Economics and English: language growth in economic perspective.

Tollison , Robert D.

And the LORD said, "Look they are one people, and they have all one language; and this is only the beginning of what they will do: nothing that they propose to do will now be impossible for them. Come, let us go down, and confuse their language there, so that they will not understand one another's speech." (Genesis 11:6-7)

The bond of language is perhaps the strongest and most lasting that can unite men. (Tocqueville [1835, 1840] 2000, p. 29)

I am a friend to neology. It is the only way to give to a language copiousness and euphony. Without it we should still be held to the vocabulary of Alfred or of Ulphilas: and held to their science also;... (Thomas Jefferson to John Adams, August 15, 1820; emphasis in the original)

POLONIUS: What do you read, my lord? HAMLET: Words, words, words. (Act 11.2)

1. Introduction

Language use is a distinguishing characteristic of Homo sapiens. (1) Human beings are programmed to construct hundreds of thousands of words from a comparative handful of sounds. a phenomenon known to linguists as duality (Pinker 1996).

From a purely economic perspective, language can be seen as an exchange-facilitating institution, just as important in its own fight as other, more deeply studied underpinnings of orderly markets, such as clearly defined property rights and rules for enforcing contracts. Exchange, of course, is not the only economic function of language since language can be used for many purposes, including coercion ("arrest that man!"). The paradigm of exchange has nonetheless proven analytically fruitful and, indeed, has been applied by economists to help explain the functioning of political markets, patterns of trade within the family, crime, drug addiction, and many other seemingly "noneconomic" aspects of human behavior (e.g., Becker 1976, 1996; Becker and Murphy 2001). Viewing language as an exchange-friendly institution locates the study of its growth and impact squarely within the singular perspective of economic science. (2) This analytical framework in no way diminishes the importance of more traditional approaches to the subject; it simply calls attention to an emerging literature in which economists have begun colonizing the study of language. (3)

Language, as Friedrich von Hayek (1960) stressed repeatedly, is an example of spontaneous order. In other words, language was not created deliberately; it evolved in the absence of human plan or conscious design. Viewed in that perspective, the evolution of language has much in common with that of another exchange-friendly institution--money. Both of these institutions emerged spontaneously as rational individuals searched for mutually acceptable ways of communicating and of acting on their natural propensities to "truck, barter, and exchange one thing for another" (Smith [1776] 1976, bk. I, chap. II, p. 17). Over time, independent responses to novel situations produced words and rules for their use that facilitated communication in much the same way that attempts to cope with the rarity of dual coincidences of wants led to the selection of one commodity as money. As Hayek (1988, p. 106) puts it,

 Trade, migration, and the increase and mixture of population must
 not only have opened people's eyes, but also loosened their
 tongues. It was not simply that tradesmen inevitably encountered,
 and sometimes mastered, foreign languages during their travels, but
 that this must have forced them also to ponder the different
 connotations of key words (if only to avoid either affronting their
 hosts or misunderstanding the terms of agreements to exchange), and
 thereby come to know new and different views about the most basic
 matters. (4)

What materialized over millennia as a result of competition among alternative media of exchange and means of communication were "efficient" monies and languages.

In modern parlance, language is the archetypal "network" good (i.e., it exhibits positive externalities in consumption). (5) Unlike money and most ordinary goods, language derives value not from scarcity but from ubiquity. Words and rules for their use become more valuable the greater is the number of people who learn and apply them consistently in everyday discourse. (6)

Language, like money, is not static. "Many different commodities," including cattle, salt, sea shells, and tobacco, "were successively both thought of and employed" as units of account, stores of value, and media of exchange (Smith [1776] 1976, bk. I, chap. IV, p, 27). New words are likewise coined or borrowed from other tongues, old words take on new meanings or become obsolete, and usage rules change as the circumstances of time and place change. In this article, we take a Hayekian or evolutionary approach to describe the development of the English language. To do so, we exploit the CD-ROM version of the second edition of the Oxford English Dictionary (OED), which among its many other valuable features identifies the date of the earliest recorded use of each word by means of an illustrative quotation. (7) These dated quotations allow us to construct an annual time series of words, beginning with the year 252 CE and ending in 1985.

English is, of course, not the world's only language. Spoken today by more than 300 million people (Bryson 1990, p. 11), it ranks a distant second to Mandarin Chinese, which counts some 750 million speakers (Bryson 1990, p. 181). English is, nevertheless, the official language of 44 countries (compared to 27 for French and 20 for Spanish), populated by 1.6 billion souls, or about one-third of humankind (Bryson 1990, p. 181). The number of English words in common use (roughly 200,000) exceeds that of German (184,000) and is double that of French (Bryson 1990, p. 13). What is more important for our purposes is that no other language has a dictionary comparable to the OED. Lacking alphabets, traditional Chinese and Japanese dictionaries are organized on the basis of radicals (semantically significant pen strokes used in composing other characters). And, as discussed here, English dictionary writers have never been subject to the stultifying regulation of a language purifier like the Academic Francaise.

We fully recognize that the procedure adopted by the editors of the OED for dating words in use at particular times is far from perfect. Dr. James Murray, the editor in chief of the dictionary's 10-volume first edition, published in 1928, began soliciting quotations in 1858 from an army of some 1,300 (eventually 2,300) scholars (Bate [1975] 1988, p. 251), including, most famously, convicted murderer Dr. W. C. Minor (Winchester 1998). Murray then selected from an estimated 5 million such submissions the earliest quotation best illustrating the meaning of each of the 252,200 entries ultimately included in the OED. (8) A similar procedure was adopted when work on the 20-volume second edition began. Nearly 2.5 million quotations were chosen to illustrate the meanings of the second edition's 291,500 headwords, which incorporated material published in a four-volume supplement to the first edition that appeared seriatim between 1972 and 1986. (9)

Words obviously had been adopted by speakers of the English language prior to their first appearance in print: "Dictionaries are but the repositories of words already legitimated by usage." (10) Just as obviously, it may well be possible to find quotations illustrating word meanings that pre-date the written passages selected for inclusion in the OED. (11) Even the definition of the word "word" is somewhat muddy: the first dated entry in the dictionary's second edition is in fact not a singular word but the compound noun construction "common prayer." Moreover, the greatest of the modern English dictionaries does not define large numbers of scientific and technical terms: it even fails to document fully the language's richness. Despite these ambiguities and omissions, the OED supplies an internally consistent data source for dating the origins of the words that serve as inputs to spoken and written English. One has to start somewhere. We start with the undisputed authority, the OED.

This article is organized as follows. Descriptive statistical analyses of the time series of words added to the English language since 252 CE are presented in the next section. (12) Section 3 advances some testable hypotheses about the economic forces that might help explain the observed growth of the language over time. There we report evidence from the 19th and 20th centuries that the production of new English words is characterized by increasing returns with respect to both population and national wealth and that neology is a function of foreign trade and the size of government. Section 4 summarizes and concludes the article.

2. The Growth of the English Language

Words are the building blocks of language but are not themselves language: "People do not just blurt out isolated words but rather combine them into phrases and sentences, in which the meaning of the combination can be inferred from the meanings of the words and the way they are arranged" (Pinker 1999, p. 4). Linguists call the rules for assembling meaningful combinations of words a grammar, the most elementary building blocks of which are noun phrases, joining a determiner, such as "a" or "the," with a noun, and verb phrases, joining a verb with its direct object, a noun phrase. A sentence, in turn, is composed of a noun phrase followed by a verb phrase. Hence, while the word "rose" is not language, it is an essential ingredient in the building of noun phrases, such as "a rose"; verb phrases, such as "is a rose"; and sentences, such as "a rose is a rose" (Pinker 1999, pp. 4-5). (13)

From an economist's perspective, words can be likened to "inputs," while a language's grammar is a "production function" that chooses the best set of inputs and rules for combining them to produce effective communication. Although there are exceptions to Pinker's assertion that isolated words are not language ("fire!"), studying changes in the input domain of the English language can nonetheless shed light on its historical development as a rich and increasingly globally accepted method of communication. (14)

The second edition of the OED uses a total of 2,436,600 quotations to illustrate word meanings. Approximately 25,000 of these authorities are taken from the Bible, the most frequently quoted work, and 33,300 of them are from the works of William Shakespeare, the most frequently quoted author (Hamlet alone is quoted nearly 1,600 times). (15) With 524 quotations, Adam Smith stands first among the economists quoted in the OED. John Stuart Mill follows with 351 quotations, although some of the attributions to him seem doubtful. Other often-quoted economists include John Maynard Keynes (79 quotations), W. Stanley Jevons (21 quotations), Joseph A. Schumpeter (13 quotations), and Francis Y. Edgeworth and John Bates Clark (eight quotations each). (16)

Table 1 shows the distribution of illustrative quotations over time. Nineteenth-century authorities account for nearly one-third of them.

Total Words

The search engine contained on the CD-ROM version of the OED allows one to identify the earliest quotation used to illustrate the meaning of each of the dictionary's main entries. Such a search produces dated quotations for 230,719 words, beginning with "common prayer" in 252 CE and ending with "hoolivan" (slang for a type of police vehicle carrying photographic and video equipment for observing crowd behavior and identifying troublemakers at football matches and other public events) in 1985. Table 2 shows the distribution of new words entering the English language from the 1st through the 20th centuries.

Some 36,589 words (15.9% of the total) pre-date the invention of the movable-type printing press by Johan Gutenberg (ca. 1455), and 136,447 (59.1%) of them were in use before Samuel Johnson completed his monumental Dictionary of the English Language, the first modern lexicon, published in 1755, containing 43,500 headwords and 118,000 illustrative quotations (Winchester 1998, p. 96). (17) Figure 1 supplies additional details. Displayed there are the total number of words added to the English language, by year, from 252 through 1985 CE, and the cumulative stock of words in place in each of those years. Two characteristics of this data set warrant clarification. First, both of the series shown in Figure 1 include all words added to the English language since the third century, including those subsequently falling out of use. According to the publishers of the OED, obsolete words account for 47,100 of the dictionary's main entries (there are an additional 240 entries for "spurious words"). (18) Because the OED's main search engine does not allow one to identify such words and to determine their dates of obsolescence, we are unable to net them out. (19) Second, we have not attempted to distinguish among multiple dictionary entries assigning different meanings to the same word. "Serve," for example, is listed four times, defined alternately as the fruit of the service tree, a noun; adoration or the act of serving, as in badminton or tennis, a noun; to be a servant, and the many variations thereof, a verb; and to deserve, a verb. We do not think this is double (or even quadruple) counting. Indeed, "'serve" has many more than four meanings--the third entry covers the gamut from waiting on tables to copulation. And so, if anything, our data set underestimates the richness of the English language.

[FIGURE 1 OMITTED]

Except for the spike at the dawn of the Common Era's second millennium (the year 1000). Figure 1 suggests that the stock of words in the English language was essentially unchanged until about 1300 CE, at which point the number of words began to increase rapidly and continuously, The language's average annual growth rate is slightly less than 1% (0.98) overall, with a median of four one-hundredths of 1% (0.04). Table 3 shows these growth rates by century.

The frequency distribution of English words is shown in Figure 2. There we see that in almost 1400 of the 1734 years of our data, 250 words or less were added to the language. Indeed, the mean number of new words added from 252 to 1985 is 133 per year, with a median of two words. (20) (Beginning in 1300, the annual average grows to nearly 318 words per year, with a median of 243 words.) Annual averages, by century, are displayed in Table 4.

[FIGURE 2 OMITTED]

The last two tables suggest that, gauged in terms of total words, the 19th century was the most lexicographically rich period in the history of the English language. (21) The highest growth rates were sustained earlier, though, particularly so from the 13th through 17th centuries. One of the most fertile periods of language growth transpired in the years closing the 16th century and opening the 17th: 923 words were added to the language in 1597, and more than 1000 words were added in each of the next three years. The sharp "spike" evident in Figure 1 shortly thereafter occurs in 1611, when the stock of words increased by 2638.

Although we make some attempt to estimate spare models explaining the time series of words in section 3, this period illustrates well the difficulty of disentangling the constellation of forces, both economic and otherwise, that shaped the development of the English language. On the one hand, England's foreign trade sector grew spectacularly during the reigns of the last monarch of the Tudor lineage, Queen Elizabeth, and the first of the Stuarts, James I. Customs receipts amounted to a modest 50,000 [pounds sterling] in 1590; they were 148,000 [pounds sterling] in 1613 and 323,000 [pounds sterling] in 1623 (Pipes 1999, p. 140), a year in which more than 1100 words were added to the language. William Shakespeare (1564-1616), perhaps the single most prolific coiner of English words and the authority most often quoted in the OED, was at the height of his powers during the same period. (22) There evidently can be no monocausal theory of language growth.

Parts of Speech

The total number of words originating in any given year can be categorized according to various parts of speech. (23) The OED's search engine allows us to identify 15 such categories: phrases, nouns, verbal nouns, pronouns, adjectives, participial adjectives, verbs, participles, adverbs, prepositions, conjunctions, particles, interjections, prefixes, and suffixes. (The data are shown in Appendix A.)

Not surprisingly, nouns--words that name or designate persons, places, things, actions, or qualities--account for over half of the additions to the English language in any given century. The impact of technological innovation, trade, and economic growth on the English-speaking world is evident in the rising share of nouns in new words added during the past two or three centuries: they make up almost 58% of the words originating in the 19th century and more than 70% of the words originating in the 20th.

Adjectives--words designating attributes and modifying or qualifying nouns to describe persons, places, or things more fully--rank next in importance, accounting for nearly 21% of the additions to the English language since the Common Era's first century. Action words--verbs--rank third (making up 12.3% of the total), followed by participial adjectives, such as "burnt," "cutting," or "engaged" (5.9% of the total); (24) adverbs, which qualify or modify another word, especially an adjective, a verb, or another adverb, so as to express a relation of place, time, circumstance, and so on (3.3%); and verbal nouns, naming words formed as inflections of verbs (2.4%). Pronouns, participles, prepositions, conjunctions, particles, interjections, and prefixes and suffixes--most of which were in place early on--account for the remainder.

In sum, descriptive words (nouns and adjectives) and action words (verbs), which together make up the building blocks of grammatical sentences, have been the engine of growth of the English language. (25) This is especially true of nouns, over half of which have been added to the tongue during the past 300 years. By means of importing and adapting words from other languages and coining of new words, the discoveries of new things and contact with new places and new peoples have enriched the English lexicon. It remains to be seen whether that growth is susceptible to systematic empirical explanation, a task to which we turn in section 3.

Obsolete Words

The CD-ROM version of the two-volume Shorter Oxford English Dictionary (SOED), which contains a total of 97,209 headwords, allows one to search both for obsolete words and for obsolete senses of words and to date their last appearances in print within intervals of 30 to 40 years. Such a search yields 2500 whole words and 17,643 meanings of words that dropped out of use over the period running from the time when Old English was the standard (prior to 1149 CE) through the late 20th century.

The raw data, which are reported in Appendix B, indicate that most of the words that disappeared from the English language did so during the 300 years commencing in the late 16th century (1570-1599) and ending in the late 19th century (1870-1899). The data also suggest that changes in the meanings of words, rather than additions or subtractions of whole words, account for the bulk of the net growth of the language over time. "Algebra" supplies a useful illustration. Achieving currency in the period of Late Middle English (1350-1469), the word formed by combining the prefix "al-" with the Arabic root jabr, which referred to the reunion of broken parts, originally meant "the surgical treatment of fractures." That meaning became obsolete in the middle of the 16th century (1530-1569) and was replaced by algebra's modern definition as a branch of mathematics. (26)

Appendix C reports two alternative ways of measuring the rate of decay of the English language over time. Obsolete whole words (and senses of words) are computed as percentages of words first used during the corresponding period and of the stock of words in use during the same period. The early 18th century (1700-1729) stands out on both of these measures of obsolescence: the 403 words whose last use can be traced to that period represent almost 17% of the new words added to the language and nearly three-quarters of 1% of the total words then in use. But if we convert these percentages to annual averages, the rate of decay is considerably smaller, amounting to about six-tenths of 1% (0.56) of the words first used between 1700 and 1729 and two one-hundredths of 1% (0.0249) of the words in use during that 30-year period.

Obsolete whole words are categorized by parts of speech in Appendix D. Although double counting is unavoidable in these data because the same word can serve more than one such function, it is clear that, just as they are the chief source of additions to the language, nouns are the words most likely to have become obsolescent.

Adjectives and verbs rank next in importance, each accounting for about 20% of the words that have disappeared from the language over time. Taken together, nouns, adjectives, and verbs make up nearly 97% of the obsolete words identified in the SOED. Appendix E compares the gross and net rates of growth of the English language from 1150 to 1985 CE. Converting these growth rates to annual averages, (27) it is apparent that, even during the periods when the most words became obsolete, the difference between the two series is quite small. Hence. although the number of obsolete words in the SOED is only about one-twentieth of the number contained in its big brother, if the distributions of these words follow similar patterns over time in the two dictionaries, the extent to which the data drawn from the OED tend to overestimate the growth of the English language is likewise quite small (compare Table 3 and Appendix E). This conclusion is reinforced by the fact, noted previously, that language growth and decay are mostly matters of changes not in the numbers of words but rather in the meanings of words. (28)

Hence, our inability to identify the obsolete words contained in the OED's larger data set seems to be of minor import. English words evidently are quite durable in comparison to other network goods, such as computer software.

3. Economics and the Growth of the English Language

Words are instruments that people are free to adapt to any use, provided they make clear their intentions. (Claude Levi-Strauss, quoted in Brandel [19871 1993. p. 3)

In what follows, we advance and test a simple argument with respect to the growth of the English language. Our purpose is to show that neology is a function of central features of economic life in the United Kingdom. Language growth, in other words, is hypothesized to be at least partly influenced by environmental factors. Thus, while language acquisition may be an innate, "hardwired" characteristic of human beings (Pinker 1999), language is also affected at the margin by constraints imposed on it by economic variables. As it were, the "software" of language changes over time in some measure as a response to changes in the nature and scope of human interactions. Indeed, if language is an evolved Hayekian order, the study of the effects of economic variables on its growth can provide useful insights into how this evolutionary process works in practice. To be sure, there are deeper linguistic forces at work, but these fall outside the more limited objectives of this economic analysis. Our goal is simply to establish that economics matters at the margin in the development of language, leaving to others the study of complementary explanations. (29)

Population, wealth, government, and trade are the key determinants of word growth in a stylized economic model of the development of the English language. We gathered annual observations on these variables for the United Kingdom for the period running from 1830 to 1969. While a richer empirical model of language growth would include measures of educational attainment, literacy rates, technological change, patterns of international migration, the global distribution of English-language publications, and many other variables, our purpose here is simply to explore in a preliminary way the question of whether there are systematic patterns in the data.

Although observations on some of our explanatory variables are available up to 1985, the last year for which the OED contains a dated quotation, consistency in the data sources and in variable definitions dictated the use of the more limited 140-year time frame ending in 1969. Our purposeful neglect of data from Australia, Canada, the United States, and other English-speaking nations has no impact on our conclusions with respect to the determinants of language growth. A more inclusive data set would be warranted if our empirical models instead focused on the stock of English words in use at a point in time. Although dictionaries of the English language certainly are available for other countries, none of these is as comprehensive--or, what is more important, as searchable--as the OED, which, after all, is a repository of British English.

We collected observations on the following explanatory variables: gross national product (GNP), 1830-1988 (in 1900 prices); total population, 1541-1980 (POP); imports (IMPORTS) to the United Kingdom and gross exports (EXPORTS) from it (both for the period 1830-1969); and government revenues (REVENUES) and expenditures (EXPENDITURES) (for the period 1830-1970). All sterling values am expressed in 1900 prices. (30) Data sources are shown in Table 5. Table 6 reports descriptive statistics.

Empirical study of the growth of the English language is hampered somewhat by the difficulty of assembling internally consistent time series of economic and demographic data that go back much before the mid-19th century. A perhaps more important constraint is that while the world economy boomed during the 140 years for which we have observations on candidate explanatory variables, the English language has grown much more modestly. The 67,457 new words added to the OED since 1830 have expanded the language by about 29.2%. Over the same period (1830-1969), the United Kingdom's population grew by 253%. Because we are able to model empirically only the tail of a very long time series, caution must be exercised in interpreting our results.

When we attempt to explain the number of words added annually to the English language, the appropriate econometric model is the count model. Count models are employed when the dependent variable takes on integer values, denoting the number of events that occur in a particular time interval. As Cameron and Trivedi (1998) point out, in such cases linear regression models often fail to account for the restricted support for the explained variable, which is always a positive integer. (31) On the other hand, when we attempt to explain the growth rate in words, using a log-linear regression specification, ordinary least squares estimating techniques can be applied.

The event modeled in our analysis is the number of words added annually to the English language as reported by the OED. Our central hypothesis is that an increase in economic activity, requiring as it does richer and more complex patterns of communication. (32) will lead to and facilitate an increase in the language base. To begin with, variables measuring the level of international trade should impact the number of words added to a language. Other things being the same, an increase in exports to other countries is hypothesized to increase the incentives of traders to incorporate additional words into the English language. Exporters arguably are compelled to familiarize themselves with the language, currency, institutions, and customs of the countries with which they trade and hence are more likely to add new words to their original language base. Such "loan-words"--words borrowed from foreign tongues--eventually are legitimated by usage. For the same reasons, imports are postulated to have the opposite effect: foreign traders selling in English-speaking countries tend to borrow words from that language in order to facilitate their transactions. Our a priori expectations are therefore that imports are negatively related, while exports are positively related, to new English words.

The number of people using a particular language presumably is very important to its growth. On the one hand, one might expect more new words to be coined the larger is the population of a language's speakers. Such a positive relationship between words and population undoubtedly holds early in a language's development. On the other hand, an argument based on the operation of network effects (positive externalities in consumption) points in the opposite direction: the more people there are who use a specific word, the more likely it is that others will know the meaning of that word and use it themselves. Network effects tend to enhance the values of existing words, thereby reducing the incentives to coin new ones. The theory of increasing returns suggests that as the stock of words increases, a point eventually will be reached at which additions to a language will be negatively correlated with population growth. The actual relationship between words and population is thus an empirical question to which the data will speak. The fact that population grew much faster than words during the time period we analyze (1830-1969) does, however, foreshadow an inverse relationship.

A similar network effect is postulated to influence the empirical relationship between the monetary value of domestic economic transactions--GNP or national wealth--and language growth. Subject to increasing returns, a given stock of words can support a larger number of market-based exchanges over time. A negative impact of increases in GNP on the number of words added to the English language is consequently predicted.

Expansions in the size and scope of government (as measured either by expenditures or by revenues) are expected to put increasing pressure on market participants to standardize their use of language and to normalize the process of contracting. More government implies more centralization and standardization. (33) Hence, it is expected that more government activity will slow down the process of language growth, producing an inverse relationship between words and the size of the public sector.

Lagged values of the dependent variables are included in the empirical specifications for two reasons. First, the number of words added in one year is likely to be highly correlated with the number of words added in the previous year. Second, this approach also allows us to account for the serial correlation in the error term as indicated by a Breusch-Godfrey Lagrange multiplier test (Greene 2000, p. 541). (34)

In estimating the count model, we apply the test for overdispersion proposed by Cameron and Trivedi (1990) for testing the appropriateness of assuming a Poisson distribution. (35) That test strongly indicates the presence of "overdispersion" in the residuals. (36) We therefore report our results from the negative binomial quasi-maximum likelihood estimator, the alternative to the Poisson suggested by Cameron and Trivedi (1986). The same authors point out that "at a different level of abstraction, an event may be thought of as the realization of a point process governed by some specific rate of occurrence of the event" (Cameron and Trivedi 1998, p. 2). In our count model specification, this would translate into an expected increase in the number of words added given a certain level of the explanatory variables. By way of contrast, the log-linear model provides coefficients that measure the constant proportional or relative changes in words for given absolute changes in the values of the regressors.

The estimates of our two model specifications are reported in Table 7 (count model results) and Table 8 (log-linear model results). As can be seen, the estimated coefficients on population and government size (measured either as total revenues or as total expenditures) are consistently negative and significant when they are included in the model. (37) The same is true for imports. Exports, on the other hand, almost always exhibit a significant positive impact on the number of words added. The coefficient on GNP is significantly negative both when measured in total and when defined in per capita terms.

The results provide evidence that words exhibit strong network effects with respect to both wealth and population--at least for the period analyzed here. Between 1830 and 1969, the number of additional words clearly is declining with increases in wealth and population. (38) Higher and higher levels of general economic activity can be supported by relatively fewer additions to the language base. (39) Over time, fewer new words are likewise required to facilitate communication among larger populations. As noted earlier, both of these negative relationships can be explained by the large stock of words already in use in 1830, the beginning of our data set. Positive relationships cannot be ruled out for earlier periods. As expected, our results for government activity suggest a standardizing effect of growing bureaucratic involvement in the economy. Other things being equal, the larger is the size of British government, the smaller is the number of new words added to the English language. (40)

The results for foreign trade comport with our a priori expectations. There is a strong indication that greater need to adjust to other countries' customs and business practices (as proxied by EXPORTS) leads to an increase in the language base as traders become more diligent in acquiring the communication skills necessary to engage in the business of shipping goods to foreign countries. Less such effort, on the other hand, is required when buying from foreigners since the burden will be on them to adhere to the local customs of trade and to observe the intricacies of the language of the receiving country. "Common culture and common language facilitate trade between individuals" (Lazear 1999, p. $95); (41) it also seems to be the case that trade impacts language growth. (42)

It is worth noting that the overall explanatory power of our models is quite high. The count and log-linear regression specifications both explain nearly 90 percent of the variation of the number of words added annually to the English language (or the growth rate in additional words) over the 18301969 period.

Table 9 displays the marginal effects of the independent variables calculated from the count model using the method suggested by Cameron and Trivedi (1998, p. 80). (43) The "Average" columns list the average of the individual responses, while, for purposes of comparison, the "Ordinary Least Squares" columns present the coefficients obtained from an OLS specification of the same model.

The entries in each column show the estimated ceteris paribus impacts of a one-unit change in the respective independent variables on the number of new words added to the English language. Over the 140 years of our data, for example, a 1 million [pounds sterling] increase in exports (in 1900 prices) caused the language base to expand by between, in model 2, two-tenths (0.21) and, in model 5, four-tenths (0.44) of a word. Stated in the reverse, one new word was added for every 2.3 [pounds sterling]-4.6 million [pounds sterling] increase in real exports. The estimates also suggest that, other things being the same, a 1 million [pounds sterling] increase in imports, again in 1900 prices, reduced language growth by between--0.17 (model 3) and --3.59 (model 1) words (i.e., one fewer word was added for every 1.7 [pounds sterling]-5.8 million [pounds sterling increase in real imports). A 160,000 increase in population or a 10 million [pounds sterling] increase in GNP likewise caused one fewer word to be added to the language base. The estimated marginal effects are generally similar across the count and OLS models, although there is a large difference in the coefficients on EXPORTS in specification 6. This variable, however, is not significant in that specification of the count model.

Table 8 presented estimates of the impact of a given absolute change in the explanatory variables on the annual growth rate of words. In the case of exports, a cautious interpretation of the results supplies evidence that, for the 140-year period covered by our data, the increase in the level of exports (from 39.8 million [pounds sterling] to 1,124.9 million [pounds sterling] in 1900 prices) contributed between 0.2% and 0.6% (135-405 words) per year to the growth of language. For imports, which increased from 50.7 million [pounds sterling] to 1,274.4 million [pounds sterling] over the same 140-year period, the results point to a negative impact on the yearly growth of language on the order of 0.44).8% (270-540 words).

Endogenous versus Exogenous Factors in Language Growth

It might be argued that the development of a language is almost exclusively an endogenous process, influenced only by earlier additions to the language base and population dynamics. If this is the case, the number of words added to a language in a given year can be characterized completely by lagged values of words and population modeled as a polynomial of arbitrarily high order. The critical question then becomes, once this polynomial regression has been estimated, is there anything left over to be explained by exogenous variables, economic or otherwise?

In order to explore this possibility, we modeled WORDS as an ARIMA process using the Box-Jenkins methodology. After employing a Dickey-Fuller test to ensure that the series is stationary, we identified population squared, the square of words lagged one period, and words lagged two periods as significant determinants of the number of new words added in the current period. We then included our economic variables in this time-series regression and found them to be significant, as earlier discussed. The same conclusions are reached even when one includes additional but statistically insignificant higher-order terms of words lagged one period and population. Hence, while language growth is apparently influenced by both endogenous and exogenous factors, the exogenous economic variables matter.

Causality

The regression results reported in Tables 7 and 8 assume that English language growth is caused by changes in the economy. Reverse causality is also possible: language growth may itself contribute to economic well-being. Indeed, as Easterly and Levine (1997) and Collier and Gunning (1999), among others, have argued, the extent of "ethnolinguistic diversity" can have a sizeable negative impact on economic growth rates. Greater linguistic diversity, in other words, hampers language's functioning as an exchange-facilitating institution. Did the growth of the English-language base in turn spur economic development over the period running from 1830 to 1969?

We investigated this possibility by applying pairwise Granger causality tests to our data, both over the 1830-1969 subsample and over larger samples for which data were available. Such tests strongly support the specifications of our models: the null hypothesis that WORDS is not Granger caused by the explanatory variables is strongly rejected, at two lags in most cases and sometimes at up to five lags. Conversely, we find little evidence of reverse causality. In the cases of exports, imports, government revenues, and GNP, we are consistently unable to reject the null that they do not Granger cause WORDS. (44) Hence, causality in our data seems to run from economics to language growth but not in the opposite direction.

Language "Efficiency"

The foregoing empirical results imply, perhaps falsely and surely unintentionally, that English-language growth is a good thing. While new words may provide benefits in the form of expressive clarity, "it is costly to learn additional vocabulary, and is socially wasteful to do so if the words are rarely, if ever, used" (Lazear 1999, p. S123). It is probably for this reason that, as discussed earlier, existing words are adapted to new uses more frequently than new ones are coined. Such adaptations can also sow confusion, of course, which is why context is all-important: "The word 'differential' means one thing to an auto mechanic and something quite different to an economist" (Lazear 1999, p. S123).

More to the point, English is widely perceived to be less "compact" than other languages, especially Latin and Greek, which are elegant in their simplicity but which, because of their fixed rules, failed to adapt. (45) Compared to Spanish, English "has many irregularities and is less consistent phonetically" (Lazear 1999, p. S122). It is a tongue "of subtlety, nuance and complexity." Consider "set,"

 an apparently simple word that takes on different meanings in a
 sporting, cooking, social or mathematical context--and that is
 before any little words are combined with it. Then, as a verb, it
 becomes "set aside," "set up," "set down," "set in," "set on," "set
 about," "set against," and so on, terms that "leave even native
 speakers bewildered about [its] core meaning." (46)

One possible explanation here is that languages are more compact (have fewer options for expressing the same idea) in economies where prices are posted than in economies where prices are subject to negotiation between buyer and seller, necessitating time-consuming "haggling and bargaining" and more nuanced talk. Casual empiricism suggests that the reverse may be true, but the hypothesis that some languages are more efficient than others merits exploration.

Although a systematic test of language efficiency is beyond the scope of the present article, we offer two suggestive observations. If "efficiency" means that the same ideas can be expressed in fewer words, then the lengths of various translations of the Holy Bible supply an indicator of the extent to which languages exhibit that property. We therefore gathered information on the total number of words required to produce the Book of Genesis in English, German, French, and Spanish versions. The data are shown in Table 10.

While we have not controlled for grammatical influences on overall scriptural length, English does seem to be less efficient by this measure. It requires 1522 more words to tell the story of Genesis in English than it does to tell it in the most "compact" of the four languages, German. a difference of about 4.3%. On the other hand, a segment from National Public Radio's Morning Edition. broadcast Monday, January 7, 2002, suggests that the opposite is true. The writers who translate dialogue in English-language films for German audiences must, in order to synchronize the dubbed voices with the original, limit themselves to the same number of syllables spoken on the screen. According to dialogue director Jurgen Neuss, "because English is every time very more quicker [sic] to speak as German," the translators frequently are required to change or to omit at least one word.

When it comes to evaluating languages, efficiency has many dimensions. As such, based on the admittedly limited evidence presented here, there is plainly more than one route to fitness in the Darwinian competition among humankind's tongues.

4. Concluding Remarks

In this article we have studied the growth of the English language by studying the OED. We recognize that "words in the dictionary" is not the same thing as "words in use." A dictionary is more than a mere catalog, however. Among other things, it establishes orthographical and definitional standards, which though by no means impervious to change, may be quite persistent. So well did Samuel Johnson's dictionary succeed in this regard "that, for over a century, the work was without serious rival" (Bate [1975] 1988, p. 251).

This standard-setting function, so valuable in exploiting the increasing returns to a "network" good, such as language, was recognized by Lord Chesterfield, to which Dr. Johnson had, in hopeful anticipation of patronage, addressed his dictionary's prospectus. Commending the dictionary to the reading public, on the eve of its appearance in print, and thereby attempting to assuage Johnson's anger over his years of willful neglect, Chesterfield observed that

 it must be owned that our language is, at present, in a state of
 anarchy and hitherto, perhaps, it may not have been the worse for
 it. During our free and open trade, many words and expressions have
 been imported, adopted, and naturalized from other languages, which
 have greatly enriched our own. Let it still preserve what real
 strength and beauty it may have borrowed from others; but let it
 not, like the Tarpeian maid, [147] be overwhelmed and crushed by
 unnecessary ornaments. The time for discrimination seems to be now
 come. Toleration, adoption, and naturalization have run their
 lengths. Good order and authority are now necessary. But where shall
 we find them, and, at, the same time, the obedience due to them? We
 must have recourse to the old Roman expedient in times of confusion,
 and chuse a dictator. Upon this principle, I give my vote to Mr.
 Johnson. (Boswell [1793] 1965, p. 73)

Fortunately for our purposes, while Dr, Johnson and his successors labored to bring order to the English language, no institution akin to the Academie Francaise ever emerged in the English-speaking world. (48) The Academie, "established primarily to 'purify'" French (Bate [1975] 1988, p. 240) by preventing its contamination by foreign words, arguably stifled its development. One would therefore not expect the (official) French language to have grown as quickly over time as English or to have been as adaptable to new circumstances of time and place. (49) Put differently, the OED supplies more useful insights into the evolution of language as a spontaneous order than its counterparts in France, Italy, and other nations where lexicography has been bureaucratized. That hypothesis is testable.

There are obviously many other opportunities for additional work in this area. Extending the analysis to include other English-speaking nations is one of them. We leave these issues for the future, being satisfied for the moment with illustrating yet another margin on which language is a worthy subject of economic analysis.

Appendices

Appendices A through E provide details supporting some of the conclusions drawn in section 2. Appendix A categorizes words added to the English language over time by parts of speech. Information on obsolescent words gleaned from the Shorter Oxford English Dictionary is presented in Appendices B to D. Finally, Appendix E reports both gross and net average annual growth rates in the English language from 1150 to 1985 CE.

Appendix A
Words Added to the English Language, Categorized by Parts of Speech,
by Century (a)

 Verbal
Century Phrases Nouns Nouns Pronouns

1st to 9 3848 184 66
 12th (0.1) (50.1) (2.4) (0.9)
13th 10 2481 175 19
 (0.2) (49.5) (3.5) (0.4)
14th 45 8074 1020 28
 (0.3) (48.8) (6.2) (0.2)
15th 32 7827 703 6
 (0.2) (48.6) (4.4) (0.04)
16th 62 16,758 1191 9
 (0.2) (46.8) (3.3) (0.03)
17th 71 21,495 945 131
 (0.2) (46.0) (2.0) (0.3)
18th 39 11,789 352 26
 (0.2) (54.1) (1.6) (0.1)
19th 149 34,952 704 7
 (0.2) (57.7) (1.2) (0.1)
20th 57 14,408 223 2
 (0.3) (70.3) (1.1) (0.1)
Totals 474 121,632 5497 294
 (0.2) (52.7) (2.4) (0.1)

 Participial
Century Adjectives Adjectives Verbs Participles

1st to 995 154 1805 24
 12th (12.9) (2.0) (23.5) (0.3)
13th 592 177 1099 28
 (11.8) (3.5) (21.9) (0.6)
14th 2153 979 2980 81
 (13.0) (5.9) (18.0) (0.5)
15th 2438 1098 2600 117
 (15.1) (6.8) (16.1) (0.7)
16th 6424 3282 5510 141
 (17.9) (9.2) (15.4) (0.4)
17th 11,151 3630 6704 96
 (23.8) (7.8) (14.3) (0.2)
18th 4840 1854 1918 24
 (22.2) (8.5) (8.8) (0.1)
19th 15,884 2189 4326 66
 (26.2) (3.6) (7.1) (0.1)
20th 3491 284 1458 9
 (17.0) (1.4) (7.1) (0.04)
Totals 47,968 13,647 28,400 586
 (20.8) (5.9) (12.3) (0.3)

Century Adverbs Prepositions Conjunctions Particles

1st to 505 46 13 1
 12th (6.6) (0.6) (0.2) (0.01)
13th 314 29 7 1
 (6.3) (0.6) (0.1) (0.02)
14th 860 35 11 0
 (5.2) (0.2) (0.1) (0.0)
15th 664 28 4 0
 (4.1) (0.2) (0.02) (0.0)
16th 1406 17 8 0
 (3.9) (0.05) (0.02) (0.0)
17th 1603 9 2 0
 (3.4) (0.02) (0.004) (0.0)
18th 491 13 0 0
 (2.3) (0.1) (0.0) (0.0)
19th 1484 19 2 0
 (2.5) (0.03) (0.003) (0.0)
20th 348 2 0 0
 (1.7) (0.01) (0.0) (0.0)
Totals 7675 198 47 2
 (3.3) (0.1) (0.02) (0.001)

Century Interjections Prefixes Suffixes

1st to 13 5 0
 12th (0.2) (0.1) (0.0)
13th 16 11 0
 (0.3) (0.2) (0.0)
14th 38 6 1
 (0.2) (0.04) (0.01)
15th 35 7 4
 (0.2) (0.04) (0.02)
16th 97 18 2
 (0.3) (0.05) (0.01)
17th 102 15 7
 (0.2) (0.03) (0.01)
18th 70 1 6
 (0.3) (0.005) (0.03)
19th 152 24 24
 (0.3) (0.04) (0.04)
20th 102 14 24
 (0.5) (0.1) (0.1)
Totals 625 101 68
 (0.3) (0.04) (0.03)

(a) Percentages of century (row) totals shown in parentheses.
Rounding may prevent totals from summing to 100.

Appendix B
New Words, Obsolete Words, and Words in Use, by Period

 First Use
Period Whole Words Any Sense

-1149
 (Old English) 4308 3836
1150-1349
 (Middle English) 6903 8718
1350-1469
 (Late Middle English) 11,755 16,666
1470-1499 1798 4052
1500-1529 1624 3906
1530-1569 4880 9877
1570-1599 6323 13,528
1600-1629 5916 12,720
1630-1669 5498 11,861
1670-1699 3042 7071
1700-1729 2378 6065
1730-1769 2724 6594
1770-1799 3520 7632
1800-1829 5181 10,940
1830-1869 10,011 19,409
1870-1899 7801 15,594
1900-1929 5823 11,455
1930-1969 6770 13,209
1970-1985 954 2401

Totals 97,209 96,739

 Last Use
Period Whole Words Any Sense

-1149
 (Old English) 0 139
1150-1349
 (Middle English) 0 0
1350-1469
 (Late Middle English) 23 1539
1470-1499 8 524
1500-1529 4 435
1530-1569 29 1117
1570-1599 173 2041
1600-1629 168 2951
1630-1669 151 3827
1670-1699 79 2793
1700-1729 402 2833
1730-1769 403 2485
1770-1799 279 1836
1800-1829 323 1832
1830-1869 301 1416
1870-1899 134 555
1900-1929 21 100
1930-1969 2 14
1970-1985 0 0

Totals 2500 17,643

 In Use
Period Whole Words Any Sense

-1149
 (Old English) 4308 3836
1150-1349
 (Middle English) 11,211 10,714
1350-1469
 (Late Middle English) 22,966 22,484
1470-1499 24,741 23,826
1500-1529 26,357 25,366
1530-1569 31,233 30,283
1570-1599 37,527 36,542
1600-1629 43,270 42,185
1630-1669 48,600 47,340
1670-1699 51,491 49,883
1700-1729 53,790 52,035
1730-1769 56,112 54,246
1770-1799 59,229 57,329
1800-1829 64,131 62,372
1830-1869 73,819 72,333
1870-1899 81,319 79,944
1900-1929 87,008 85,676
1930-1969 93,757 92,529
1970-1985 94,709 93,498

Totals

Appendix C
Obsolete Words as Percentages of New Words and Words in Use

 First Use
Period Whole Words Any Sense

-1149
 (Old English) 0 3.62
1150-1349
 (Middle English) 0 0
1350-1469
 (Late Middle English) 0.20 9.23
1470-1499 0.45 12.93
1500-1529 0.25 11.14
1530-1569 0.59 11.31
1570-1599 2.74 15.09
1600-1629 2.84 23.20
1630-1669 2.75 32.27
1670-1699 2.60 39.50
1700-1729 16.90 46.71
1730-1769 14.79 37.69
1770-1799 7.93 23.75
1800-1829 6.23 16.75
1830-1869 3.01 7.30
1870-1899 1.72 3.56
1900-1929 0.36 0.87
1930-1969 0.03 0.11
1970-1985 0 0

Overall averages 2.57 18.24

 In Use
Period Whole Words Any Sense

-1149
 (Old English) 0 3.62
1150-1349
 (Middle English) 0 0
1350-1469
 (Late Middle English) 0.10 6.84
1470-1499 0.03 2.20
1500-1529 0.02 1.71
1530-1569 0.09 3.69
1570-1599 0.46 5.59
1600-1629 0.39 7.00
1630-1669 0.31 8.08
1670-1699 0.15 5.60
1700-1729 0.75 5.44
1730-1769 0.72 4.58
1770-1799 0.47 3.20
1800-1829 0.50 2.94
1830-1869 0.41 1.96
1870-1899 0.17 0.69
1900-1929 0.02 0.12
1930-1969 0.002 0.02
1970-1985 0 0

Overall averages 2.64 18.87

Appendix D
Obsolete Whole Words, Categorized by Parts of Speech, 1350-1985 CE (a)

Period Nouns Pronouns Adjectives

1350-1469
 (Late Middle
 English) 11 (47.8) 0 (0.0) 5 (21.2)
1470-1499 5 (55.5) 0 (0.0) 1 (11.1)
1500-1529 3 (75.0) 0 (0.0) 1 (25.0)
1530-1569 11 (37.9) 0 (0.0) 4 (13.8)
1570-1599 83 (47.4) 0 (0.0) 22 (12.6)
1600-1629 64 (36.8) 0 (0.0) 47 (27.0)
1630-1669 63 (40.1) 2 (1.3) 28 (17.8)
1670-1699 35 (43.2) 0 (0.0) 18 (22.2)
1700-1729 236 (57.7) 0 (0.0) 103 (25.2)
1730-1769 230 (56.5) 0 (0.0) 97 (23.8)
1770-1799 145 (51.2) 0 (0.0) 78 (27.6)
1800-1829 201 (60.0) 0 (0.0) 73 (21.8)
1830-1869 193 (62.1) 0 (0.0) 61 (19.6)
1870-1899 94 (68.1) 0 (0.0) 25 (18.1)
1900-1929 16 (76.2) 0 (0.0) 3 (14.3)
1930-1969 2 (100.0) 0 (0.0) 0 (0.0)
1970-1985 0 (--) 0 (--) 0 (--)

Totals 1392 (54.4) 2 (0.1) 566 (22.1)

Period Verbs Adverbs Prepositions

1350-1469
 (Late Middle
 English) 3 (13.0) 3 (13.0) 1 (4.3)
1470-1499 0 (0.0) 2 (22.2) 0 (0.0)
1500-1529 0 (0.0) 0 (0.0) 0 (0.0)
1530-1569 13 (44.8) 1 (3.4) 0 (0.0)
1570-1599 62 (35.4) 5 (2.9) 0 (0.0)
1600-1629 55 (31.6) 4 (2.3) 0 (0.0)
1630-1669 52 (33.1) 6 (3.8) 1 (0.6)
1670-1699 24 (29.6) 2 (2.5) 1 (1.2)
1700-1729 63 (15.4) 6 (1.5) 0 (0.0)
1730-1769 70 (17.2) 6 (1.5) 0 (0.0)
1770-1799 58 (20.5) 1 (0.4) 0 (0.0)
1800-1829 55 (16.4) 2 (0.6) 0 (0.0)
1830-1869 52 (16.7) 0 (0.0) 1 (0.3)
1870-1899 16 (11.6) 3 (2.2) 0 (0.0)
1900-1929 0 (0.0) 2 (9.5) 0 (0.0)
1930-1969 0 (0.0) 0 (0.0) 0 (0.0)
1970-1985 0 (--) 0 (--) 0 (--)

Totals 523 (20.4) 43 (1.6) 4 (0.2)

Period Conjunctions Interjections Prefixes

1350-1469
 (Late Middle
 English) 0 (0.0) 0 (0.0) 0 (0.0)
1470-1499 1 (1.1) 0 (0.0) 0 (0.0)
1500-1529 0 (0.0) 0 (0.0) 0 (0.0)
1530-1569 0 (0.0) 0 (0.0) 0 (0.0)
1570-1599 2 (1.1) 1 (0.6) 0 (0.0)
1600-1629 1 (0.6) 3 (1.7) 0 (0.0)
1630-1669 1 (0.6) 3 (1.9) 1 (0.6)
1670-1699 0 (0.0) 1 (1.2) 0 (0.0)
1700-1729 0 (0.0) 1 (0.2) 0 (0.0)
1730-1769 0 (0.0) 4 (1.0) 0 (0.0)
1770-1799 0 (0.0) 1 (0.4) 0 (0.0)
1800-1829 1 (0.3) 3 (0.9) 0 (0.0)
1830-1869 1 (0.3) 3 (1.0) 0 (0.0)
1870-1899 0 (0.0) 0 (0.0) 0 (0.0)
1900-1929 0 (0.0) 0 (0.0) 0 (0.0)
1930-1969 0 (0.0) 0 (0.0) 0 (0.0)
1970-1985 0 (--) 0 (--) 0 (--)

Totals 7 (0.3) 20 (0.8) 1 (0.04)

Period Suffixes

1350-1469
 (Late Middle
 English) 0 (0.0)
1470-1499 0 (0.0)
1500-1529 0 (0.0)
1530-1569 0 (0.0)
1570-1599 0 (0.0)
1600-1629 0 (0.0)
1630-1669 0 (0.0)
1670-1699 0 (0.0)
1700-1729 0 (0.0)
1730-1769 0 (0.0)
1770-1799 0 (0.0)
1800-1829 0 (0.0)
1830-1869 0 (0.0)
1870-1899 0 (0.0)
1900-1929 0 (0.0)
1930-1969 0 (0.0)
1970-1985 0 (--)

Totals 0 (0.0)

(a) Percentages of century (row) totals--adjusted for 58 words
counting as more than one part of speech--shown in parentheses.
Rounding may prevent totals from summing to 100.

Apenndix E
Gross and Net Growth Rates in the English Language, Annual
Averages. 1150-1985 CE (Percentages)

 First Use
Period Whole Words Any Sense

1150-1349
 (Late Middle English) 0.80 1.14
1350-1469 0.87 1.11
1470-1499 0.26 0.46
1500-1529 0.22 0.39
1530-1569 0.46 0.66
1570-1599 0.67 0.96
1600-1629 0.52 0.70
1630-1669 0.32 0.40
1670-1699 0.21 0.28
1700-1729 0.15 0.22
1730-1769 0.13 0.17
1770-1799 0.21 0.25
1800-1829 0.28 0.32
1830-1869 0.38 0.39
1870-1899 0.34 0.36
1900-1929 0.23 0.24
1930-1969 0.19 0.19
1970-1985 0.07 0.09
1150-1985 0.92 1.65

 In Use
Period Whole Words Any Sense

1150-1349
 (Late Middle English) 0.80 0.90
1350-1469 0.87 0.87
1470-1499 0.25 0.20
1500-1529 0.22 0.22
1530-1569 0.46 0.48
1570-1599 0.67 0.69
1600-1629 0.51 0.51
1630-1669 0.31 0.31
1670-1699 0.20 0.18
1700-1729 0.15 0.14
1730-1769 0.11 0.11
1770-1799 0.19 0.19

1800-1829 0.28 0.29
1830-1869 0.38 0.40
1870-1899 0.34 0.35
1900-1929 0.23 0.24
1930-1969 0.19 0.20
1970-1985 0.07 0.07
1150-1985 0.89 0.92

Table 1. Percentage of Authorities Quoted in the Oxford English
Dictionary, by Century, 252-1985 CE

Period Percentage Cumulative Percentage

1st to 12th centuries 1.0 1.0
13th century 1.0 2.0
14th century 3.5 5.5
15th century 4.5 10.0
16th century 10.0 20.0
17th century 16.0 36.0
18th century 11.0 47.0
19th century 31.0 78.0
20th century 20.0 98.0
Undated (a) 0.5 98.5

Source: "Facts about Oxford English Dictionary," p. 3. Accessed 1
May 2004. Available http://oed.com/public/inside/funfacts.htm.

(a) "Undated" includes approximately 1250 quotations from Beowulf.
The remainder consists of proverbs, nursery rhymes, "made-up"
illustrations, and references to the appearance of word forms
"in modern dictionaries."

Table 2. New Words Added to the English Language, by Century,
252-1985 CE

 Cumulative Cumulative
 Period Words Words Percentage Percentage

1st to 12th centuries (a) 7681 7681 3.33 3.33
13th century 5014 12,695 2.17 5.50
14th century 16,561 29,256 7.18 12.68
15th century 16,102 45,358 6.98 19.66
16th century 35,810 81,168 15.52 35.18
17th century 46,773 127,941 20.27 55.45
18th century 21,735 149,676 9.42 64.87
19th century 60,559 210,235 26.25 91.12
20th century 20,484 230,719 8.88 100.0

(a) The following are the number of words added per century from the
1st through the 12th centuries: 3rd century (1), 4th century (2),
5th century (2), 6th century (7), 7th century (30), 8th century
(556), 9th century (1943), 10th century (1307), 11th century (3096),
and 12th century (737).

Table 3. Average Annual Growth Rates of the English Language by Century
(Percentages)

 Standard
Period Mean Median Deviation Minimum Maximum

1st to 12th centuries 1.43 0 16.98 0 476.19
13th century 0.52 0.03 1.85 0 11.63
14th century 0.89 0.06 3.36 0.01 31.99
15th century 0.45 0.11 1.21 0.01 10.57
16th century 0.58 0.50 0.41 0.06 2.01
17th century 0.46 0.38 0.37 0.11 2.96
18th century 0.16 0.13 0.10 0.06 0.73
19th century 0.34 0.35 0.10 0.17 0.60
20th century 0.11 0.11 0.06 0.0009 0.23
Totals 0.98 0.04 12.60 0 476.19

Table 4. Average Annual Number of Words Added to the English Language,
by Century

 Standard
Period Mean Median Deviation Minimum Maximum

1st to 12th centuries 8.10 0 97.23 0 2776
13th century 50.14 3 166.95 0 1064
14th century 165.61 13 473.26 3 4061
15th century 161.02 41.5 381.43 5 3091
16th century 358.10 308 269.11 28 1425
17th century 467.73 415.5 330.60 135 2638
18th century 217.35 174.5 138.93 76 971
19th century 605.59 593.5 189.34 260 1051
20th century 238.19 245.5 120.26 2 476
Totals 133.06 2 272.72 0 4061

Table 5. Variable Definitions and Data Sources

Variable Data Sources (a)

WORDS CD-ROM version of the second edition of the Oxford
 English Dictionary
POP Total UK population in thousands of persons (Mitchell
 1988, pp. 7-14)
GNP UK gross national product in millions of pounds sterling
 at 1900 prices (Mitchell 1978, pp. 407-22)
IMPORTS (b) UK imports in millions of pounds sterling at 1900 prices
 (Mitchell 1978, pp. 298-310)
EXPORTS (b) UK exports in millions of pounds sterling at 1900
 prices; EXPORT is the sum of exports and reexports
 (Mitchell 1978, pp. 298-310)
REVENUES UK government revenues in millions of pounds sterling at
 1900 prices (Mitchell 1978, pp. 370-83)
EXPENDITURES UK government expenditures in millions of pounds
 sterling at 1900 prices (Mitchell 1978, pp. 370-83)

(a) Mitchell in turn relies to some extent on data from
Feinstein (1972).

(b) Numbers for exports and imports are for the United Kingdom
(including England, Ireland, Scotland, and Wales). Southern Ireland
was treated as foreign from April 1, 1923.

Table 6. Descriptive Statistics

 WORDS POP GNP

Mean 133.09 14,828 2,795.02 [pounds sterling]
Median 2 6124.5 2,220.92
Maximum 4061 49,603 9374.43
Minimum 0 2774 397
Standard
 deviation 272.73 14,453.91 2256.27
Skewness 4.79 1.17 1.16
Kurtosis 46.80 2.88 3.42
Jarque-Bern 145,206 100.65 36.82
Probability 0 0 0
Sum 230,773 6,524,318 4,444,08.7
Sum of
 squared
 deviations 1.29E + 08 9.17E + 10 8.04E + 08
Observations 1734 440 159

 IMPORTS EXPORTS

Mean 471.78 [pounds sterling] 343.49 [pounds sterling]
Median 441 297.15
Maximum 1274.4 1124.9
Minimum 50.7 39.8
Standard
 deviation 283.47 235.08
Skewness 0.52 0.97
Kurtosis 2.83 3.49
Jarque-Bern 6.50 23.44
Probability 0.04 8.00E - 06
Sum 66,049.70 48,088.3
Sum of
 squared
 deviations 1.11E + 08 7,681,690
Observations 140 140

 REVENUES EXPENDITURES

Mean 442.37 [pounds sterling] 505.78 [pounds sterling]
Median 137 141.7
Maximum 2330.5 2394.4
Minimum 48.3 47
Standard
 deviation 546.76 643.90
Skewness 1.48 1.42
Kurtosis 4.26 3.83
Jarque-Bern 60.64 51.55
Probability 0 0
Sum 62,373.5 71,315.
Sum of
 squared
 deviations 4.19E + 07 5.80E + 07
Observations 141 141

Table 7. Estimates of Negative Binomial Count Models (a)

 1 2 3

Constant 5.772868 *** 6.098018 *** 5.974138 ***
 (0.094053) (0.141420) (0.091817)
POP -1.32E - 05 ***
 (4.36E.06)
GNP -0.000175 ***
 (3.05E - 05)
IMPORTS -0.001227 *** -0.000714 ** -0.000642 ***
 (0.000249) (0.000299) (0.000249)
EXPORTS 0.000664 ** (0.000448) 0.000737 ***
 (0.000275) (0.000279) (0.000245)
GNP/POP

EXPENDITURES

REVENUES

WORDS(-1) 0.001340 *** 0.001167 *** 0.001064 ***
 (0.000112) (0.000122) (0.000112)
Likelihood
 ratio index
 (pseudo-[R.sup.2]) (0.855295) (0.871885) 0.894224

 4 5 6

Constant 6.088937 *** 5.880400 *** 5.852874 ***
 (0.115885) (0.085461) (0.086906)
POP
GNP
IMPORTS -0.000585 ** -0.001099 *** -0.000498 *
 (0.000281) (0.000222) (0.000263)
EXPORTS 0.000591 ** 0.000913 *** 2.89E - 05
 (0.000256) (0.000244) (0.000273)
GNP/POP -8.671900 ***
 (2.031803)
EXPENDITURES -0.000231 ***
 (4.22E - 05)
REVENUES -0.000323 ***
 (5.39E - 05)
WORDS(-1) 0.001209 *** 0.001094 *** 0.001143 ***
 (0.000110) (0.000108) (0.000108)
Likelihood
 ratio index
 (pseudo-[R.sup.2]) (0.868061) 0.898417 0.894128

(a) GLM-robust covariance estimators; dependent variable is the number
of words added to the English language, 1830-1969 CE. Standard errors
shown in parentheses; asterisks indicate significance at the
1% (***), 5% (**). and 10% (*) levels, respectively.

Table 8. Estimates of Log-Linear Models (a)

 1 2 3

Constant 1.795141 *** 2.405485 *** 2.626799 ***
 (0.408988) (0.527838) (0.498052)
POP -9.26E - 06 ***
 (2.74E - 06)
GNP -0.000112 ***
 (2.33E - 05)
IMPORTS -0.000889 *** -0.000564 *** -0.000615 ***
 (0.000173) (0.000188) (0.000187)
EXPORTS 0.000580 *** 0.000432 ** 0.000662 ***
 (0.000194) (0.000196) (0.000160)
GNP/POP

EXPENDITURES

REVENUES

Log WORDS(-1) 0.738711 *** 0.668548 *** 0.614892 ***
 (0.062235) (0.075943) (0.075916)
Adjusted [R.sup.2] 0.871238 0.875323 0.881631
F-statistic 314.5028 244.9692 259.8243

 4 5 6

Constant 2.323003 *** 2.473683 *** 2.336419 ***
 (0.494092) (0.480370) (0.474168)
POP

GNP

IMPORTS -0.000581 *** -0.000918 *** -0.000587 ***
 (0.000160) (0.000164) (0.000166)
EXPORTS 0.000554 *** 0.000754 *** 0.000278 *
 (0.000133) (0.000157) (0.000151)
GNP/POP -4.878896 ***
 (1.380347)
EXPENDITURES -0.000124 ***
 (3.55E - 05)
REVENUES -0.000179 ***
 (5.47E - 05)
Log WORDS(-1) 0.678475 *** 0.631737 *** 0.653070 ***
 (0.072150) -0.074128 (0.072740)
Adjusted [R.sup.2] 0.876022 0.879235 0.877467
F-statistic 246.5410 253.9984 249.8465

(a) Newey-West heteroscedasticity and autocorrelation consistent
standard errors and covariance estimators; dependent variable is the
natural logarithm of words added to the English language, 1830-1969
CE. Standard errors shown in parentheses; asterisks indicate
significance at the 1% (***), 5% (**), and 10% (*) levels,
respectively.

Table 9. Marginal Effects from the Negative Binomial Count Models (a)

 1 2

 Average OLS Average OLS

POP -0.00641 -0.00631
GNP
IMPORTS -0.59592 -0.42075 -0.34681 -0.18211
EXPORTS 0.32249 0.280134 0.21760 0.17621
GNP/POP
EXPENDITURES
REVENUES
WORDS(- 1) 0.6508 0.695242 0.56684 0.60554

 3 4

 Average OLS Average OLS

POP
GNP -0.10328 -0.03991
IMPORTS -0.17237 -0.28831 -0.28371 -0.30372
EXPORTS 0.30559 0.29510 0.28634 0.26364
GNP/POP -4204.87 -1563.698
EXPENDITURES
REVENUES
WORDS(- 1) 0.47749 0.62918 0.58603 0.66891

 5 6

 Average OLS Average OLS

POP
GNP
IMPORTS -0.53054 -0.39801 -0.2404 -0.31320
EXPORTS 0.44070 0.31608 0.01395 0.18601
GNP/POP
EXPENDITURES -0.1117 -0.03383
REVENUES -0.15607 -0.05150
WORDS(- 1) 0.52799 0.65528 0.55201 0.66608

(a) GLM-robust covariance estimators; dependent variable is the number
of words added to the English language, 1830-1969 CE. Marginal effects
are computed by aggregating over all individuals and calculating the
average response (see Cameron and Trivedi 1990, p. 80). The results
from simple OLS specifications are reported for comparison purposes.

Table 10. The Book of Genesis in English, German, French, and Spanish

Language Edition of the Holy Bible Total Word Count

English New International Version 36,931
German Luther Bible 35,409
French Louis Segond 36,302
Spanish Nueva Version Internacional 36,409

Source: http://www.bible.gospelcom.net. Accessed 1 May 2004.

(1) Gertrude Stein did once say, though, that "the thing that differentiates animals and man is money" (quoted in "Crisp and Even." The Economist. December 22, 2001-January, 4 2002, p. 87).

(2) Related aspects of language use may also be amenable to economic interpretation. According to McCloskey (1985, p. xviii), for example, "rhetoric is an economics of language, the study of how scarce means are allocated to the insatiable desires of people to be heard."

(3) Economists" colonization of the study of language began with Adam Smith, who offered a rational choice account of changes provoked by speakers of different tongues attempting to communicate with one another. Smith's account clearly predicts what is known to modern linguists as pidgin (Levy 1997). Marschak (1965) pioneered the modern economics literature on language, conceiving of it as a currency whose use reduces trading costs. That analogy has been criticized by Grin (1996, p. 28), who nevertheless recognizes the positive contributions of economics to the study of language. One aspect of the economics literature, summarized cogently by Lazear (1999). addresses the incentives of the speakers of one tongue to learn another (many of the relevant papers are collected in Lamberton 2002). Lazear models the benefits of language acquisition in terms of the size of the pool of trading partners to which fluency yields access. It follows that "the value of [linguistic] assimilation is larger to an individual from a small minority than to one from a large minority group" (Lazear 1999, p. S95). And, indeed, the empirical evidence (e.g., Chiswick 1978; McManus, Gould, and Welch 1983; Dustmann and Fabbri 2003) suggests that the ability to speak the majority language increases earnings. Relatedly, Choi (2002) presents a two-country model in which free trade encourages workers in the low-wage country to learn the language of the high-wage country. Absent a reversal in relative wages, the language of the high-wage country is universally adopted in the long run. This seems to be a special case of Lazear's more general model. For an alternative application of the methods of economics to the study of language, grounded in evolutionary game theory, see Rubinstein (2000).

(4) Hayek's observation echoes Samuel Johnson, who wrote in the preface to his 1755 Dictionary of the English Language that

 commerce, however necessary, however lucrative, as it depraves the
 manners, corrupts the language; they that bare frequent intercourse
 with strangers, to whom they endeavour to accommodate themselves,
 must in time learn a mingled dialect, like the jargon which serves
 the traffickers on the Mediterranean and Indian coasts. This will
 not always be confined to the exchange, the warehouse, or the port,
 but will be communicated by degrees to other ranks of the people,
 and be at last incorporated with the current speech. (Lynch 2002,
 p. 41; emphasis in original)

(5) The classic reference to language as a network good is Church and King (1993). See Shy (2001, pp. 252-9) for a summary of their analysis. Laitin (1994) also treats communication as a coordination problem whose solution is facilitated by the adoption of a common technological standard (a language). Arabic served such a function in the medieval Islamic world,

 where circumstances ... were uniquely favorable to the development
 of long-range, large-scale commerce. For the first time ever, a
 vast region of ancient civilizations, from Morocco across North
 Africa to the Middle East and as far as--and later beyond--the
 borders of China and India, was united in a single political and
 cultural system, for a while even under a single central authority.
 The Arabic language, in universal use at least as a medium of
 international and inter-regional communication, was understood from
 one end of the Islamic world to the other, and provided a subtle,
 rich and sophisticated medium of communication. (Lewis 1995, p. 172)

(7) Samuel Johnson apparently was the first lexicographer systematically to employ authoritative quotations to illustrate the meanings of words in the English language. In doing so, Johnson was following the precedent set by the six-volume Italian dictionary, first published in 1621 under the auspices of the Accademia della Crusca (Bate [1975] 1988, p. 247). Boswell ([1793] 1965, pp. 51-2) writes that Johnson, in the prospectus for his Dictionary of the English Language. credited Alexander Pope with selecting "many of the writers whose testimonies were to be produced as authorities," suggesting that Pope had contributed ideas to an earlier English dictionary project planned along these same lines but ultimately abandoned. Even earlier precedents can be found in "the great Arabic dictionaries of the Middle Ages, listing the different meanings of words and illustrating them with examples of their occurrence in classical texts" (Lewis 1995, p. 264).

(8) "Facts about Oxford English Dictionary," p. 1. Accessed 1 May 2004. Available http://oed.com/public/inside/funfacts.htm.

(9) Ibid., pp. 23.

(10) Thomas Jefferson to John Adams, August 15, 1820 (Cappon 1959, vol. 11, p. 567). As Dr. Johnson put it, "as language was at its beginning merely oral, all words of necessary or common use were spoken before they were written" (Lynch 2002, p.

(11) To supply just one of many possible examples.

 the word "civilization"--a neologism--emerged late, and
 unobtrusively, in eighteenth-century France. It was formed from
 "civilized" and "to civilize." which had long existed and were in
 general use in the sixteenth century. In about 1732, "civilization"
 was still only a term in jurisprudence: it denoted an act of justice
 or a judgement which mined a criminal trial into civil proceedings.
 Its modern meaning, "the process of becoming civilized," appeared
 later, in 1752, from the pen of the French statesman and economist
 Anne Robert Jacques Turgot, who was then preparing a universal
 history, although he did not publish it himself. The official debut
 of the word in print occurred in 1756. in a work entitled A Treatise
 on Population by Victor Riqueti, Marquis of Mirabeau, the father of
 the celebrated revolutionary Honore, Count Mirabeau.... From France,
 the word "civilization" rapidly spread through Europe.
 The word "culture" went with it. By 1772 and probably earlier, the
 word "civilization" had reached England and replaced "civility,"
 despite its long history (Braudel [1987] 1993, pp. 3-4).

(12) Of course, English, as such, did not exist at the beginning of our data set. What was spoken in 252 CE was a Saxon dialect of northern Germany. In the period prior to 1470 CE, according to the OED. the language evolved in the following sequence: Old English (before 1150), Middle English (1150-1349), and Late Middle English (1350-1469). About 4500 words in the OED are survivors from Old English, including "every one of the one hundred most common" (Bryson 1990. p. 58).

(13) Pinker's distinction between nouns and verbs may offend linguists, who prefer to talk about morphemes--indivisible units, such as "in," "come" and "-ing"--as the building blocks of language.

(14) English was the language used on 78.3% of the Interact Web pages in 1999. Next in importance were Japanese (2.5% of the Web pages). German (2.0%), Spanish (1.7%), French (1.2%), and Chinese (0.6%). All other languages collectively accounted for 13.7% of the Web pages ("Le Cyber Challenge," The Economist, March 11, 2000, p. 55). Similarly, English has for some time been the lingua franca of air traffic control (Lazear 1999, p. S100). For a general discussion of the globalization of the English language, see "A World Empire by Other Means," The Economist, December 22, 2001-January 4, 2002, pp. 65-7. and Bryson (1990. pp. 35-45). It would not be surprising to find that the global diffusion of English follows a logistic function, as have the adoption of technological and legal innovations. See Grilliches (1957) on hybrid corn and Shughart and Tollison (1985) on corporate chartering laws.

(15) "Facts about Oxford English Dictionary," pp. 2-3.

(16) Irving Fisher is quoted six times, and Thomas R. Malthus and Bernard

Mandeville are quoted twice each. Karl Marx is quoted once, and so are Vilfredo Pareto and Alfred Marshall. Paul Samuelson ranks first among living economists with four quotations, the same number attributed to David Ricardo.

(17) Dr. Johnson may in fact have gathered over twice that number of quotations. He "was forced to drop somewhat more than half of them lest "the bulk of my volumes ... fright away the student'" (Bate [1975] 1988, p. 247). Although it ultimately took Johnson thrice the time to complete his dictionary he estimated it would take at the outset (three years), comparisons with the national dictionaries of Italy and France, both of which had been the products of learned academies with many members, illustrate the magnitude of his scholarly achievement. Aided in his efforts only by "six humble assistants" (Bate [1975] 1988, p. 2431, Johnson produced the first major dictionary of the English language in less than half the 20 years it had taken the Accademia della Crusca to prepare its six-volume work. The French Academy took even longer. It "spent four years deliberating how to proceed with a dictionary that could match or excel the Italian. Then, in 1639, with its original eight members expanded to forty, it began actual work and finished it fifty-five years later (1694)" (Bate [19751 1998. p. 240). The French dictionary's protracted development cannot be explained by a painstaking search for quotations to illustrate the meanings of words: that "practice had been rejected by the French Academy with the excuse that membership in the Academy automatically indicated that one was already an 'authority,' and that other authorities did not need to be cited" (Bate [1975] 1988, p. 248). Asked by Dr. Adams how he thought he could possibly finish in three years a project on the scale of one that had taken the 40-member French Academy 40 years to bring to fruition, Johnson famously responded. "let me see; forty times forty is sixteen hundred. As three to sixteen hundred, so is the proportion of an Englishman to a Frenchman" (Boswell [1793] 1965, p. 54).

(18) "Facts about Oxford English Dictionary," p. 2. Many of the spurious words owe their existence to typographical errors or other mistakes: "One such is messuage, a legal term used to describe a house, its land, and buildings. It is thought to be simply a careless transcription of the French menage" (Bryson 1990, p. 71).

(19) Some analysis of decay rates based on data from the CD-ROM version of The New Shorter Oxford English Dictionary is reported later in this article, however.

(20) Goodness-of-fit tests do not reject the hypothesis that the series of new words added annually to the English language follows a Poisson distribution with a mean of 133 and a standard deviation of 273. The Poisson distribution describes the number of times a specific event occurs in a particular time interval and is often used by industrial engineers to model physical processes such as the arrivals of people joining a queue, of jobs at a work station, and of telephone calls at a central switch. We exploit the properties of this distribution in section 3. However. because the goodness-of-fit tests do not distinguish sharply between the Poisson and alternative distributions (e.g., normal and exponential), we do not restrict ourselves to that assumption.

(21) The nineteenth century marked the pinnacle of success of the English novel). Publishing thrived because of both a decline in the costs of paper and printing and a literate public's insatiable demand for the writings of Charles Dickens, Anthony Trollope, the Bronte sisters, William Thackeray, and George Eliot. to name a few of the leading lights of British fiction. Between 1837 and 1901, according to one estimate, approximately 50,000 novels were published in the United Kingdom (Zimmerman 2003. p. 1128).

(22) The Bard of Avon "used 17,677 words in his writings, of which at least one-tenth had never been used before" (Bryson 1990. p. 76).

(23) The categories used here are based on the OED's definitions. Linguists, by contrast, do not identify parts of speech by semantic criteria. A word is linguistically a noun, for example, if it can take a noun suffix, such as an article and plural marker, Verbs are likewise words that can take a tense marker.

(24) More formally, a participial adjective is an adjective that is participial in origin and form, where a participle is a nonfinite part of a verb used with an auxiliary verb in expressing tense and voice, such as "has gone." "had been kicked," and "will be working." Strictly speaking, "engaged" is both a participle and a past tense of "engage."

(25) One of English's distinctive characteristics is its wealth of different words having the same meaning: "English ... is the only language that has, or needs, books of synonyms like Roget's Thesaurus, 'Most speakers of other languages are not aware that such books exist'" (Bryson 1990, p. 14, quoting Laird 1953, p. 54).

(26) This is the etymology given by the OED. According to Dr. Johnson (Lynch 2002, pp. 58-9). "algebra" is "derived, by some. from Geber, the philosopher; by some, from gefr, parchment; by others, from algehista, a bone-setter; by Menage, from algiatarat, the restoration of things broken." An alternate view of the word discussed in the OED is that it has maintained its mathematical meaning since it first appeared in a book titled ibn al-jabr wa'l mukabala by al-Kwarizimi, an Arab of the ninth century. Drift in a word's meaning over time is called catachresis (Bryson 1990, p. 78).

(27) This calculation assumes a constant, linear decay rate.

(28) Another way of seeing the same thing is to calculate the average age of words, Using the series of words in use identified in the SOED, starting with the year 1150 CE and dating words from the midpoints of the intervals shown in Appendix B, yields an average age of 298 years. When the same calculation is performed for words dated by their first use (i.e., unadjusted for obsolescence), the average declines by slightly more than one year (to 296.6 years). A similar calculation for the more comprehensive OED yields an average age of 307.3 years for the 223.524 words added to the English language since 1150. As of 1985. the average age of all words contained in the OED is 331.5 years.

(29) For a sociological perspective on some of these issues, see Lieberson (2000). who studies trends in the popularities of given names.

(30) Nominal values were converted to constant 1900 prices using the implicit UK GNP price deflator calculable from Mitchell (1978, pp, 407-22).

(31) Similarly. Greene (2000, p. 880) argues that "the preponderance of zeros and the small values and clearly discrete nature of the dependent variable suggest that we could improve on least squares and the linear model with a specification that accounts for these characteristics."

(32) "The general formula for the number of two-way communications necessary to link up all the members of a group is n(n-1)/ 2, where n is the number of members of the group" (Posner 2001, p. 66).

(33) Such standardization materialized in the medieval Islamic world, where, as a result of Caliphate hegemony. Arabic. along with one or two other major languages (Persian and Turkish), united the far-flung and ethnically diverse Muslim peoples, "serving not only as the media of a narrow clerical class, like Latin in Western Europe, but as the effective means of universal communication. supplanting local languages and dialects at all but the lowest levels." The "linguistic unity of the Muslim world" had no "parallel in medieval Christendom. 'The Franks.' says Rashid al-Din, 'speak twenty-five languages, and no people understands the language of any other'" (Lewis 1995. p. 173). Dictionaries themselves help contribute to standardization, especially with respect to orthography. Pronunciation is another matter entirely, however (Bryson 1990, pp. 99-116).

(34) Including the lagged dependent variable on the right-hand side essentially models the growth of words as a dynamic-adjustment process, thereby introducing the possibility that the ceteris paribus effects differ as between the long run and the short run. It turns out, however, that because the estimated coefficients on lagged words are very small numerically, the estimated coefficients on the other explanatory variables are nearly identical over both time horizons. Moreover. estimating the same models without including lagged values of the dependent variable does in no case change the algebraic signs of the reported coefficients or lead to a reduction in the precision with which they are estimated. To the contrary, statistical significance improves in a number of instances.

(35) The Poisson model is restrictive in that it assumes that the conditional mean of the dependent variable is equal to its variance. A test for overdispersion in the residuals first estimates the Poisson model and then obtains fitted values. In the next step. the difference between the squared regression residuals and the actual values of the dependent variable are regressed on the squared estimated values. If the resulting coefficient is significantly different from zero. one rejects the Poisson restriction. In addition, if the coefficient is greater than zero, one concludes that the variance is greater than the conditional mean. which is defined as overdispersion.

(36) The coefficients in the overdispersion tests are highly significant, leading us to reject the Poisson restriction. In addition, the estimated coefficients are significantly positive, supplying another indicator of overdispersion in the residuals.

(37) A specification including both EXPENDITURES and REVENUES was also estimated. However, it is not reported here because of the not unexpectedly high partial correlation between those two variables.

(38) The log-linear model suggests that this result also holds for changes in these variables.

(39) An alternative interpretation of the negative relationship between words added to the English language and GNP is that words are an income-inferior good.

(40) The growth of government was paralleled by the growth of the commercial media, first newspapers, then radio and television. which would also tend to standardize language use. As Dr. Johnson remarked, "from ... uncertain pronunciation arise in great part the various dialects of the same country, which will always be observed to grow fewer, and less different, as books are multiplied" (Lynch 2002, p, 26). A richer model of language growth, perhaps including a time series of paper prices, might disentangle these effects.

(41) Adam Smith's "argument in Wealth of Nations is that trade and language are two aspects of the same process; humans trade because we have language, nonhumans do not trade because they do not" (Levy 1997, p. 672). As pointed out by one of the referees, the direction of causation alluded to here might be reversed by modern biopaleontologists: wade may well have given rise to language as a survival mechanism.

(42) The results reported in Tables 7 and 8 generally me robust to the inclusion of YEAR and YEAR (2) as explanatory variables, which allow for diminishing returns to language growth. We also tested the hypotheses that the annual rate of inflation, the percentage of the population in the British armed forces (as a proxy for the impacts of war and of British colonialism), and immigration flows (both to and from the United Kingdom) influence language growth. This. however, turned out not to be the case. None of the estimated coefficients was different from zero at standard levels of statistical significance. Additionally. an attempt was made to incorporate patent applications, a proxy for the impact of technological change on language growth, into the models. An internally consistent time series of patents applied for (or granted) was unfortunately not available for the period in question. Nor could we locate data on internal population mobility, which would be expected to promote uniformity in language use.

(43) The authors characterize two methods of calculating these marginal effects but state that one method, although computationally more challenging, is preferable: "It is conceptually better ... to report the average response ... over all individuals" (Cameron and Trivedi 1998. p. 81), which is the technique employed here.

(44) The results for population and government expenditure are somewhat ambiguous. The hypothesis that WORDS does not cause population is rejected at two lags but not at one lag in the full sample; it is rejected at all five lags in the 1830-1969 subsample, however. The hypothesis that WORDS Granger cause EXPENDITURES cannot be ruled out in either sample.

(45) "A World Empire by Other Means," p. 66.

(46) Ibid., p. 65. The condition of one word having many meanings is called polysemy. On this definition, set is the "polysemic champion.... [I]t has 58 uses as a noun, 126 as a verb, and 10 as a participial adjective. Its meanings are so various and scattered that it takes the OED 60.000 words--the length of a short novel--to discuss them all" (Bryson 1990, pp. 69- 70).

(47) Boswell ([1793] 1965. p. 73) explains in a footnote that

 Tarpeia, daughter of Spurius Tarpeius, governor of the citadel on
 the Capitoline Hill, traitorously opened the gates to the Sabines
 for the "ornaments on their arms." The soldiers crushed her to death
 with their shields, saying that these were the ornaments they had
 promised her. Traitors were afterwards put to death in Rome by being
 hurled from a rock that crowned this hill.

(48) Three 18th-century English writers--Joseph Addison, Daniel Defoe, and Jonathan Swift--did propose such regulation. though:

 Like a good protectionist, Addison wrote: "I have often wished that
 ... certain Men might be set apart, as Superintendents of our
 Language, to hinder any Words of Foreign Coin from passing among us;
 and in particular to prohibit any French Phrases from becoming
 current in this Kingdom, when those of our own stamp are altogether
 as valuable."

"Fortunately," according to The Economist, "the principles of free trade triumphed." (see "A World Empire by Other Means," p. 65)

(49) Remarking on the temporary respite from the Academy's hegemony occasioned by the regular exercise of Dr. Guillotin's renowned invention. Thomas Jefferson wrote to John Adams, "What a language has the French become since the date of their revolution, by the free introduction of new words!" (Cappon 1959, vol. II, p. 567).

References

Bate, W. Jackson. [1975] 1988. Samuel Johnson. Washington, DC: Counterpoint.

Becker, Gary S. 1976. The economic approach to human behavior. Chicago: University of Chicago Press.

Becker, Gary S. 1996. Accounting for tastes. Cambridge, MA: Harvard University Press.

Becker, Gary S., and Kevin Murphy. 2001. Social economics: Market behavior in a social environment. Cambridge, MA: Belknap Press of Harvard University Press.

Boswell, James. [1793] 1965. The life of Samuel Johnson. Abridged, with an introduction, by Bergan Evans. New York: Modern Library.

Braudel, Fernand. [1987] 1993. A history of civilizations, translated by Richard Mayne. New York: Penguin.

Bryson, Bill. 1990. The mother tongue: English and how it got that way. New York: HarperCollins.

Cameron, A. Colin, and Pravin K. Trivedi. 1986. Econometric models based on count data: Comparisons and applications of some estimators and tests. Journal of Applied Econometrics 1:29-53.

Cameron, A. Colin, and Pravin K. Trivedi. 1990. Regression-based tests for overdispersion in the Poisson model. Journal of Econometrics 46:347-64.

Cameron, A. Colin, and Pravin K. Trivedi. 1998. Regression analysis of count data. Cambridge: Cambridge University Press.

Cappon, Lester J., ed. 1959. The Adams-Jefferson letters: The complete correspondence between Thomas Jefferson and Abigail and John Adams. 2 vols. Chapel Hill: University of Noah Carolina Press.

Chiswick, Barry R. 1978. The effect of Americanization on the earnings of foreign-born men. Journal of Political Economy 86:897-921.

Choi, E. Kwan. 2002. Trade and the adoption of a universal language. International Review of Economics and Finance 11: 265-75.

Church, Jeffrey, and Ian King. 1993. Bilingualism and network externalities. Canadian Journal of Economics 26:337-45.

Collier, Paul, and Jan Willem Gunning. 1999. Why has Africa grown slowly? Journal of Economic Perspectives 13:3-22.

Dustmann, Christian, and Francesca Fabbri. 2003. Language proficiency and labour market performance of immigrants in the UK. Economic Journal 113:695-717.

Easterly, William, and Ross Levine. 1997. Africa's growth tragedy: Policies and ethnic divisions. Quarterly Journal of Economics 112:1203-50.

Feinstein, Charles H. 1972. National income, expenditure and output of the United Kingdom, 1855-1965. Cambridge: Cambridge University Press.

Greene, William H. 2000. Econometric analysis, 4th ed. Englewood Cliffs, NJ: Prentice Hall.

Grilliches, Zvi. 1957. Hybrid corn: An exploration in the economics of technological change. Econometrica 25:501-22.

Grin, Francois. 1996. The economics of language: Survey, assessment, and prospects. International Journal of the Sociology of Language 121:17-44.

Hayek, Friedrich A. von. 1960. The constitution of liberty. London: Routledge & Kegan Paul.

Hayek, Friedrich A. von. 1988. The fatal conceit: The errors of socialism. The Collected Works of F. A. Hayek, vol. I. edited by W. W. Bartley III. Chicago: University of Chicago Press.

Laird, Charlton. 1953. The miracle of language. Greenwich, CT: Fawcett.

Laitin, David D. 1994. The Tower of Babel as a coordination game: Political linguistics in Ghana. American Political Science Review 88:622-34.

Lamberton, Donald M. 2002. The economics of language. International Library of Critical Writings in Economics, vol. 150, edited by Mark Blaug. Cheltenham, UK: Edward Elgar.

Lazear, Edward P. 1999. Culture and language. Journal of Political Economy 107:S95-S126.

Levy, David M. 1997. Adam Smith's rational choice linguistics. Economic Inquiry 35:672-8.

Lewis, Bernard. 1995. The Middle East: A brief history of the last 2,000 years. New York: Touchstone.

Lieberson, Stanley. 2000. A matter of taste: How names, fashions, and culture change. New Haven, CT: Yale University Press.

Lynch, Jack, ed. 2002. Samuel Johnson's dictionary: Selections from the 1755 work that defined the English language. Delray Beach, FL: Levenger Press.

Marschak, Jacob. 1965. Economics of language. Behavioral Science 10:135-40.

McCloskey, D. N. 1985. The rhetoric of economics. Madison: University of Wisconsin Press.

McManus, Walter S., William Gould, and Finis Welch. 1983. Earnings of Hispanic men: The role of English language proficiency. Journal of Labor Economics 1:101-30.

Mitchell, Brian R. 1978. European historical statistics 1750-1970. Abridged edition. New York: Columbia University Press.

Mitchell, Brian R. 1988. British historical statistics. Cambridge: Cambridge University Press.

Pinker, Steven. 1996. The language instinct: How the mind creates language. New York: William Morrow.

Pinker, Steven. 1999. Words and rules: The ingredients of language. New York: Basic Books.

Pipes, Richard. 1999. Property and freedom. New York: Knopf.

Posner, Richard A. 2001. Antitrust law, 2nd ed. Chicago: University of Chicago Press.

Rubinstein, Ariel. 2000. Economics and language: Five essays. The Churchill Lectures in Economic Theory. Cambridge: Cambridge University Press.

Shughart, William F., II, and Robert D. Tollison. 1985. Corporate chartering: An exploration in economics of legal change. Economic Inquiry 23:585-99.

Shy, Oz. 2001. The economics of network industries. Cambridge: Cambridge University Press.

Smith. Adam. [1776] 1976. An inquiry into the nature and causes of the wealth of nations, edited by Edwin Cannan. Chicago: University of Chicago Press.

Tocqueville, Alexis de. [1835, 1840] 2000. Democracy in America, translated and edited by Henry C. Mansfield and Delba Winthrop. Chicago: University of Chicago Press.

Winchester, Simon. 1998. The professor and the madman: A tale of murder, insanity, and the making of the Oxford English Dictionary. New York: HarperCollins.

Zimmerman, Diane Leenheer. 2003. Authorship without ownership: Reconsidering incentives in a digital age. DePaul Law Review 52:1121-68.

Michael Reksulak, * William F. Shughart II, ([dagger]) and Robert D. Tollison ([double dagger])

* School of Economic Development. Georgia Southern University, P.O. Box 8152, Statesboro. GA 30460. USA: E-mail [email protected].

([dagger]) Department of Economics. University of Mississippi, P.O. Box 1848. University, MS 38677-1848, USA: E-mail [email protected]; corresponding author,

([double dagger]) Department of Economics, Clemson University, 222 Sirrine Hall, Clemson. SC 29634, USA; E-mail [email protected].

We benefited from comments on earlier drafts by John Conlon. Tyler Cowen, Mark Crain. Robert Ekelund Jr., Jack Hirshleiter, John Johnson, Walt Mayer, Robert McCormick. Charles Rowley, Hilary Shughart, Hsiao-Ting Su, Robert Tamura, and seminar participants at Virginia Military Institute. Two anonymous referees helped improve the article considerably. Thanks also to Winoto Pinata, Baoping Shang, and Birsel Tavukeu for able research assistance. As is customary, however, we accept full responsibility for any remaining errors.

Received March 2004: accepted April 2004.