Interlingual Communication in the Sciences
The writer of this paper is not a scientist, but a linguist who is interested in the sciences and, particularly, in communication.
First, then, science. Unlike all other study and discovery, the sciences have, with hardly any territorial or national exceptions (perhaps Soviet biology), a global acceptance: what chemistry, or any other science, says in English it says also in Japanese, uninfluenced by any personal, economic, political, national, religious, or other “ideology” or “myth” (the old word for what is now called ideology). In the words of the Report of the National Science Foundation (October 1957):
Science is not contained within national boundaries. The concepts, objectives, and methods of scholarship and scientific research, if not its language, are common to all nations. It is one of the few areas in which there is international understanding. Hence, better communication in this area has possibilities of assisting toward better understanding elsewhere.
Science is, in fact, in a strategic position to attack problems of global communication. Yet the primary and crucial attack is still to be made; and the time seems to be ripe for making it, now that we have swift, sure, and worldwide systems of intercommunication from any one part of the globe to any other or others that have modern transmitting and receiving stations.
The traffic jam that has developed in documentation is the consequence of trying to do first what should come last of all—abstracting, indexing, cataloguing, storage, retrieval and the like—and this in a variety of languages, over and over again, sometimes even in one and the same language. The result is that the entire structure has become superhumanly if not supermechanically top-heavy and overloaded, and is about to collapse of its own weight. The primary research necessary to create a truly interlingual (so-called “international”) means of communication is still to be undertaken: but it is the first step, and the problem of finding such a means of communication that will be
JOSHUA WHATMOUGH Harvard University, Cambridge, Massachusetts.
satisfactory and also supralinguistic (by which I mean as far superior to language as language is to the subhuman communication of the birds, beasts, and insects) calls aloud for solution.
We hear much, far too much from educationists, who repeat their complaint parrot-like one after another ad nauseam, of the dangers of specialization. But except for the rare few who have a gift for synthesis, specialization is necessary and inevitable: the specialist certainly has his place, and an important place. The contrast is brought into most vivid relief when a specialist undertakes to criticize a work of broad scope. During historic times there has been no fundamental change in human physiological and psychological capacities, or in human aptitudes produced by biological adaptation. What man has done in the twentieth century of the present era was inherent in man in the fifth century of the preceding era. What has changed has been the extension of his “memory” through the means of making and preserving human records, all the way from simple pictographic devices to the most recent and elaborate machinery for recording, storing, coding, and retrieving information. This is the problem of modern man: his brain remains what it has been all along, but the amount of data which he is feeding into it, with its limited input channels, has become vastly larger, causing what is essentially a breakdown of the system, freely recognized as such in individuals. Hitherto this modern problem has been met, but not solved, by the device of specialization, since so few possess the necessary faculty of synthesis, and (more important) since progress in any field depends on specialization. Who, nowadays, can take all knowledge for his province?
But it is the modern extension, in breadth and in depth, of human knowledge, more than specialization as such, that is dangerous, unless it can be overcome by the diffusion of information and by adequate means of intercommunication at a global, not merely national, level. This, so far as I see, cannot be done by merely developing existing methods of communication and documentation, but only by transforming the medium itself, the “code;” for then we shall attack the causes of the trouble, instead of its consequences; and instead, by so doing, of aggravating the present situation which we must seek rather to combat. For the work potential of the human organism is more or less fixed: the deficiency is in the “mechanics” of communication. Abstracting, as now conducted, has become an endless Penelope’s web, a taking apart by the less able of what other, and better, men have put together. Much more is required than documentation, which is inadequate, and sometimes quite inferior. A cataloguer in the Harvard College Library, supposing Charles Galton Darwin’s The Next Million Years to be concerned with divination, actually classified it as Folklore, and I had to point out that the book is an attempt at
scientific prediction by a sober biologist! (Incidentally it may be observed that one of the worst muddles in documentation is produced by the attempts, made in large library catalogues, to give, under the guise of a subject index, a bibliography. Those who do not know how to get at the bibliography of a subject ought not to be allowed inside a large library at all.)
In short, the primary research in interlingual communication, in the sciences (and in other fields), should be to reduce the medium of communications to a single form first, before indexing, abstracting, cataloguing, and documentation are attempted—these are the final steps if confusion and duplication are to be avoided.
Next, communication. Communication is a relation between persons, provided that they share one and the same means of communication, and this in its turn is essentially a matter of convention. In the ancient middle east cuneiform writing performed this purpose and the “interpreter” (dragoman) was a rarity; in Chinese the literate now use a single form of writing in the same way, no matter what form (dialect) of the language they speak; and in modern times Bishop Wilkins (1668) and Neurath (1937) have proposed this very solution of Babel, that is, a symbolism once removed from linguistic symbolism.
Linguistic symbolism proper has never succeeded as a means of worldwide communication, even partially. The common Greek of post-Hellenistic times was heard from Spain to the Punjab; imperial Latin from Britain to Asia Minor, and its successor, the written Latin of the church has circled the globe; Aramaic, as a written language of clerks and traders, from Armenia to Egypt and Persia; Arabic, as the language of Moslems, has been in use from Spain to Indonesia. Of these Latin and Arabic have been the languages of the learned, of scholars, and of scientists; but there is no turning back, despite modern attempts at the revival of Latin (Nuntius Latinus Internationalis and others). However effectively all these symbolisms may have served the Oecumene or inhabited world in the past, even of international scope, they can no longer do so, and certainly not Arabic, the only actual survivor in normal use. The British Association for the Advancement of Science explored the possibilities of Latin in 1921, and decided against it.
Anything that involves the use of a code book, whether based on numbers or words (e.g., simplified English) or an artificial language constructed for interlingual communication involves translation, and therefore is subject to all the delays, expense, inadequacy, and difficulties that all translation involves.
The discovery of babel came with the expansion of Europe, the age of voyageurs, missionaries, and traders; the invention of printing, and, in modern times, of the steamship, the telegraph, submarine cable, of the airplane, and now of radio, which have led to a movement for an “international” or “uni-
versal” language. To the Churchmen of the fifteenth and following centuries there have succeeded the philosophers of the eighteenth, who sought a rational system of signs to convey meaning, the anthropologist, and the scientist of the nineteenth and twentieth centuries.
Attempts to achieve the desired end have been based (1) a priori, on first principles—seventeen such attempts were made between 1629 (Descartes) and 1902 (Dietrich); (2) on the use of existing materials out of which to create an artificial language for worldwide use—over thirty of these are known from the “langue nouvelle” (1765, Faiguet) to Interlingua (1951) of the Internation Auxiliary Language Association; (3) on the revival of a language no longer spoken, usually Latin; (4) on the advocacy of a national language, with modification (e.g., basic English) or without it, i.e., depending on the historic spread of a language that normally accompanies the spread of a common civilization or culture; and in fact already a large number of words, especially of scientific terminology, are available and in actual use; (5) polyglot dictionaries.
In the past it has been the course of events, not the arguments of politicians, philosophers, or philologists, amateur or expert, that have made “world languages;” now the time has come for the interests of scientists, technicians, and businessmen to be served.
The difficulties of translation are notorious. Woodrow Wilson spoke of it as the compound fracture of an idea; perhaps he was thinking of the Chinese version, turned back into English as “invisible idiot,” for the original proverb “out of sight, out of mind.” The costly devices, or the time-consuming use of interpreters, at the League of Nations and, since then, in meetings of the United Nations, or at international congresses, are equally well known—and disliked. The cost of translations, or even of abstracts in translation, is staggering; there is great delay in making them available, and frequently they are not trustworthy. But the actual trend, the course of events, is and, for at least fifty decades, has been such as to make some form of interlingual communication daily more urgent.
Technically this ought to be a simple matter with modern methods of communication, which are quick and certain. Theoretically also it ought to be simple, for it is an axiom that the use of a common means of communication accompanies ease of communication. Yet linguistically serious barriers exist in the number of languages in use, and in the difficulties to be overcome, and the effort expended, in mastering more than two or three of them. This is true of Europe and of the Americas alone; the difficulties are merely multiplied if the continent of Asia also is considered, as of course it must be. The state of affairs is worse than paradoxical—it is exasperating, frustrating, and vexatious, a hindrance to the spread of scientific knowledge alone, all other considerations
apart. Even proper names are not exempt—Pfatten (in the old Austrian Tyrol) is now Vádena, and what used to be called Agram is now called Zagreb.
As for artificial languages, it must be conceded that they have failed. Out of thirty such that have been proposed, beginning with Esperanto, only six contenders are left in the field, and their supporters not only cannot agree which to accept but they even attack one another. Since 1889 over 700 periodicals devoted to the cause of Esperanto alone have been started; of these none lasted more than a few years, only fifteen survived in 1939, and today the number is even less. All artificial languages so far have been biased in favor of European ways of saying things, but there are other ways, more than one, of saying the same thing, and Asiatics and Orientals are not favorably impressed by what purports to be international and then turns out to be entirely Occidental and European.
A language is a code in which messages are transmitted. In other words it is a system of like or partly like recurrent features, together with their corresponding meanings, that is, it is a systematic symbolism which can be transformed into other systems, e.g., electrical (as in the telegraph or telephone) or electronic (as in the machines which have been built to reproduce “synthetic” speech, e.g., the Edinburgh PAT (parametric artificial talking device) built by W.Lawrence for the English Ministry of Supply); as a process, it generates a controlled sequence of symbols from a finite set, with an indefinite population of combinations and permutations (upon which variations of meaning depend). “Language” in the abstract is not a part of nature, linguistic events are; and also those events which the linguistic events (utterances), at first remove, symbolize. The “language” that anyone knows is, physically speaking, at the utmost the residues of traces of his lifetime of experience left as a series of routings and junctions in the nervous pathways of his brain, to form a statistical storehouse of memory, a cerebral assembly of patterns and of links between them, which provide for the controls and impulses that govern his acts of speech each time he “opens his mouth.” The mystery of all this lies in the matching with the actualities (meanings) which they symbolize, and with one another, of (1) the speech acts, the phenomena with which linguists deal; (2) the structual forms which these acts may variously take in different languages, sometimes with almost complete agreement (e.g., the first person singular pronoun, “I”), usually with far less, or none at all; (3) the corresponding written or printed marks on paper; or (4) any form of systematic symbolization which may be substituted for these last, and here there is almost unlimited scope for invention and improvement, most of all at an interlingual level. Symbolism (not to be confused with semiology) intervenes not only in protocol or basic statements, but also in subsequent, i.e., logically derived
statements, over and above primary situations, and this, as we shall see, introduces further difficulties.
Linguistic change, and interlingual equivalence, are both manifestations not of independent and isolated phenomena in series, but parts of a system or systems. Historical linguistic phenomena, that is to say, are parts of a given synchronic structure or status, and notwithstanding the possibility, and fact, of occasional “holes” in the pattern, there is such a thing as the “economy” of linguistic change. This has been revealed most clearly by the shift of interest from exclusively qualitative phenomena to the increasing study of quantitative linguistic phenomena.
Here I venture to refer the reader to some paragraphs in my book Language (London and New York, 1956, pp. 251–257; or in the paper cover edition, New American Library, 1957, pp. 224–229) on the general topics of the relation between language and the sciences, and to urge the reader to ponder them before preceding with the rest of this paper.
Since there is only one human species, it is obvious that all its members may well and ought to have a single means of communication. But all past experience shows that is not likely to be merely linguistic, which inevitably involves either prestige or translation. Human translation is riddled with shortcomings; and it has been asserted that “the simple basic fact that the meaning of every linguistic expression depends only upon incomplete, uncertain, and temporally shifting convention within a community of language has as its necessary consequence that equivalent reproduction of a text in a foreign language is impossible.” Gross blunders may have, and have had, serious consequences, even though the translators were believed to know what they were doing. The mathematical theory of communication furnishes proof of what was already well known, the degradation of any translated message.
In the meantime, that is, until a supralinguistic procedure can be devised and perfected, it therefore seems best to depend on the hope that mechanical translation may be achieved. Such translation has limited aims; it does not aim at perfection, but practical utility. Of the prospects of success, and of the uses of mechanical translation, I must leave Professor Oettinger to speak. He is an acknowledged expert and an active worker in the field. Here I insist only that documentation, abstracting, retrieval, and the like, should be performed on one language only, after foreign sources have been converted into it: this in the interests of economy both of effort and of money, as well as of simplicity and, in the long run, of effectiveness, granted that word-for-word translation usually needs exegesis—witness the early translations of the Bible into Gothic, Armenian, Old Church Slavonic, and other languages, all of which were done on a word-for-word basis.
To put the matter another way, linguistic analysis above the levels of phonology and morphology is still largely to be done: such things as word order, syntax, and even meaning are largely unexplored, at least in such a way as to be practically useful in mechanical translation. Attempts made by one or two logicians to cope with the problems presented have not so far given much promise of filling the need; indeed one of the foremost of them, Chomsky, explicitly declines to consider the problem of meaning at all! More promising are statistical investigations, which are more in accord with the mathematical theory of communication.
Here let me quote from a report on Mathematical Linguistics which I presented to the Eighth International Congress of Linguists held at Oslo in August 1957, and from some of my remarks made in the course of the discussion that followed.
In modern times statistical and mathematical methods applied to language were introduced by George Boole, who in his Laws of Thought (London, 1854, p. 245) records how the principle of determinate frequency was applied over a hundred years ago to deciphering Ogam and cuneiform; a number of frequency counts were made during the last century, e.g., by Förstemann, Lanman, and E.V.Arnold, to name no others; quite recently appeal was made to the same principle (of determinate frequency) in helping to decipher Linear B. It was the subject of a Harvard doctoral thesis in Linguistics in 1929 (HSCP 40, 1929, 1–95), begun two years earlier. The doctrine was not well received at the time, although the actual objections were far from being reasons (or, if there were any, the reasons for objection were well hidden below the surface). What quickened matters, and brought a mathematical statement, was the enormous progress made by communication theory and information theory during and since the war of 1939–1945, to which, on the linguistic side, Mandelbrot’s Contribution à la Théorie Mathematique des Jeux de Communication (Compt. rend. acad. sci., 232 (1951), 1638–1640, 2003–2005; afterwards Publ. de l’institut de statistique de l’Univ. de Paris, II, fasc. 1–2, 1953) and Oettinger’s recent studies (Harvard dissertation in Applied Mathematics, 1954) have made notable contributions, now being further developed in the seminar in Mathematical Linguistics conducted by Dr. Oettinger, some of the fruits of which it is hoped will be published in the near future.
Then there is the work of Shannon, both his Mathematical Theory of Communication (Urbana, Illinois, 1949) and a more recent paper on Prediction and Entropy of Printed English (Bell System Technical Journal, 30 (1951), 50–64), which raise two important questions. In the first place there is the question of the relation of the written language to the spoken. As to this, observe not merely that writing is systematic in its own right, as has been shown by Pul-
gram and by Uldall independently, but also that even a system of writing that is not phonematic obeys the law of great numbers—this was shown specifically for English “spelling” by Hultzén and Allen at the Chicago meeting of the Linguistic Society of America in December 1955. Then there is the matter of the transformation of one form of energy (and language is an as the Greek philosophers recognized, in whatever sense they used the term of language) into another (e.g., electrical), and since Shannon’s work brought this problem to the forefront there has been a great deal of misunderstanding. So far I know no one has identified physical entropy with the quantification of “information” (see my specific denial, Language, London, 1956, p. 201 note), where not entropy but ectropy (so called “negentropy”) is involved, and statements to the contrary, implying such an identification, are false. What has been said is that “Information” theory is a useful model of the working of language. It is also false to insinuate that the model, and the theory, infer a purely mechanistic view of language, as if it were all chance and no choice, restrained and directed as the choices are. But attention should be called to recent views of energy as developed by the Danish scientist J.N. Brønsted (died 17 December 1947) in his papers (1936 to 1946) on Principles and Problems in Energetics (New York and London, 1955) which deal with the transmission and transformation of energy, equilibrium, and coupling of different systems in ways more refined than the classical treatment of heat and energy. It might well be fruitful to enlist the cooperation of an expert in this field with an open-minded linguist, a neurologist, and a physiologist in order to discover exactly what correlations there are, if any, between language (both spoken and written) and other forms of human activity and consumption of energy. By contrast the work of linguists alone, who must depend upon more elementary knowledge, especially in mathematics and statistics, necessarily leaves something to be desired. Yet there are some praiseworthy, courageous, and unpredjudiced endeavors being made to reduce linguistic phenomena and findings to mathematical formulae, a few of which deserve to be mentioned here, such as the work of Lees and Swadesh in lexicostastics (since about 1952); of Greenberg’s quantitative approach to the morphological typology of languages (1954)—producing arithmetic indices that enable the degree of “polysynthesis,” “inflexion,” or “analytical” method (when word and morphome are identical and frequently monosyllabic) to be compared with precision, both with one another and in various languages; the objective criteria (devised by Annemarie Schlismann) for comparing the styles of authors (Innsbruck, 1955, cf. Word, 12 (1956), 293–298); correlation methods of comparing transitional ideolects by D.W.Reed and J.L.Spicer (presented to the Linguistic Society of America (Language, 28 (1952), 348–359); H.H.Paper’s
formulae for the depiction of symmetry in language (also presented to LSA); Pulgram’s formula for prehistoric linguistic expansion (1956); the older works of Zipf, Chrétien, Kroeber, and others. That there is ample scope for the development or correction of these results and, above all, for far greater rigor of method, as well as for new investigations, by as many as feel themselves attracted to or competent in this new and rapidly growing field of Mathematical Linguistics, goes without saying. It has the great virtue of revealing new, unsuspected, and wider horizons, exactly like the application of mathematical logic to biology in recent decades. Improvement of scientific method has always been a matter of science lifting itself up by its own intellectual and linguistic bootstraps; when these are also mathematical (which after all is only the Greek for intellectual), a subject, no matter what prior claims may have been advanced for it as a “science,” has at last become truly scientific.
I wish also to call attention to some previous presentations of the relativity of distribution of vocabulary, in particular those of Mlle. M.-L.Dufrénoy (on orientalizing influences in French romance), and of Professor Josephine Miles (on the continuity of poetic discourse in English).
I must, however, point out that nearly all the work done so far, by Yule, Guiraud, Herdan, Zipf, and others is not, everything considered, altogether above the criticism, either from the linguistic side or from the mathematical side, of being somewhat amateurish. For something more rigorous it is necessary to turn to the communication engineers, and their reports as published in the Bell System Technical Journal, in the Transactions of the Professional Group on Information Theory (Publications of the Institute of Radio Engineers, New York), and of the Symposia on Communication Theory and on Information Theory (London, 1953, 1956). It is not, indeed, clear that there has yet been a meeting of minds even here. Some of the linguists seem to me to be hidebound by old fashioned notions of language, and most of the communication engineers to be groping in the dark to get some notion of linguistic structure and function, and some others of them to be too much occupied with non-linguistic problems to concern themselves with purely linguistic theory, scientific or not. Moreover, when mathematics is despaired of, as it is by some investigators, it is always possible to resort to symbolic logic, perhaps a necessary preliminary way of formulating problems and solutions, especially in syntax, that may lead through mathematical logic to pure and not merely applied mathematics.
It is, however, quite clear that by means of the methods of statistical description of languages and of applied mathematics, for example as employed in attempts to devise automatic dictionaries and, even, by means of switching theory, automatic composition and translation, closer approach is being made to understanding and formulating the sequential processes, coding systems, and
functional symmetries of languages and of language, difficult as problems of context and syntax, idiomatic expression, variants of vocabulary (choice), and semantic alternants (synonomy and homonymy) are. In the widest sense of the term the processes are mathematical, and come within the domain of modern algebras. It is equally clear that there is need of mathematical linguists, that is, workers whose training lies both in modern linguistics and in modern mathematics, subjects not likely to be united in recent times once schooldays are over; and that there are great and important discoveries in store, far more important than anything to which comparing (say) the length of sentences in Marlowe and Shakespeare, is likely to lead.
Most continuous discourse is a matter of the making of routine linguistic decisions according to rules of habit and of choice, no matter what the language. Moreover it is not difficult to formulate these rules: they are to be found, if not in every grammar book, then in every complete description of a language prepared on structural principles. Each sequence in the stream of speech may be seen as a binary choice between that which is accepted (1) and everything else which is not (0): each phoneme, e.g., s or not -s; or each morphome, ness or not -ness; or each part of speech, a noun or not a noun; each construct, of phonemes (a phonematic construct, i.e., a morphome) or not; or of morphomes (a morphomatic construct, i.e., a word or phrase) or not, active or not, a sentence (a construct of epilegmata, “words”) or not—and so on, to the end of the utterance, whatever its length, a paragraph, a poem, a chapter, an act of a play, a book, an entire play, a whole library, all the libraries in the world. Such a mass of material, however, is unmanageably great: then we take random (i.e., unbiased) samples. For the nature of language is such that a sufficiently large sample will portray the whole, and languages obey laws as stable as any that have been found in human behavior, if not as stable as those of astrophysics (which are not entirely stable). If languages did not so obey, thanks to grammatical rules and lexical conventions, linguistic communication would be impossible. The number of linguistic phenomena in a language is not infinite—phonemes, morphomes, epilegmata, constructions; even the number of orders of arrangement itself, though much larger, still is not infinite: restraints are always imposed in the interest of understanding, of sense, of meaning. A highly convincing mathematical basis, underpinning the metastable equilibrium, historical and contemporary, of language, already found empirically by statistical investigations of linguists and communication engineers, is now actively being developed along lines of investigation that may fairly be called purely theoretical, that is, concerned with the formal and abstract linguistic order (pattern) that exists, and purely for its own sake. This task will not be carried much further, much less completed, without
enlisting the interest of the best mathematical minds. The theory of abstract sets is perhaps the most promising point of departure, since in language the one-to-many relation is at least as prominent as the one-to-one. Variation in categories, parts of speech, orders of arrangement, reversed order (as in interrogation in some languages), the presence or absence of infixation, inflexion, compounding, even rhythm and the like, the alternants presented in syntactical rules and a number of other variants that readily come to mind, all make for a considerable number of separate problems, none of them easy, that must be solved before a comprehensive theory can even be attempted. Context, nuance, style present quite terrifying, and still very remote, quests for the future. Alternations, already shown by sampling to rank high in frequency of occurrence, may not be the simplest to solve, practically useful as such solutions would be. Function and “meaning” will have to be differentiated, and especially units or categories that are concerned solely with functional relationships (e.g., prepositions) from those which are concerned with content of meaning, to use for the present terms that are convenient rather than trenchant. The identification of context, and perhaps even more, of word classes (as distinguished from class functions) and patterns of word classes, will call for recursive rules (rules of replication) in the formulation of tokens of sentence types with their recurrent structures. Some preliminary work of a descriptive kind (the analysis of discourse) only serves to show how much still remains to be done. The implications of conventional punctuation, the degree of internal coherence in a text (determined by subject matter), idiomatic expressions (that is, with respect to a given language), redundancy, unusual sentence patterns, partial incoherence or garbling—all these will have to be considered. It is a large program, but a stimulating one that calls for many workers, and may well produce theorems of elegance as well as of conviction; and, at last, a true mathematical theory of discourse.
Actually we may justifiably see, statistically and mathematically, mass linguistic regularity, when apparently some individual linguistic elements follow no regular laws, and without any appeal whatever to metaphysics. Language is regular; individual units of language show variation within this regularity. That is all.
In the meantime, and as a starting point, it may confidently be asserted that the mathematics of communication systems serves as a suitable model, beginning with any linguistic system. The factors involved not only in the linguistic system, but also in speech and in actual speakers, even the physical factors, articulatory and acoustic, in terms of force, intensity, energy, and other physical properties, will all have to be obtained before the time comes for anything like a general theory. This is a program of work for years to
come and for generations of workers. Only the merest beginning has yet been made to obtain physical correlates to analytical descriptions, to determine morphome or word boundaries; yet these are among many necessary preliminaries. Any text is a pragmatic affair, for it is a sequence of physical events, spoken or written, produced, transmitted, received, and identified—even sometimes reproduced and retransmitted—by physiological and neurophysiological organs; or it may be handled at certain points electrically or electronically. The preconceived assumption that linguistics, physics, physiology and neurology, force and energy are all completely independent of one another is precisely what has hindered and still hinders progress, most of all progress in linguistics. Actually they appear to be correlated in the most intimate fashion. But the only kind of unitary theory that will be able completely to comprehend this intimate correlation will be mathematical.
The statistical and mathematical methods do comprehend a complete description of the constraints operative upon combinations of the basic elements of linguistic systems, including meaningful utterances, that is utterances permitted, or at least not prohibited, by the system. In theory this may seem an impossible task; in practice, therefore, it is necessary to set, at least for the present, a practical goal, namely that of reasonable approximations. To do so is not unscientific, but rather characteristic of scientific method, and a clear evidence of confidence in linguistics as a discipline that is at last becoming both scientific and mature.
Notoriously syntax is difficult to formulate both accurately and exhaustively. Descriptive structural method has declined the task for thirty years, and still shows little appetite (or should I say aptitude?) for it. But mathematical logic already has made some notable, if only initial, attacks upon syntactical problems—the work largely of Spang-Hanssen and Yngve (on the analysis of gaps), of Bar-Hillel and Chomsky (on sentence analysis), and of Epstein (on grammatical categories)—in each case with results that are stated mathematically or symbolically.
The problem, most broadly stated, is to find the statistical structure of natural linguistic messages, all of which, in all languages, are decomposable into epilegmata, language being a sequence of discrete entities,1 as de Saussure and others have held. A message may be represented by a continuous carrier (e.g., magnetic tape), but it must be divisible into a set of discrete or symbolic units if it is to be received, indefinitely retransmitted, and understood. If language did
not exist, and we set about to create it, it would be necessary to string together sets of discrete segments—phonemes, morphomes, words, and so on, sentences, paragraphs, chapters. To deal with the totality of all texts is the unattainable ideal of the structuralists; mathematically and statistically this is neither an ideal nor even a necessity, since a sample text, or sampling of texts, suffices, provided that the sample is long enough to exhibit all the characteristics of a linguistic status. Difficulty arises from individual variations; but, as we have seen, these are covered by mass regularities (macroscopic), however great in practice must be the freedom permitted in a given microscopic problem—and “every case is a given case when you come to it.” Given the range of the number of epilegmata in a sequence of a selected size, and the weighted average of their properties of transmission (e.g., spoken or written), in all permitted combinations, we may determine all the possible messages. These will correspond to actual messages, i.e., agree with the mathematical theory (cf. Mandelbrot’s formula of a text),2 which is confirmed experimentally if we smooth out the steps in a crude plotting of the data.
A language may be either a closed system (e.g., classical Latin) or an open system (e.g. modern English). The status of the former (closed) may be described as a stable equilibrium, of the latter (open) as metastable, i.e., such that an infinitely small change will disturb the equilibrium. The fact of the universal existence of a linguistic status is known statistically.
Statistics distinguishes between finite, hypothetical (i.e., indefinite) and infinite populations (so Yule and Kendall). A language may be said to constitute a hypothetical population—only so can we explain the ability of a speaker to produce and understand certain new sentences and to reject others. That is, there is a finite state grammar, with a finite grammatical apparatus, that generates an indefinite number of sentences.
The system consists of a sequence of binary choices.
Language is a sequence of physical events, controlled by the nervous system which certainly behaves as an all or nothing mechanism. The conditions of communication depend upon a correlation of linguistic, physical, and physiological features. “Information” (in the technical sense) is concerned with non-conformation, surprisal, change in pro-presentation, the unexpectedness of the event (i.e., choice). Even “nonsense” is communicable, but at high cost, e.g.,
William Blake, James Joyce, Dylan Thomas, Gerard Manley Hopkins (all of whom have been regarded as eccentric, if not demented). Their “open” vocabulary has been described (by Mandelbrot) as a linguistic “high temperature;” in somewhat similar manner Yngve speaks of “open” positions within the regular syntactic structure, a sort of cement, this last constituting those aspects of a sentence which have the attributes of a calculus.
Shannon’s Mathematical Theory (1949) was misunderstood by some early linguistic reviewers (e.g., Fischer-Jorgensen). More recently (1957) Shannon has written as follows: “Language gradually evolves under continued use as an efficient communication code, i.e., human beings tend to adopt an approximation to the ideal encoding of communication theory.” This is what was implied, without a mathematical proof at the time, by my Theory of Selective Variation (1941, Harvard examination papers; 1948 Actes VIe Congrès des Linguistes; presented also at the summer meeting of the Linguistic Society of America, Berkeley, California, 1951), viz:
A linguistic status is produced by consistent selection which preserves the system from all possible or actual inconsistency of variation, thus giving a coherent, homologous, and significant pattern.
It is an elementary observation that transformations of communication are easy, not only from one code to another (as in translation) or by heliograph or semaphore, but also electrically or electronically (telephone, radio, teletype, or radiophoto, e.g., of the N.Y. Times to San Francisco during the Republican convention of 1956. These are all transformations of one form of energy (the spoken or written utterance) into another and then back again. Information theory treats a language system as a sequence of physical quanta of energy. Its mathematical formulation is applicable precisely because mathematics is concerned with structure, exactly as any language system is; the mathematical formulation, however, is both more exhaustive and less exhausting than articulate structural methods (as witness the work of Harris and Hockett).
Moreover, using methods analogous to those of combinatorial mathematics, we may apply it at different structural levels, as of phonemes, morphomes, epilegmata, and so forth. The continuous carrier, whatever it is, is segmented into uniform but discrete entities, whatever they are, and it is these which are received, retransmitted, and understood. The evidence so far available indicates that it is the epilegmata which are decoded and encoded one by one, with some slight time delay. No doubt the process would be more efficient if the entire text were coded simultaneously, but this is perhaps not practicable except in radiophotography of short texts. A compromise might be sentence by sentence. Hence arises the need for knowing the structure of sentences. In
what follows English sentences have been used; but the procedures are applicable in any language. The text should contain continuous samples long enough to exhibit all the basic characteristics of the language, but not so short as to exclude variations.
It is at once apparent that conditional probabilities are involved. Sentence structure is not haphazard, but subject to syntactic restraints. “Artificial” languages, including the utterances of children who have not acquired complete control of their language, and perhaps also basic English with its limited vocabulary, do not satisfy a general law, but the free-flowing language of schizophrenics is said to do so, a curious phenomenon the explanation of which is still to be found.
Two points are to be noted: (1) the maximization of “information” is completely analogous to the minimization of free energy (i.e., we have to do with ectropy, so-called negentropy or negative entropy); and (2) there are marked transitional probabilities (i.e., the transmission of each quantum restricts the choice of subsequent quanta that may be transmitted).
Coming now to structure, we must ask how this can be described. It is difficult to see how meaning can be used for this purpose, though Mackay’s definition of meaning (Information Theory, 1956, p. 219) is suggestive, viz:
the selective function of the message on an ensemble of the possible states of the conditional-probability matrix [of behavior patterns] in relative circumstances.
Is the number of meanings limited? And if so, by what? Roget classified the range of English epilegmatic meanings under 1000–1200 headings, Buck the meanings of I.Eu. roots under something over 1300. And if each such unit is regarded as having one (and only one) meaning, then a sequence of three such units (a sentence of three “words”) would occur as one of 1.7×109 theoretically possible sequences, of which not all, indeed only a fraction, occur in practice, since “nonsense” is excluded. Moreover it is necessary to distinguish between signs and symbols, symbols being open-ended (the number of meanings not being merely one-to-one), and also different symbols with identical tokens (e.g., bank, walk) not commonly distinguished except by context. So far, therefore, no means has been found of analyzing structure solely on a semantic basis. More promising is the attempt at analysis on the basis of word classes and functions (see Harvard Papers: Seminar in Mathematical Linguistics).
Theoretically a sentence of 50 words each of which is a token of two symbols would be one of 1015 possible sequences, but the actual (i.e., permissible) number is much less, as this sampling showed. It is higher, however, than proposed by Fries, Bar-Hillel, or the Wundheilers, none of whom found any solution to the problem of discovering a correct set of indices.
Sentences show structural ambiguity, and at first glance this calls for a complicated analysis. However, orders of arrangement govern not only sequences of epilegmata, but also sequences of word classes. There seems a remote possibility of classifying tokens uniquely (i.e., into a given class), and this presumably might be accomplished by contextual analysis, using redundancy (i.e., the presence of more clues than are actually necessary) in structure to compensate for ambiguity. Any sequence consists of dyadic combinations of binary-choices and the problem is to describe the degree of selective variation in the structural symbols, i.e., where there is identity of relation but multiple structure.
The constraints are greatest at or near the sentence boundary. Some other observations: if W2 is class 12 (verb), this fact supplies much information containing the occupant of the position W1. Class 3 (adjective) has low influence both forward and backward; class 5 (article; “no,” “each,” “some,” “all,” “any”—Russell’s logical words, i.e., they indicate structure, and may not be changed without changing it) has an exceptional forward influence. Rather high variation is permitted by classes 1 (substantive) and 4 (adverb). X2 tests indicate that no two adjacent positions are sufficiently similar to be considered as having been drawn at random from a given population (Shannon obtained the same conclusion in his study of English word order).
What are called “creative acts of the imagination” seem more likely to be variations permitted within the natural laws of language, the intrinsic regularity of which is such that, given sufficient bulk in sufficient variety, no intercepted messages remain forever undeciphered. Since its regular macroscopic properties represent a metastable relationship, which preserves equilibrium for equilibrium’s sake (i.e., for reasons connected with intelligibility), notwithstanding the random behavior of its microscopic constituents, the same mathematical methods may be applied as in dealing with the properties of matter in relation to the behavior of its constituents. Large scale studies of sentence structure will in all likelihood reveal (1) logical words, few in number, short in length, frequent in occurrence; (2) object words (“dictionary” words), numerous, variable in length, less frequent in occurrence. The former class, if changed, changes the structure of the sentence, and therefore its meaning; any of the latter may be replaced by the dictionary definition.
How far is it possible to attach quantitative measures to the relationship “language:meaning”? This problem has for the most part been shunned; perhaps the time is coming when it must be attacked more deeply by “information” theory, which so far has barely got beneath the surface.
Symbolic logic, insomuch as it also (like the algebra of abstract sets) deals pre-eminently with structure, may be serviceable in solving some problems:
for example, symbols enter into relations which are reflected in the sequential arrangements of their corresponding tokens; then these relations become symbols in their turn, the features of which may readily be described by logical methods. Syntactical relations (among the tokens) should not be confused with operations, e.g., 2+2 is not a sentence, but a structured name for the integer 4, no matter how stated (Housman, Asquith, Chaplin).
More attention should be paid to punctuation. There was distributed at the Fifth International Congress of Linguists a monograph (printed at Göteborg, 1939) in which patterns of structure indicated by phrasal boundaries were identified in fourteen principal European languages. It was apparent that transition probabilities are involved here also. Hence a sentence may be viewed as a sequence not only of word classes but also of phrasal and clausal classes.
A number of schemes using single syllables, combined with numbers (1 to 9), but demanding the use of a code book, have been devised for multilingual communication, particularly at airfields; none has proved satisfactory, because of the delay involved; attempts to manufacture more or less artificial languages for international use are equally unsuccessful (see Language, 1956, pp. 53–58). Some method of supralinguistic symbolism, to which, as we have seen, the sciences are already much inclined, will have to be devised—of this more below. Meanwhile translation is all that can be done, notwithstanding all its imperfections, but it is perhaps the best hope, especially if mechanical translation can be perfected for practical use, as no doubt it can. So far it proceeds practically as word-for-word translation, and this is often roundly condemned. There is, however, something to be said for a word-for-word procedure. In the use of language memory, which depends on input, output, and feedback, plays a large role, even in “creative” writing. It uses chiefly the existing pathways and switching points in the brain; in original composition it creates new ones, which the reader or hearer will decode. All the evidence thus far available tends to show that this will be done, in any particular language word-by-word, not, as might superficially have been supposed, in groups of words, or idea by idea, except possibly in formulaic composition, or in some linguistic types, where the arrangement of words, in accordance with “grammatical rules,” may go into more or less stereotyped order (phrases). New neurological investigation may compel a modification of this statement; but the best research so far conducted views very unfavorably the notion that we receive language, physically speaking, by “ideas” rather than through the words which infer them.
Mandelbrot has shown that linguistic structures, characterized as they are by regularity of pattern, which is governed by probabilities of choice, are adapted to word coding with a minimum of effort, the “economy” of the mechanics
of language (glossodynamics, or glossostatics); in his words a message is purposively produced in such a way as to be decoded word by word in the easiest or “least costly” possible fashion.
Difficulties arise because of the difference in pattern in different languages, as well as from differences in meaning in different contexts even in one and the same language.
The adoption of a single language (English, Spanish, Russian, Latin) for global use is not only unlikely in a divided world of many cultures that are still poles apart: it has been attempted, and found wanting. The same is true of “artificial” languages, all the way from Esperanto to Interlingua (and basic English). Yet we understand well enough what the nature of communication is and what the nature of language is. Both are characterized by symbolization, by system or good order, and therefore both lend themselves to coding on a single, universal, and global level. There are grave difficulties; but they are not insuperable—what the human mind has done in creating babel, the human mind (ex hypothesi) can cope with.
Lines of attack that suggest themselves are: (1) study not only of the nature of symbols and tokens of symbols but also of the combinations and relations of symbols one to another, even of the relations of such combinations one to another, which then become symbols and tokens of symbols at a higher level in the hierarchy of communication; (2) study of the nature of the cerebral processes that take place when overt meaning (as in public utterance, written or spoken) is suppressed (as in “thought”), which should lead (3) to a more profound understanding of the nature of meaning itself; (4) the invention of means of combining form (symbolization analogous to that which is used in logical syntax) with content (what is commonly meant by meaning).3
This last is a challenge to communication engineers. One thinks of switching theory as providing solutions to not dissimilar problems, problems of a far more elementary kind. One imagines (for imagination is needed at this stage) that the final result will no more resemble language than a sewing machine looks like eight fingers and two thumbs. It goes without saying that electronics is the springboard for a leap to the universal “grammar” that I venture to foresee, if I am not quite bold enough to prophesy. I do not believe that it is be-
yond the ingenuity, courage, and persistence of modern communication theory, or of the builders of those machines, Univac and the rest, that, at present in limited fields, in the future in all fields in which men think, do better than unaided humans.
Once created, the universal “grammar” (I use the word for want of a better) of communication will reduce the labor and cost of documentation, storage, and retrieval, by at least half; and will at least double its accuracy and trust-worthiness.
It has commonly been held that one cannot communicate, at least at any except the lowest levels and in the simplest terms, without communicating in a particular language. This dogma, which is valid to a certain point, lies at the base of all the strife over the language question in Indian, African, and Asian universities. Macaulay held that English and English literature should be taught in the Indian universities the establishment of which he advocated—in those days literature, significantly enough, was not taught at all at Cambridge, or in any self-respecting university. There are those who consider that the use of Maltese in the University of Malta is a brake on the island’s progress, and Livingstone held that the survival of native languages (to which Afrikaans may now be added) was a dividing force in Africa. But even language, with all its fine qualities as an instrument of communication, is far from perfect; this unique power among animate creation, is certainly capable of improvement, of sharpening, of finer perception and clearer expression.
Language represents an enormous step forward, above and beyond the means of communication used by sub-human animals. Is it not conceivable that methods over and above language can be achieved? The development of Mathematical Linguistics, of parametric artificial talking devices, of a universal grammar that would truly be infinite in scope—all depending upon further development of switching theory and a better understanding of the nature of meaning, both private and overt, i.e., of the content of expression—this is the real crux, together with better means of relating symbols and tokens, and of these, in their turn, as symbols and tokens of symbols, in a hierarchy of language (here switching theory might surely play a part)—all this suggests that such an achievement is not inconceivable, however far it may be in the future. It is moreover a necessary preliminary step to true interlingual communication, which should be simpler in the sciences than elsewhere; and this a necessary preliminary step to be taken before abstracting and indexing can be anything but chaotic. The recent announcement that computators can develop the capacity to learn, as well as to operate, is definitely encouraging.
If the notion of evolutionary emergence is correct, there is no reason why the present symbolic level should not be vastly improved, if not surpassed.
Human experience is not limited to a mere slice of the universe in the way in which it is to specialized creatures (biologically speaking), each species of which by its very specialization is restricted in what, and how much, may enter its “ambient world.” What reason is there for supposing that man has reached the limits of scientific and technological progress in the scope of his peculiarly symbolic activity? In any event all that we call science and technology is systematic symbolization, not merely the concatenation of symbols, but combinations and arrangements of them in such a way as to enhance their symbolic value, a grammar, not just a vocabulary. This may be fertile and productive without limit. A formula is a symbolization devised for a particular symbolic realm so that it becomes part of a great “thinking” mechanism, which, in its turn, imposes what I have called language engineering (see my forthcoming Lowell Lectures, 1957)—the next forward step.