Constraint, cognition, and written numeration

Constraint, cognition, and written numeration Stephen Chrisomalis Wayne State University he world’s diverse written numeral systems are afected by human cognition; in turn, written numeral systems afect mathematical cognition in social environments. he present study investigates the constraints on graphic numerical notation, treating it neither as a byproduct of lexical numeration, nor a mere adjunct to writing, but as a speciic written modality with its own cognitive properties. Constraints do not refute the notion of ininite cultural variability; rather, they recognize the ininity of variability within deined limits, thus transcending the universalist/particularist dichotomy. In place of strictly innatist perspectives on mathematical cognition, a model is proposed that invokes domain-speciic and notationally-speciic constraints to explain patterns in numerical notations. he analysis of exceptions to cross-cultural generalizations makes the study of nearuniversals highly productive theoretically. he cross-cultural study of patterns in written numbers thus provides a rich complement to the cognitive analysis of writing systems. Keywords: cognitive anthropology, writing, constraints, numerals, number systems 1. Introduction Out of the darkness, Funes’ voice went on talking to me. He told me that in 1886 he had invented an original system of numbering and that in a very few days he had gone beyond the twenty-four-thousand mark. He had not written it down, since anything he thought of once would never be lost to him. His irst stimulus was, I think, his discomfort at the fact that the famous thirty-three gauchos of Uruguayan history should require two signs and two words, in place of a single word and a single sign. He then applied this absurd principle to the other numbers. In place of seven thousand thirteen, he would say (for example) Máximo Pérez; in place of seven thousand fourteen, he Railroad; other numbers were Luis Melián Lainur, Olimar, sulphur, the reins, the whale, the gas, the caldron, Napoleon, Agustín de Vedia. In place of ive hundred, he would say nine. Each word had a particular sign, a kind of mark; the last in the series were very complicated … Pragmatics & Cognition 21:3 (2013), 552–572. doi 10.1075/pc.21.3.08chr issn 0929–0907 / e-issn 1569–9943 © John Benjamins Publishing Company Constraint, cognition, and written numeration 553 I tried to explain to him that this rhapsody of incoherent terms was precisely the opposite of a system of numbers. I told him that saying 365 meant saying three hundreds, six tens, ive ones, an analysis which is not found in the ‘numbers’ he Negro Timoteo or meat blanket. Funes did not understand me or refused to understand me (Borges 1964: 64–5). In his short story, Funes, the Memorious, Jorge Luis Borges presents an account of a person blessed or cursed with an apparently limitless memory and who, among other things, constructs a bizarre numerical system. Funes’ memory allows him to ignore the constraints that apply to those of us whose memories are less prodigious than his own, thus producing a system of numeral words and number symbols1 lacking any structure, and in so doing loses the ability to communicate with others or even to perform arithmetic. Borges reminds us that the ability to forget is fundamental to humanity, and that the limitations of human cognitive capacities matter a great deal. Whether we say seven thousand thirteen or Máximo Pérez is not simply a stylistic choice. Number systems are subject to constraints that limit their range of variability. Below, I will outline an argument relating to the concept of constraint as a framework under which comparative questions in the study of written notations can be asked and answered. I refer to recent cross-cultural work on numerical notation systems (graphic, nonphonetic systems of number symbols) to address longstanding debates over the degree to which human behavioral plasticity is limitless. While there is ininite variability in human social life, this does not entail an extreme particularism, nor does it render futile the search for pattern and generalization across time and space. Constraint-based approaches better explain probabilistic generalizations about human behavior than do universalistic ones because they allow for exceptions, and indeed are best explained by examining those exceptions. In examining the interaction of written representations and cognition, there are two general sets of relevant constraints to be considered. First, we must ask how perceptual and cognitive constraints afect the structure of graphic notations (considering numerical notation here as a special case of written representations). In other words, how are the cognitive architecture of humans and the properties of the immediate environments with which they engage, relevant to the graphic systems they are likely to employ? his topic has been amply addressed in contemporary Western contexts by cognitive psychologists (Dehaene 2011) but does 1. I do not know what to make of the contradiction between ‘He had not written it down…’ and ‘Each word had a particular sign, a kind of mark…’ he resolution does not appear to be a translation issue (orig. ‘No lo había escrito…’ and ‘Cada palabra tenía un signo particular, una especie de marca; las últimas eran muy complicadas…’). 554 Stephen Chrisomalis not consider the wide range of numerical notations used worldwide over the past 5,000 years. Equally importantly, however, we must ask how graphic notations, once invented and used in speciic contexts, afect their users’ mathematical cognition. For instance, we might inquire, following Goody (1977: 12–13; 1986: 52–54), how diferent numerical systems (such as those of the LoDagaa of Ghana or predynastic Sumer) afect arithmetical practices. Establishing how these two processes intersect is critical for a full understanding of how written numeration relates to cognition — considering cognition both as cause and efect. Numerical notation has properties that are in some ways similar to those of lexical numeral systems (the number words in any particular language), whose structures are generally regarded by generative linguists as deriving from properties of universal grammar. Numerical notation, however, in contrast to language, is not cross-culturally universal and emerges only in speciic social contexts. Because it is a graphic representation of number that has no necessary or rigid connection to the number words of any particular language, its properties are diferent from those of lexical number. hat there are two distinct modalities for representing a single domain of activity (number), each with its own structural properties provides an opportunity to examine the speciic role of graphic representation in affecting the structure of cognitive frameworks such as number systems. 2. Constraints against universals and particulars A constraint is any non-deterministic factor, operating through one or more processes, which makes some outcome more or less likely to occur. A constraint theory helps generate statements about tendencies toward or against certain outcomes, as well as limitations on some sorts of outcomes, in some area of activity, without requiring or compelling a particular one. It helps us escape from phrases such as “lawlike generalizations”, which at best serve only as a euphemism for “laws” (Watson 1976), and at worst, conceal a willingness to ignore variation. Constraints are not merely tendencies; a constraint is not simply something that tends (not) to happen, but rather, is a factor that helps explain or model that tendency. his deinition also accounts for Franz Boas’ (1896) important recognition of equiinality, that the same efect may be produced by multiple diferent causes. One can identify likely constraints from outcomes, but they are not outcomes themselves. A variety of constraint models have been proposed in the social sciences over the past century, but perhaps the earliest does not use the term ‘constraint’ at all: Alexander Goldenweiser’s ‘principle of limited possibilities’ (Goldenweiser 1913). his is the very sensible suggestion that for most cultural phenomena, there are not a limitless number, nor only one, but a few distinct conigurations. Goldenweiser Constraint, cognition, and written numeration 555 noted (1942: 124–5), for instance, that the functional considerations of use limited the possibilities for oars — that despite variability in length, weight, cross-section, material, and so on, there are numerous similarities between all workable oars due to physiological, technical, and functional factors. Similarly, you can have a slotted (lat) screwdriver, or a Phillips (cross-head) screwdriver, and many others. But you cannot have a workable round-head screwdriver, even though it is easy to imagine one. Even though each context, individual, and history is unique, unrelated groups converge on similar solutions to problems. Written number systems, similarly, are a notational technology upon which a variety of constraints operate to limit crosscultural variability. Goldenweiser’s friend Edward Sapir used similar argumentation to explain why, although there are theoretically an ininite number of speech sounds, human languages use only a small fraction of these, because of the cognitive and perceptual limitations of the brain and the muscular limitations of the human mouth and vocal tract (Sapir 1921: 45–52). In contemporary linguistics, Optimality heory has explicitly used the concept of constraint since its inception (Prince and Smolensky 1993). However, such theories are usually subsumed today under universalizing models that identify absolute and knowable limits on the range of possible languages. Constraint, in this context, frequently means constraint down to a single solution based on a universal constraint set present in all languages. But constraints on graphic notations like writing systems and written numbers cannot be simply derived from universal grammar, because they are not direct, unmediated representations of language. Just as it matters what the shape of an oar is, it matters what the features of a sign are. he ecological psychologist J.J. Gibson (1979) developed the concept of afordances to describe potentials or possibilities for action that emerge from an object or environmental stimulus. A lat surface afords having something placed upon it, a round object afords being rolled, and ingers aford being counted (and counted upon). Afordance is, in some sense, the opposite of constraint, which has a limiting efect, but I see no reason to decouple them. Constraints and afordances oten co-occur — there is not much diference between a strong constraint against some behaviors and a strong afordance for one particular outcome. Gibson regarded affordances as emerging from the environment (including both exteroperception, the perception of the environment, and proprioception, the perception of the observer him/herself as part of that environment), but was suspicious of mental models that are not environmentally motivated (Gibson 1979: 141). his relected his anti-cognitivist perspective, and here we must part ways. An afordance-based model may explain why so many languages’ numeral words are decimal (because humans have ten ingers) but it does not explain why number words and graphic number signs tend to be ordered from the highest to the lowest powers of the base, for instance. 556 Stephen Chrisomalis In his distinguished lecture ‘Constraint and Freedom’, the archaeologist Bruce Trigger noted that most aspects of human behavior are less orderly than universalism would suggest but more orderly than particularism would suggest (Trigger 1991). Aiming explicitly for a middle ground, he envisioned behavior as the product of a set of overlapping constraints, deined as factors “that human beings must take into account to varying degrees when selecting an appropriate course of action” (1991: 555–6). Trigger distinguished external, universalizing constraints like ecology and technology from internal, particularist ones like symbolic practices and cultural traditions. Trigger’s account is useful, but is problematic in two ways. First, it presumes that we are consciously aware of the constraints under which we make decisions, which is not always the case. Just as individuals do not readily understand the cognitive constraints underlying their phonological processes, most people are unaware of the constraints on their numerical systems. Second, there is no reason to think that internal, mental constraints are less likely to be universal than external, environmental ones. Ecological and environmental variability is enormous, but there are considerable commonalities behind all human brains. he key will be to determine which constraints are relevant, and how. 3. Constraining ininity In an old magician’s trick, which I irst learned about from the popular mathematics of Martin Gardner, the magician asks an audience member to think of a number: “I want you to name a two-digit number between 1 and 50. Both digits must be odd, and they must not be alike. For instance, you cannot name 11” (Gardner 1956: 174). he magician then immediately asks, “Is your number 37?” very frequently to the astonishment of the subject. Partly this trick works because the magician creates rules that restrict the subject’s choices, but obscures the severity of the restriction through subtle language that makes it appear as if the subject has more choice. here are actually only eight answers that fulill these criteria (13, 15, 17, 19, 31, 35, 37, 39). But this cannot be the only explanation, because far more than one-eighth of respondents will respond 37 (otherwise, it would not be a very reliable trick). Of course, not everyone will choose 37. hat would suggest that there were no other options at all. Similarly, it is uninteresting if somebody chooses to answer 53 or 8 or some other number that violates the speciied constraints. he fact that few people choose 15, while lots of people choose 37, is best explained by the hypothesis that three and seven are more prototypically odd than one, ive, and nine — at least in Western societies where three and seven are also symbolically evocative. When some outcome is possible, but is attested signiicantly more rarely (or more oten) than expected, constraints are probably involved. Constraint, cognition, and written numeration 557 Now to revise the game somewhat, I might ask you to think of a number that is greater than one, but that is less than two (i.e. neither one nor two themselves). In virtually any sizeable group to whose members such a question would be meaningful, it is a very reasonable bet that some respondents will choose one and a half (1.5). Halves and quarters, and to a lesser degree tenths, are the numerals between the whole numbers that most people use most oten. But note that while I did impose a constraint of magnitude, you were free to choose any rational number, like 1.832, or an irrational number, like the square root of 2 (roughly 1.414). In fact, there are ininitely many numbers between 1 and 2, and you could have chosen any one of them — far greater than the eight choices you had in the irst game. So we have, on the one hand, a sort of constraint, leading most of you to choose familiar numbers within the prescribed limits, but on the other hand, ininite variety — plausibly, you could have chosen, as I did, 1.6402895036 — although you didn’t. However, there are also an ininite variety of numbers that you could not have chosen — for instance, 6 or π. Human cultural variability is analogous to the second game, for many domains of experience. It is oten treated as an anthropological axiom that culture is ininitely variable. I do not wish to disagree with this at all — in fact, it is selfevident that human cultural variability is not only ininite but increasing over time. Every time two strangers meet, every time someone is born or dies or does anything, something new happens. But “ininity” does not equal “anything can happen”. Cultural variability is ininite, but it is not limitlessly ininite. A theory of constraints regards it as useful to understand variation, but also asks what overarching factors or processes limit that variation. his is neither a universalist nor a particularist approach, but a comparative one. It does not assume in advance, but rather asks, to what degree and by what processes any aspect of human behavior is constrained. he importance of a theory of constraint is not its moderateness; it is not a Solomonic compromise. Just as you could not have answered 6 within the rules of my second game, there is an ininity of potential human behaviors that are not attested, despite the ininity of those that are. he role of constraint theory is to bridge the gap between the imaginable and the attested, the commonplace and the rare. Without denying ininite variability, it forces us to ask, given considerable cross-cultural variability, what is it that we do not see — and why. One useful example of this principle comes from Brown and Witkowski’s (1981) linguistic and anthropological examination of igurative expressions in 118 languages, which revealed that the igurative association between the pupil of the eye and a small person (as in English pupil ‘student’) is far more common than would be expected by chance — around one-third of the languages in their sample did so. he explanation can hardly be genetic, can hardly be random, and does 558 Stephen Chrisomalis not result chiely from language contact. Instead, there is a perceptual constraint at play: if one looks into the pupil of someone’s eye, one sees a small relection of oneself (Tagliavini 1949; Brown and Witkowski 1981: 601). While this frequent conceptual linkage invokes general principles that extend beyond any one cultural context, this inding is hardly ‘universalist’, as Brown and Witkowski describe it. A universal that applies to only one-third of the world’s languages is hardly a universal, and I do not ind their rephrasing of the matter as ‘universalist tendencies’ to be analytically helpful. While Brown (1976) shows that there are universals of body-part nomenclature, they are of the form ‘All languages have a word for eye’ rather than these less universal yet more interesting patterns. Everett’s (2005) controversial account of what he calls ‘cultural constraint’ in the language and behavior of the Pirahã of Brazil is the most startling example of a sort of a theory of constraints in the domain of number. Everett believes the Pirahã to be the only exception to a wide range of otherwise human universals (color terms, embedding), to have the simplest kinship terminology, phonological inventory, and pronominal systems of any language, and to lack number words completely (2005: 622). His explanation for this is decidedly particularistic: he Pirahã possess a unique yet unspoken ‘immediacy of experience’ cultural constraint that restricts their speech and behavior in numerous respects. Everett sees his work as refuting Chomsky’s universal grammar, yet it is not clear what exactly he means by ‘cultural constraint’ in this account. If he means that it is a cultural value that is acquired through socialization, then it is not clear how it is acquired, whether it is ever expressed explicitly, whether Pirahã children violate this principle, and what happens if they do. If he means that in acquiring the Pirahã language, individuals are precluded or constrained from certain forms of thought, then this is not a cultural constraint, but a linguistic constraint, and becomes circular (language constrains language). In fact, what exactly Everett means by ‘culture’ is altogether unclear; accounts of limited-numeral languages generally invoke a social environment in which counting is irrelevant to most decision-making, but Everett does not take this functionalist approach. Everett’s work has sparked a host of responses and counterarguments, and is still being debated on evidentiary grounds (Frank et al 2008; Nevins et al 2009; de Cruz et al 2010; Reboul 2012). However, the chief issue is that there is no explanatory mechanism proposed by which this constraint, and its consequent linguistic expressions, comes into existence and is reproduced and transmitted socially. Without it, Everett’s model does not actually detail a constraint, and he cannot establish to what degree other, similar societies possess similar features (or not), and for what reasons. Constraint, cognition, and written numeration 559 4. Constraining written numerals In 1978, Joseph Greenberg published his now justly renowned paper, ‘Generalizations about numeral systems’ (Greenberg 1978). It is a well-known exposition of a universalist position in linguistics, describing 54 generalizations that Greenberg felt held true of all (or most) number word systems. While Greenberg did not himself link it to Chomsky’s model of universal grammar, it has been conceptualized as contributing to that body of theory in that it proposes that there are strong and unexpected regularities in language structure (particularly syntactic and morphological ones, in this case) that suggest a universal, hardwired mechanism for their production. While Greenberg presented his rules along with any exceptions he had found, his account is clearly a universalist one in which exceptions are downplayed. Some of Greenberg’s regularities are universal but unsurprising; for instance, that there is no language in the world that has a numeral for two, but no word for one.2 On the other hand, some of his regularities are mind-bogglingly complex, such as the appropriately, ‘oddly’ numbered #37: If a numeral expression contains a complex constituent, then the numerical value of the complex constituent itself in isolation receives either simple lexical expression or is expressed by the same function and in the same phonological shape, except for possible automatic phonological alternations, stress shits, or overt expressions of coordination (Greenberg 1978: 279–280). he fact that Greenberg allows two separate classes of outcome, and three separate classes of exception, suggests that this is not much of a generalization. But other generalizations are exceptionless and nontrivial: for instance, every natural language has a inite set of number words; every known language has a highest number beyond which there is no agreed-upon successor. his does not forestall the possibility that new, higher ones could be developed, but there will always come a point where one will once again reach the end of the system and need to make up yet another highest number. Although it is not one of Greenberg’s generalizations, one might ask: are there, as Funes the Memorious devised, languages in which the word for ive hundred is nine? Or, to put the matter more generally, are there languages that use the same word for two or more numbers? his is very odd to think about — unlike homonyms in other aspects of language, numerical homonyms raise a very strong prospect of ambiguity. Are these in fact homonyms, synonyms, both, or neither? 2. But note that there are English words like googol (1 followed by 100 zeroes) that, if considered part of the numeral system (which Greenberg would not), do violate this generalization — there is no well-formed or accepted number word for ‘one less than a googol’. 560 Stephen Chrisomalis Conversely, is the word for nine, ive hundred? his creates a fundamental semantic ambiguity that leads to surreal statements like here are ive hundred innings in a baseball game. And yet, surprisingly enough, there are languages that use the same word for two or more numbers. he resolution to the ambiguity is that such languages do not use those words in isolation, but in combination with gestures, or pointing to parts of the body, that make the intended meaning clear: as in ‘one, another one, another one, another one’, raising or pointing to a diferent inger each time, as Greenberg (1978: 257) postulates for Kaliana (Sapé), spoken in southern Venezuela. Seiler (1995: 144) notes that several languages have the same word for 1 and 6, 2 and 7, and so on, because the names of the numerals are the names of speciic ingers (as if English used pinkie for both 5 and 10). Such languages are rare, but highly informative. hey highlight the constraint that leads to the generalization — the massive potential for ambiguity that emerges from numerical homonymy — by demonstrating how exceptions deal with that diiculty. Chrisomalis (2010) investigates the constraints of numerical notation systems in a way partially parallel to Greenberg. Numerical notation systems are structured, graphic, principally non-phonetic representation systems for numbers. hey are not essentially linguistic, although they relate to language — they are a very diferent modality, representing numbers graphically rather than orally. While they frequently co-occur with writing systems, writing is neither a necessary nor a suicient condition for the existence and widespread use of numerical notation. While tallying, the use of marks, knots or other signs in one-to-one correspondence with some quantity being counted, may be a precursor to numerical notation structured by a numerical base and its powers, they are not the same. he use of notches or cutmarks on bone, as is now abundantly attested in the Upper Paleolithic archaeological record and potentially earlier (Marshack 1972; Elkins 1996; d’Errico 1998; Hofecker 2007), is fascinating and worthy of attention, but is not numerical notation per se, and falls outside this analysis. More than 100 numerical notation systems are attested to have been used. hey range from the very old, such as proto-cuneiform, developed ca. 3500 BCE (Nissen et al. 1993) to the ultra-new — the Inupiaq numerals developed in 1995 by schoolchildren in Kaktovik, Alaska, and used more widely thereater (Bartley 2002). Some, like the Hindu-Arabic, or Western numerals, are used by billions; others, such as the Cherokee numerals developed by Sequoyah in the early nineteenth century, were rejected almost at the time of their invention (Holmes and Smith 1977; Walker and Sarbaugh 1993). Yet all these attested systems are variants of one of ive basic types (Chrisomalis 2010: 13): – – cumulative-additive systems, like the Roman numerals cumulative-positional systems, like the Babylonian numerals Constraint, cognition, and written numeration 561 – – – ciphered-additive systems, like the Hebrew alphabetic numerals ciphered-positional systems, like the Western (Hindu-Arabic) numerals multiplicative-additive systems, like the Chinese numerals You can imagine others quite easily, but such aberrant systems were never used by anyone as far as can be ascertained. here are powerful cognitive constraints limiting the possibilities for actual, workable numerical notation systems (Chrisomalis 2010: 360–400). here are eighteen strong regularities among these numerical notation systems, a few of which are universal but most of which, while generally valid, have one or more exceptions. Some of these follow Greenberg’s regularities for lexical numerals quite closely: there is no numerical notation system that just skips a number, for instance. Such a thing is not impossible to imagine. Isak Dinesen (1992: 281–3) imagines just such a thing in Out of Africa, having been told by a translator, embarrassed because it had a ‘dubious ring’ in Swedish, that Swahili has no word for nine nor any compound of nine, although it had all the other numerals. However, her wonderment and bemusement at this possibility serves to demonstrate just how unlikely it is that such a circumstance could arise. In other ways, however, the set of regularities in numerical notation diverges from Greenberg’s considerably. For instance, while all lexical numeral systems have a limit beyond which newly invented words are needed to extend the system, many numerical notation systems are truly ininite: they consist of a small set of symbols and a rule-set allowing the construction a numeral representation of indeinite length, without ever having to develop a new symbol. In decimal numerals with place-value like our own Western (Hindu-Arabic) numerals, you can simply add another zero. It is certainly imaginable that English could have a lexical numeral system that simply replaces the digits with the words zero through nine, and thus would become ininite; for instance, if we simply said two six three instead of two hundred and sixty-three. And, of course, English speakers do sometimes use such readings: for instance, when reading phone numbers. However, these digital readings are secondary and derive from graphic notation — in English, they emerged only in the late nineteenth century — and are most oten employed when reading directly, rather than in ordinary conversation. No language has as its basic, ordinary number system anything like this. Similarly, while many lexical numeral systems are not ordered strictly by descending order of powers, numerical notations virtually always are strictly ordered. Strict ordering of linguistic numeral-phrases by descending powers is common, and Miller and Zhu (1991) show that children acquire numeral words with strict power-ordering (such as Mandarin) somewhat more quickly than speakers of languages with irregularities (such as English). Yet there are many exceptions 562 Stephen Chrisomalis to this linguistic generalization — for example, German vierundsiebenzig (4+70) for 74 instead of *siebenzigundvier or in English sixteen (6+10) instead of *ten-six. But in numerical notation systems, strict ordering by powers is virtually always followed.3 his divergence between the two representations of numbers is best explained by the fact that lexical numerals frequently have two morphemes for the same number. hirteen and thirty both are morphologically ‘3 10’ but -teen indicates a number from 11–19 is to be formed through addition (3+10), whereas -ty indicates that a decade from 20–90 is to be formed through multiplication (3x10). In numerical notation, where such morphological variability does not exist, strict ordering is necessary to avoid ambiguity. Numerical notations consist of small sign-inventories of distinct graphemes, and to have multiple signs for 10 would defeat one of their key elements, to communicate numbers rapidly with discrete and easily-diferentiable signs. hus, because of their graphic modality, numerical notation needs power-ordering to a greater degree than lexical numeration. he graphic modality of numerical notation also helps to explain a major structural diference from lexical numerals, namely the much greater role of signrepetition. Repetition (e.g., reduplication) is quite rare in lexical numeral systems, so, for instance, there is no language where three hundred is expressed as ‘hundred hundred hundred’. In contrast, repetition of signs to indicate addition is extremely common in numerical notation — it is the fundamental principle of both nonpositional systems such as the Roman numerals and the Egyptian hieroglyphs and also positional systems such as the Inka khipu and the Babylonian sexagesimal numerals. Virtually all lexical numeral systems use multiplication using powers of one or more numerical bases ‘three hundred’, and while this is sometimes used in numerical notation, it is not the predominant representational mode. Here, the constraint is the visual nature of graphic representations in contrast to the vocalauditory channel used to speak and listen. Not only is ‘ten ten ten ten ten ten one one one one’ much longer than 64, because of the impermanent nature of vocal-auditory signals, such numerals would be diicult for a listener to comprehend and retain, or to distinguish from, say, 54 or 65. With numerical notation, in contrast, because graphic notations are relatively permanent and can be examined in their totality by the reader, a notation like XXXXXXIIII is not too diicult to perceive. Further evidence for this position is found from accommodations of graphic notations to the capacity (common to humans and some other animals) to subitize, or rapidly perceive small numerical quantities as wholes (Mandler and 3. he use of subtractive notation, as in Roman MCMXCIX for 1999, does not violate this principle because the powers (thousands — M, hundreds — CM, tens — XC, ones — IX) are still in order even though, within each power, the signs are inverted. If one could write ICILMXIXC for the Roman numeral 1273, this would be an exception. Constraint, cognition, and written numeration 563 Shebo 1982). For instance, instead of writing 8 as eight lines in a row (––––––––), which presents cognitive challenges to successful enumeration, numerical notation systems tend to group repeated units into easily-readable chunks of three to ive units (= = = =), or, as in the Roman numerals, to add a sub-base at 5 to prevent groupings of ive or more signs. he constraint here is visual and perceptual, and so it operates only on numerical notation, not on lexical numerals. For generativist linguists, universals of verbal numerals (and universals of language more broadly) help to demonstrate that speciic aspects of language are hardwired in the human brain. If they are hardwired, they must be evolutionarily as old as anatomically modern humans, many tens of thousands if not hundreds of thousands of years old (Wiese 2003).4 However, regularities in numerical notation also exist, some in parallel with and some divergent from those that exist for lexical numerals. Because there was no numerical notation before 6000 years ago, when the earliest systems developed alongside early writing, this is far too recent to be an evolved capacity, and the functional association with the origins of writing is not coincidental (Schmandt-Besserat 1992). here are universals, near-universals, and statistical generalizations about numerical notation, just as there are for (presumably UG-related) lexical numerals. In other words, we have two sets of diferent universals for the domain of number, one of which cannot possibly be biologically innate. here are certainly innate evolved capacities (such as subitizing) that relate to numerical notation, but these are not suicient, by themselves, to explain the regularities in question. And if there is no possibility that the regularities of numerical notation are fully innate, there is no reason to assume that the regularities of lexical numerals are fully innate. If there is a universal grammar for lexical numerals, therefore, it is quite plausibly universal in part because of the kinds of non-biological constraints that produce regularities in numerical notation. Because number is a domain where there are two or more diferent representations in many diferent societies, we can investigate why some regularities might be true of number words but not of number symbols, or vice versa. For numerical systems, one of the most crucial constraints is to avoid ambiguity. In most situations, where there are few contextual cues to distinguish the meaning of one number from another in an utterance or text, it is vital to avoid ambiguity. Combining a verbal or graphic representation with gesture, as described above, is one solution. Consistently using a limited set of graphic signs, ordered from highest to lowest power, is another. Comparing these two modalities, we see some constraints that result in true universals across both modalities, others that result in universals in only one or 4. See, however, Andersen 2005 for a counterargument that sees numerals as historically more recent than other word categories. 564 Stephen Chrisomalis the other, but not both, and many others in which we see patterns and strong generalizations but also exceptions. he question of how variable any particular domain is, then, is interrelated with the question of the number and strength of constraints operant in that domain. How, then, do constraint theories help? 5. he 99% problem Nicholas Evans and Stephen Levinson have recently issued a major set of statements against universalism in linguistics that challenges the predominant view that linguistic universals are numerous and important (Evans and Levinson 2009; Levinson and Evans 2010). hey argue that despite over a century of serious work on the question, the number of true, exceptionless universals of language is vanishingly small. And yet, far from retreating into particularism, Evans and Levinson note that a wide variety of features are common to virtually all, but not quite all, known languages. Everett’s example of numeral-less Pirahã is one of them, but certainly not the only one. For instance, Margetts (2007) shows that Saliba, a language of northern Colombia, violates the near-universal that the verb ‘give’ takes three arguments (X gives Y to Z). he study of the rarely attested is a small but important area in contemporary linguistics (Pericliev 2004; Wohlgemuth and Cysouw 2010). his is what we might call ‘the 99% problem’. If something is true of 99% of languages, or 99% of numerical notation systems, or 99% of screwdrivers, but there are one or two exceptions, what do we do with this fact? For a universalist, the answer is easy: ignore the exception. Ater all, we all ignore exceptions and outliers in our data all the time. For a particularist, the answer is also easy: use the exception as a demonstration of variability in the face of the apparently universal. But this hardly seems satisfactory as an accounting of the degree of variability in human societies. For a comparativist, however, such a situation invites an opportunity — and a challenge. For Evans and Levinson, near-universals (strong but not quite exceptionless regularities) are evidence of “stable engineering solutions satisfying multiple design constraints, relecting both cultural-historical factors and the constraints of human cognition” (Evans and Levinson 2009: 1). For any phenomenon you might want to look at, there are multiple constraints, some of which are inevitably going to be in conlict with one another, and multiple solutions, which weight some of those constraints more heavily than others. For each phenomenon, we must remain agnostic about whether these constraints are likely to be stringent or lax, whether they will be numerous or few, and what particular form they will take. We must also remain aware that any situation with multiple constraints is likely to produce exceptions, and to take those exceptions seriously in our analysis. Constraint, cognition, and written numeration 565 Within a constraint theory, in the literal and original sense of the phrase, the exception proves the rule — not ‘prove’ in the sense of demonstrate it to be true, but prove as in ‘proving ground’. Recall that the Latin version of the saying is exceptio probat regulam — the exception probes the rule might be a better translation. It does not deny the rule — rather, it clariies its scope and conirms its general validity otherwise. So, for instance, if you park your car beside a sign that says “No parking 8:30 am — 5:30 pm”, the time prohibition equally conirms that at other times, parking is permitted. And if you were to see that sign and think to yourself, “Why those times?” you would probably be able to reason that those are business hours. Similarly, for cultural phenomena, analyzing exceptions to general patterns is a powerful analytical tool for understanding why the pattern emerges in the irst place. he exception probes the rule, and also helps rationalize its existence, by explaining cases when it doesn’t apply. Exceptionless universals, where they exist, are important but do not provide any clue as to why they might be true. In Donald Brown’s important study of human universals, he notes quite rightly that universals are not necessarily hardwired or inevitable — for instance, that there may be ‘new universals’ such as tobacco and metal tools, as well as ‘former universals’ such as breastfeeding infants (Brown 1991: 50). But we only know that these categories are interesting because we are aware of attested societies (past or present, respectively) where the universal does not hold. hus, near-universals — widespread generalizations with a few exceptions — are extraordinarily important (Dryer 1997, 2003), because they allow anthropologists and linguists to probe for explanations without having to go to other disciplines (such as psychology and biology) to explain the broader patterns being observed. I thus disagree with Brown’s (1991: 45) claim that, “he distinction between a near-universal and a universal (or absolute universal), then, is not signiicant, or imposes an artiicial break in an unbroken natural continuum. A near-universal is universal enough.” his reasoning promotes the idea that exceptions are uninteresting or meant to be ignored. When evaluating rarities in numerical notations, a further consideration is that some kinds of numerical notation systems are easy to imagine, and in fact have been invented, but their structures lead them to be quickly rejected. Many of these systems in antiquity simply will not have survived; in other words, their rarity makes their archaeological or archival preservation unlikely. When we look in the modern era, we ind many such systems, simply because preservation is so much better. Take, for instance, the Codex Seraphinianus, Luigi Seraini’s ictional surrealist encyclopedia (Seraini 1983). Its script, though presenting the appearance of decipherability, is meaningless, and it is illed with arcane diagrams of impossibilities. Yet its written numerals can be securely deciphered, because like all numerical notation systems, the Seraphinianus system follows patterned rules that 566 Stephen Chrisomalis allow one to predict the next in the sequence, and because it is paginated consistently with enumerated tables of contents. Yet the Seraphinianus numerals violate several of the regularities that apply to numerical systems that are actually used for any amount of time in actual social contexts (for instance, it has a base of 21). It appears, then, that the constraints on human imagination are far less than the constraints on human behavior. If we had perfect knowledge of the past, we might ind dozens, possibly even hundreds of structurally anomalous numerical notations that were rapidly invented and rapidly discarded. he constraints that move us from the imaginable to the attested are primarily cognitive and functional. Only ive general types of numerical notation are, as Evans and Levinson call them, ‘stable engineering solutions’. Systems that vary from these signiicantly either rapidly converge to one of those ive, or else disappear entirely. here is still plenty of variability, because humans are relatively lexible. As the social anthropologist Christopher Hallpike puts it, it is not about survival of the ittest, but rather ‘survival of the mediocre’ (Hallpike 1986: 81–145).5 here is no one optimal solution, but a variety of workable solutions, and only the considerably unworkable are discarded, unused. 6. Constraint, history, and cognition A further advantage of a constraint-based approach to written numeration and cognition is that it permits us to investigate the role of non-cognitive constraints on the history of numerical notations. Here I again follow Evans and Levinson (2009), who similarly insist that not only cognitive constraints but also social, cultural, and historical constraints afect the distribution of linguistic features worldwide. In particular, for numerical notation systems, if we hope to know anything about the way that numerical notations, once invented, are adopted, transformed, and relate to cognitive activities such as arithmetic, we need to know not only what the structure of a system is, but how it was intended to be used, and how those functions changed over time. his parallels the general distinction between Goody (1977) and Olson (1994) on the one hand, and Scribner and Cole (1981) on the other hand, with respect to the consequences of literacy. Just as we recognize that a toddler and a mathematician may both use the same set of numerals 0–9, but with enormously diferent cognitive outcomes, we should not presume that any particular notation automatically or universally has any particular cognitive efect. Once again, we return to the deinition of constraints as processes that limit but do 5. Or, as Brooks and Wiley (1988: 103) call it, slightly more optimistically, ‘survival of the adequate’. Constraint, cognition, and written numeration 567 not specify a particular outcome as a solution. We can then analyze how variability in their features is selected for or against in particular contexts of use, and understand how, within the general constraints ofered by universal cognitive processes and environmental preferences, speciic numerical notations may lourish (or not) under particular social and historical constraints. Let us return to the Roman numerals for a moment, and consider the widespread assumption that they were replaced because they are characteristic of an earlier stage of human cognition — as Dehaene (2011: 66) calls them, a ‘living fossil’ — or at worst, as Murray (1978) asserts, that they actively inhibit mathematical cognition. his claim is seen frequently in the cognitive literature on numeracy (e.g. Zhang and Norman 1995) under the guise of the assumption that it is reasonable to compare numerical notations on the basis of their eiciency for written computations. Even if we grant that Roman numerals could not have been used as the basis for higher mathematics — a claim on which Schlimm and Neth (2008) cast some doubt — we must contend with the fact that they thrived in Western Europe for over two thousand years, the last ive hundred of which they coexisted with the Hindu-Arabic (Western) numerals. In fact, to know why the Roman numerals were replaced, for instance, we cannot simply compare their structure to that of their successor, the Hindu-Arabic numerals, but we must also ask for what functions it was used — for instance, the Roman numerals were never used for written arithmetic, as the Romans had the perfectly serviceable abacus in its place. Only when written arithmetic, associated with widespread literacy and the printed book, became feasible and common, was the Roman system vulnerable to being replaced by the Hindu-Arabic system in Western Europe; it was simply too well-entrenched and the modes and technologies of its use too well-known. For any communication system to be useful (whether an alphabet, or a music notation system, or a system of traic signs), it needs to be shared in common by a body of users, and one reason a system may be rejected is that it is perceived to be awkward for some function, which might be arithmetical but might be something else entirely. In the case of Hindu-Arabic numerals, xenophobia may have played a role, but also the fact that the zero was considered particularly enigmatic and diicult to work with.6 Another is that there is a pre-existing system, and abandoning it would mean that you could no longer communicate well with those who do not know the new system. Either factor is a constraint, but we cannot predict which will be most important in any particular case. I have serious doubts that it is ever possible to predict the cognitive efects of a numerical notation directly from its structure. Damerow’s (1996) careful work on 6. his perplexity is demonstrated by the European adoption of the Arabic word sifr ‘zero’ not only as zero, but also as cipher ‘secret mode of writing’. 568 Stephen Chrisomalis archaic Mesopotamian numeration is the most consistent and thorough attempt to argue that the stages of the evolution of notation parallel cognitive stages, in this case from a developmental-psychological perspective. For instance, the presence of multiple (perhaps as many as 30) diferent numerical notations in the Uruk period (ca. 3000 BCE) is seen as evidence of the absence of abstract number concepts, because there was no way to refer to an abstract number without simultaneously designating some concrete object being enumerated (grain, cattle, people, etc.). However, the lexical evidence from Sumerian suggests that there was, even at this early date, a single, perfectly ordinary set of number words. Taken literally, this would imply that nonliterate Sumerians with their single abstract number system were more abstract in their numerical cognition than literate ones — a position Damerow surely does not wish to take. A far more reasonable explanation is that, just as we count time in base-60 and use hexadecimal (base-16) notation in computing, the technical needs of administering the Uruk city-state made it advantageous to use diferent notations that had diferent systems of weights and measures. However, to say that one cannot simply read cognitive efects from written number notations is not to say that there are no such efects, but rather to demand further empirical investigation, including social context. Goody (1977: 85– 89) rightly emphasizes the importance of numbered lists as a cognitive tool for organizing information that better permits the visual inspection of knowledge. Indeed, while arithmetic is cross-culturally quite a rare function of written numerals, and thus not a good candidate for identifying cognitive efects, list-making is a widespread phenomenon in which written numerals are employed. Potentially — though this is not yet demonstrated — number systems that more readily allow the reader to evaluate the size of numbers are better suited for such a function. Because every three-digit Hindu-Arabic numeral is larger than every two-digit numeral, one can use the length of numeral-phrases as a proxy for numerical size. For Roman numerals, it is not immediately apparent that CXXI is larger than CXVIII. While this does not explain the replacement of Roman numerals, it does suggest fruitful avenues for research. Because the cognitive literature to date has focused almost exclusively on Hindu-Arabic numerals, there has been no opportunity to consider these potentially important efects. In a global network of societies in which there is a strong interest, particularly among elites, in science, commerce, technology, and education, social and communicative constraints make the adoption of a standard notation to facilitate social networks very likely, though not inevitable, and even systems of great longevity can be replaced very rapidly. his is the case not only with the Roman numerals, but many other systems, which were rendered obsolete at the advent of the modern world-system between 1450 and 1650 (Chrisomalis 2010: 423). his was Constraint, cognition, and written numeration 569 a period which saw a wide range of newly literate users of written numbers (e.g. the mercantile professions) choosing the Hindu-Arabic numerals because of their speciic utility for pen-and-paper computation, and the role of written arithmetic texts in transmitting this information. An abacus is a remarkable computational tool, but it leaves no trace of immediate steps in computation, which is hardly advantageous to commercial enterprises on a national or global scale. he fact that Roman numerals are unsuited for written computation was irrelevant until such time as written arithmetic became a highly desirable function in the early modern capitalist social context. Once that had occurred, however, the Roman numerals declined in use quite rapidly. he historical pattern of invention and decline of numerical notations shows that in some key ways, the past ive hundred years (or, should I say, following Funes, nine years?) may be unlike the preceding ive thousand years. Most notably, the massive integration of societies has made it very dificult for novel numerical systems to gain a foothold due to these newly important social constraints, which exist alongside cognitive and graphic/notational ones. Yet, even though numerical notations are disappearing more rapidly than in the past, the systems that remain, and recently-invented and used systems, conform generally to the structural principles observed in the pre-modern world. To abandon the study of the cognitive and structural constraints governing written numerals, and to view the choices underlying the adoption, transformation, and decline of numerical notations (and scripts) as motivated solely by social constraints, would miss the point. We should remain attentive to the fact that numerals are used by people, in real social contexts, for a variety of reasons. In fact, only by doing so can we integrate the two perspectives outlined at the beginning of this essay: how cognitive factors constrain numerical systems, and then how in turn numerical systems constrain cognitive outcomes. he next and vital step is to extend this work cross-culturally and cross-linguistically to generate testable general theories for the relationship of cognition and representational systems as a central aspect of the human experience of mathematical ideas. References Andersen, H. 2005. “he plasticity of universal grammar”. In W. Østreng (ed), Convergence: Interdisciplinary Communications 2004/2005. Oslo: Centre for Advanced Studies at the Norwegian Academy of Sciences and Letters, 21–26. Bartley, W.C. 2002. “Counting on tradition: Iñupiaq numbers in the school setting”. In J.E. Hankes and G.R. Fast (eds), Perspectives on Indigenous People of North America. Reston, VA: National Council of Teachers of Mathematics, 225–236. Boas, F. 1896. “he limitations of the comparative method of anthropology”. Science 4: 901–908. DOI: 10.1126/science.4.103.901 570 Stephen Chrisomalis Borges, J.L. 1964. Labyrinths. D. A. Yates and J. E. Irby (eds). New York: New Directions. Brooks, D.R. and Wiley, E.O. 1988. Evolution as Entropy: Toward a Uniied heory of Biology. Chicago: University of Chicago Press. Brown, C.H. 1976. “General principles of human anatomical partonomy and speculations on the growth of partonomic nomenclature”. American Ethnologist 3(3): 400–424. DOI: 10.1525/ae.1976.3.3.02a00020 Brown, C.H. and Witkowski, S.R. 1981. “Figurative language in a universalist perspective”. American Ethnologist 8(3): 596–615. DOI: 10.1525/ae.1981.8.3.02a00110 Brown, D.E. 1991. Human Universals. New York: McGraw-Hill. Chrisomalis, S. 2010. Numerical Notation: A Comparative History. New York: Cambridge University Press. DOI: 10.1017/CBO9780511676062 d’Errico, F. 1998. “Palaeolithic origins of artiicial memory systems: An evolutionary perspective”. In C. Renfrew and C. Scarre (eds), Cognition and Material Culture: he Archaeology of Symbolic Storage. Cambridge: McDonald Institute for Archaeological Research, 19–50. Damerow, P. 1996. Abstraction and Representation: Essays on the Cultural Evolution of hinking. Dordrecht: Kluwer Academic Publishers. DOI: 10.1007/978-94-015-8624-5 Dehaene, S. 2011. he Number Sense. 2nd edition. Oxford: Oxford University Press. de Cruz, H., Neth, H., and Schlimm, D. 2010. “he cognitive basis of arithmetic”. In B. Löwe and T. Müller (eds), PhiMSAMP. Philosophy of Mathematics: Sociological Aspects and Mathematical Practice. Texts in Philosophy 11. London: College Publications, 59–106. Dinesen, I. 1992. Out of Africa. New York: Modern Library. Dryer, M. 1997. “Why statistical universals are better than absolute universals”. Proceedings, Chicago Linguistic Society 33:123–145. Dryer, M. 2003. “Signiicant and non-signiicant implicational universals”. Linguistic Typology 7(1): 108–128. DOI: 10.1515/lity.2003.007 Elkins, J. 1996. “he impossibility of close reading”. Current Anthropology 37(2): 185–226. DOI: 10.1086/204490 Evans, N. and Levinson, S.C. 2009. “he myth of language universals: Language diversity and its importance for cognitive science”. Behavioral and Brain Sciences 32(5): 429–448. DOI: 10.1017/S0140525X0999094X Everett, D.L. 2005. “Cultural constraints on grammar and cognition in Pirahã”. Current Anthropology 46(4): 621–646. DOI: 10.1086/431525 Frank, M.C., Everett, D.L., Fedorenko, E., and Gibson, E. 2008. “Number as a cognitive technology: Evidence from Pirahã language and cognition”. Cognition 108(3): 819–824. DOI: 10.1016/j.cognition.2008.04.007 Gardner, M. 1956. Mathematics, Magic and Mystery. New York: Dover Publications. Gibson, J.J. 1979. he Ecological Approach to Visual Perception. Boston: Houghton Milin. Goldenweiser, A. 1913. “he principle of limited possibilities in the development of culture”. Journal of American Folklore 26(101): 259–290. DOI: 10.2307/534816 Goldenweiser, A. 1942. Anthropology: An Introduction to Primitive Culture. New York: F.S. Crots. Goody, J. 1977. he Domestication of the Savage Mind. Cambridge: Cambridge University Press. Goody, J. 1986. he Logic of Writing and the Organization of Society. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511621598 Greenberg, J.H. 1978. “Generalizations about numeral systems”. In J. H. Greenberg (ed), Universals of Human Language. Stanford: Stanford University Press, 249–297. Hallpike, C.R. 1986. he Principles of Social Evolution. Oxford: Clarendon. Constraint, cognition, and written numeration 571 Hofecker, J.F. 2007. “Representation and recursion in the archaeological record”. Journal of Archaeological Method and heory 14(4): 359–387. DOI: 10.1007/s10816-007-9041-5 Holmes, R.B. and Smith, B.S. 1977. Beginning Cherokee. 2nd ed. Norman: University of Oklahoma Press. Levinson, S.C. and Evans, N. 2010. “Time for a sea-change in linguistics: Response to comments on he Myth of Language Universals”. Lingua 120(12): 2733–2758. DOI: 10.1016/j.lingua.2010.08.001 Mandler, G. and Shebo, B.J. 1982. “Subitizing: an analysis of its component processes”. Journal of Experimental Psychology: General 111(1): 1–22. DOI: 10.1037/0096-3445.111.1.1 Margetts, A. 2007. “Learning verbs without boots and straps? he problem of ‘give’ in Saliba”. In M. Bowerman and P. Brown (eds), Cross-linguistic Perspectives on Argument Structure: Implications for Learnability. Mahwah, NJ: Lawrence Erlbaum, 111–41. Marshack, A. 1972. “Upper Paleolithic notation and symbol”. Science 178: 817–828. DOI: 10.1126/science.178.4063.817 Miller, K.F. and Jianjun Zhu. 1991. “he trouble with teens: Accessing the structure of number names”. Journal of Memory and Language 30(1): 48–68. DOI: 10.1016/0749-596X(91)90010-H Murray, A. 1978. Reason and Society in the Middle Ages. Oxford: Clarendon Press. Nevins, A., Pesetsky, D., and Rodrigues, C. 2009. “Pirahã exceptionality: A reassessment”. Language 85(2): 355–404. DOI: 10.1353/lan.0.0107 Nissen, H.J., Damerow, P., and Englund, R.K. 1993. Archaic Bookkeeping: Early Writing and Techniques of Economic Administration in the Ancient Near East. Chicago: University of Chicago Press. Olson, D.R. 1994. he World on Paper: he Conceptual and Cognitive Implications of Writing and Reading. Cambridge: Cambridge University Press. Pericliev, V. 2004. “Universals, their violation and the notion of phonologically peculiar languages”. Journal of Universal Language 5: 1–28. Prince, A. and Smolensky, P. 1993. Optimality heory: Constraint Interaction in Generative Grammar. Report RuCCS-TR-2, Rutgers University Center for Cognitive Science, New Brunswick, NJ. Reboul, A. 2012. “Language: Between cognition, communication and culture”. Pragmatics & Cognition 20(2): 295–316. DOI: 10.1075/pc.20.2.06reb Sapir, E. 1921. Language: An Introduction to the Study of Speech. New York: Harcourt, Brace and Company. Schlimm, D. and Neth, H. 2008. “Modeling ancient and modern arithmetic practices: Addition and multiplication with Arabic and Roman numerals”. In V. Sloutsky, B. Love and K. McRae (eds), Proceedings of the 30th annual meeting of the Cognitive Science Society, Austin, TX: Cognitive Science Society, 2097–2102. Schmandt-Besserat, D. 1992. Before Writing: From Counting to Cuneiform. Austin: University of Texas Press. Scribner, S. and Cole, M. 1981. he Psychology of Literacy. Cambridge, MA: Harvard University Press. DOI: 10.4159/harvard.9780674433014 Seiler, H. 1995. “Iconicity between indicativity and predicativity”. In S. Rafaele (ed), Iconicity in Language. Amsterdam: John Benjamins, 141–151. DOI: 10.1075/cilt.110.12sei Seraini, L. 1983. Codex Seraphinianus. 1st American edition. New York: Abbeville Press. Tagliavini, C. 1949. “Di alcune denominazioni della papilla”. Annali dell’Instituto Universitario di Napoli NS 3: 341–378. 572 Stephen Chrisomalis Trigger, B.G. 1991. “Distinguished lecture in archeology: Constraint and freedom — a new synthesis for archeological explanation”. American Anthropologist 93(3): 551–569. DOI: 10.1525/aa.1991.93.3.02a00010 Walker, W. and Sarbaugh, J. 1993. “he early history of the Cherokee syllabary”. Ethnohistory 40(1): 70–94. DOI: 10.2307/482159 Watson, R.A. 1976. “Laws, systems, certainty, and particularities”. American Anthropologist 78(2): 341–344. DOI: 10.1525/aa.1976.78.2.02a00140 Wiese, H. 2003. Numbers, Language, and the Human Mind. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511486562 Wohlgemuth, J. and Cysouw, M. (eds). 2010. Rethinking Universals: How Rarities Afect Linguistic heory. Mouton: de Gruyter. DOI: 10.1515/9783110220933 Zhang, J. and Norman, D.A. 1995. “A representational analysis of numeration systems”. Cognition 57: 271–295. DOI: 10.1016/0010-0277(95)00674-3 Author’s address Stephen Chrisomalis Department of Anthropology Wayne State University 3054 F/AB, 656 Reuther Mall Detroit, MI 48202 USA chrisomalis@wayne.edu http://clas.wayne.edu/chrisomalis About the author Stephen Chrisomalis earned his Ph.D. from McGill University in 2003 and is presently Associate Professor of Anthropology at Wayne State University in Detroit, Michigan. He is the author of Numerical Notation: A Comparative History (Cambridge, 2010). He specializes in the cross-cultural cognitive anthropology of mathematics and numeral systems, the anthropology of writing and literacy, and linguistic anthropology. Copyright of Pragmatics & Cognition is the property of John Benjamins Publishing Co. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.