Modified Tree of Porphyry for the Entry Bachelor

What do we learn from demonstrating a high degree of isomorphism between Katz’s semantic categories on the one hand, and those of Aristotle (and, derivatively, Wilkins) on the other? If we believe that Aristotle’s classification of nature is correct, we should have no objection to Katz’s adoption of the scheme for contemporary linguistic analysis.

We do not know whether Katz intends to follow an Aristotelian scheme consistently, since he has not overtly identified his schematic model with Aristotle. Furthermore, there are many points on which contemporary science would disagree with Aristotle’s classification. Finally, we have yet to establish that divisions made in nature are necessarily appropriate as linguistic categories. We may “know” that whales are mammals, but this academic knowledge does not prevent 11s from talking about the “gigantic fish” we saw in the aquarium.