[This is an electronic edition of a paper originally presented at a meeting of the Linguistic Circle of Madison at the University of Wisconsin (Madison, Wisconsin, U.S.A.) on 12 December 1967 and then at meetings elsewhere in 1968 and 1973. In the meantime, it circulated widely in mimeographed form and finally appeared in print in Notes from the Linguistic Underground, edited by James D. McCawley, Syntax and Semantics, vol. 7 (New York: Academic Press, 1976), pp. 229-242. It is posted here with the kind permission of Elsevier, the present copyright holder. For the home page of Syntax and Semantics go to ScienceDirect. In order that the reader may locate corresponding passages in the volume in which it was published in 1976, I have indicated the original page numbers in square brackets. A change that I have had to make affects the references, which in the published version were given in-text in abbreviated form, with full bibliographical entries provided only at the end of the volume. Here, for the convenience of the reader, I have placed all references in the footnotes. This has necessitated renumbering the original notes. I should also point out that as far as I recall there were no notes of any kind in the typescript of the paper when I delivered it in Madison in 1967. I regret that no copy of that version of the paper has survived. This is unfortunate because some of the endnotes added to the published edition nine years later are not consistent with the theorerical and historiographical positions that I defended in the original paper. It seems clear to me now that by 1976 I had begun to question some of my earlier presuppositions. In this electronic version, please note that I reproduce an example of the branching diagrams that Wundt uses in his book Völkerpsychologie (see, for instance, Volume 1, Die Sprache, 2nd part, pp. 322, 327, 333-335). This use of branching diagrams obviously points ahead to their later extensive use by generative grammarians to indicate constituent structure, although the historical links are still unclear. Throughout this Web version of the article I have made minor textual and orthographical corrections. The reader should be warned, therefore, that this is not a verbatim transcript of the 1976 article. However, I have abstained from updating the content of the article. Instead of that, I have added new comments at the end in two sections, headed "Further Thoughts" and "Provisional Conclusions." There I summarize my ideas on these questions at the present time. I should like to take this opportunity to thank colleagues who have commented in print and otherwise on my 1976 article. [Among these contributions, I might particularly single out Giorgio Graffi's stimulating "L'analisi in costituenti immediati prima di Bloomfield," Lingua e stile, 25 (1990), 457-469, from which I have derived much benefit.] Needless to say, I welcome further comments now that a version of the article is being made available on the World Wide Web. The copyright holder requests that single copies of this electronic version of the article be downloaded and printed for the reader’s personal research and study only.]
THE TEXT OF THE 1976 ARTICLE
There is at present a widespread notion that immediate-constituent analysis is a modern version of traditional syntactic analysis. Zellig Harris, for example, has recently asserted: "Traditional grammar established various distinguished segments of sentences which were hierarchically subdivided into smaller |[p. 230] segments (in a manner made explicit by Leonard Bloomfield, as the method of immediate constituents)." Similarly, John Lyons sees "an obvious parallelism between immediate constituent analysis and the traditional procedure of parsing sentences into subject and predicate, and each of these, where appropriate, into words, phrases, and clauses of various types." I shall argue in this paper that traditional grammar and immediate-constituent analysis are logically incompatible, and that immediate-constituent analysis has a non-traditional origin.
Traditional grammar is a family of linguistic theories represented in the grammars written before the advent of scientific linguistics. I use the expression "family of theories" rather than the word "theory," since traditional grammar is not a single, unchanging conceptual object. I assume, however, that it has certain fairly stable defining features. For convenience, I take many of my examples from the Latin grammar of Allen and Greenough (1931) and the Greek grammar of H. W. Smyth (1916), since both these works are still in print and can be consulted by interested readers.
The aim of traditional syntactic description is, to quote Smyth, to show "how the different parts of speech and their different inflectional forms are employed to form sentences" (Greek Grammar, p. 255). The sentence, it may be noted, is regarded here as a combination of words, i.e., it is defined synthetically. In fact, the four basic units of traditional grammar: letter, syllable, word, and sentence, form a straightforward ascending hierarchy, and the Greek word syntaxis itself suggests the idea of arranging things in an ordered array.
Thus, sentences are thought of in traditional grammar as combinations of words, not combinations of phrases. However, certain phrasal elements are recognized by most traditional grammarians. For instance, periphrastic verb forms of the type amatus est 'he was loved' are treated as single words for morphological purposes and listed in paradigms along with forms consisting of single words, like amatur 'he is loved.' It is as if such periphrastic forms are considered to be substitutes for non-existent single words. In grammars of the vernacular languages, this notion is sometimes made explicit. For example, Adelung in his Deutsche Sprachlehre für Schulen ("German Grammar for Schools") defines auxiliary verbs as "diejenigen Verba ... welche die mangelhafte Deutsche Conjugation in Vergleichung mit den vollständigern anderer Sprachen ergänzen helfen" ("those verbs that help to make up for the deficiency of the verb morphology of German in comparison with the more complete morphologies of other languages").
Another type of phrasal constituent that has been recognized by traditional grammarians, especially in the last century or so |[p. 231] can be exemplified by the English expression near the window in a sentence such as He was standing near the window. This type of expression is said to function as a single syntactic unit but to consist formally of a preposition and a noun. It is useful to draw a distinction between "form" and "function" since the same form may have several different functions. Prepositional phrases, for instance, occur as predicate complements (as in The carriage is in good condition), as objective complements (as in He found the carriage in good condition), and as adjectival modifiers (as in the lass with the delicate air).
But note that in traditional grammar the head and its attribute are not said to constitute a phrase. In the last example quoted, with the delicate air may be called a phrase, but neither the delicate air nor delicate air are referred to as phrases. There is in fact no such thing in traditional grammar as a noun phrase in the sense in which this term is used today by professional linguists. Traditional grammarians do not divide sentences into phrases without residue; traditional grammar knows nothing of phrase structure.
In most traditions, both classical and vernacular, the sentence is said to contain two necessary elements, namely a subject and a predicate, or in some traditions three: subject, verb, and object. However, there has always been a persistent tendency to think of these essential elements as single words, not as strings of words. For instance, Allen and Greenough (New Latin Grammar, p. 164) cite canis currit 'the dog runs' as a minimal complete sentence, and similarly Smyth (Greek Grammar, p. 255) quotes êlthe kêryx 'a herald came.' At the same time, however, the terms "subject" and "predicate" are usually defined semantically: the subject as the person or thing spoken about, and the predicate as that which is said about the subject. One might reasonably expect, therefore, that where the subject noun or predicate verb is accompanied by modifiers, the entire resultant phrase would constitute the subject or predicate, as the case might be. But this has for the most part not been the case. In a sentence such as Latin vir fortis patienter fert 'a brave man endures patiently,' the subject is the word vir 'man,' not the phrase vir fortis 'a brave man,' and the predicate is fert 'endures,' not patienter fert 'endures patiently.' Quoting this sentence Allen and Greenough (New Latin Grammar, p. 166) state "the adjective fortis 'brave' modifies the subject vir 'man.'" Thus, not only is the expression vir fortis not a subject, it is not a phrase either, a phrase being defined by Allen and Greenough (New Latin Grammar, p. 166) as "a group of words, without subject or predicate of its own, which may be used as an Adjective or an Adverb." Vir fortis is a group of words without subject or predicate of its own, but it is being used |[p. 232] here neither as an adjective nor as an adverb. Similar arguments apply to the expression patienter fert; it is neither a predicate nor a phrase.
It should also be noted that a subject is a subject with respect to some finite verb, not with respect to the sentence as a whole. Thus, a student attempting to construe an unfamiliar Latin sentence is told to "look for the finite verb, and then find its subject." Note that he is not invited to try dividing the sentence in two and to call one portion the subject and the other the predicate.
In addition to these principal sentence elements, other subordinate elements are recognized in traditional grammar, namely the so-called modifiers and complements. Adjectives are said to be modifiers when they appear attributively in close association with a substantive; direct objects are considered complements, as also are most prepositional phrases.
The notion of modification has an interesting peculiarity. An adjective is said to modify the substantive it accompanies. In certain circumstances, therefore, an adverb may modify an attributive adjective, i.e., an adjective that itself modifies another word. Thus, in an English expression such as very hot water, the adverb very is said to modify hot, which itself modifies water. The result, then, in cases of this type is a hierarchy of modification. But it is also possible for a word to be modified by several words between which no relation of modification exists. For example, in the English expression an old man with leprosy, the substantive man is said to be modified by the adjective old, by the indefinite article an, and by the prepositional phrase with leprosy. In such cases, no hierarchical structure of modification is recognized.
So much for traditional syntactic theory. Let us now examine immediate-constituent analysis as expounded in Bloomfield's Language and further elaborated by Rulon S. Wells and C. E. Bazell. In particular, I should like to focus attention on what are, I believe, three fundamental characteristics of this kind of syntactic analysis. These are the notions:
1. That any sentence breaks down or can be split binarily into a subject part and a predicate part. Thus, the English sentence Poor John ran away breaks down into poor John, the subject part, and ran away, the predicate part.
2. That some groups of words are syntactically equivalent to single words. Thus, the group of words very rich is equivalent syntactically to the single word poor in the expressions very rich man and poor man.|[p. 233]
3. That the analysis of a sentence yields a single unbroken hierarchy of groups. For example, the sentence Poor John ran away is analysed first into poor John and ran away. The first of these two constituent parts of the sentence is in turn analysed into poor and John, and the second into ran and away. In no instance are discontinuous constituents recognized, say poor ... away, nor do any constituents overlap, as they would if we posited poor John and John ran as constituents. Note also that two constituents are recognized each time the process of analysis is applied to a sequence. To divide a sentence such as English John loves Mary into three immediate constituents (John, loves, and Mary would not be considered a normal analysis by the practitioners of this theory.
If we scrutinize these three basic features of immediate-constituent analysis from the vantage point of traditional grammar, we must conclude that the two theories are fundamentally incompatible. Let us consider first the notion of a binary split of the sentence into subject and predicate. This would make terminological sense to a traditional grammarian only for sentences consisting of two words (such as Latin canis currit 'the dog runs'). In all other instances, the traditional grammarian would baulk since he applies the terms "subject" and "predicate" to single words. Moreover, he has no way of referring to the constituents that the immediate-constituent analyst calls by these same terms. This is because his terminology in the area of phrasal constituents is rudimentary. Moreover, for the traditional grammarian, word order and syntax are separate topics. In Latin, for example, word order is relatively independent of syntactic relations, and the operation of isolating the constituents that the immediate-constituent analyst calls subjects and predicates would be a useless exercise in the few cases where it could be done. It is, therefore, no accident that immediate-constituent analysis did not form part of traditional syntax.
As for the syntactic equivalence of word groups and single words, this has some support in traditional grammar. Prepositional phrases were considered equivalent to adverbs or adjectives functionally, and periphrastic verb forms equivalent to non-periphrastic ones. It is also noteworthy that periphrastic forms consisted in most cases of two words, as did prepositional phrases (minus any modifiers, of course). But, as I have already pointed out, equivalence of noun to noun phrase or of verb to verb phrase was absent from traditional grammar.
Finally, let us consider the hierarchical features recognized in traditional grammar. While there is some conceptual affinity between the traditional noton of |[p. 234] hierarchies of modification and the hierarchical principle basic to immediate-constituent analysis, the two are nevertheless separate notions and lead to different analyses in concrete cases. In a phrase such as English very cold beer, the grammatical tradition said merely that beer is a substantive, that cold was an adjective modifying the substantive beer, that very was an adverb modifying the adjective cold, and therefore that the same relation of modification that obtained between cold and beer also obtained between very and cold. Implied in the traditional theory was, of course, the notion that cold is the locus of two relations, one obtaining with beer and the other with very. In immediate-constituent analysis, on the other hand, cold had no relation to beer whatever, but rather the phrase very cold had a relation to the word beer, which paralleled the relation holding between the single words very and cold.
These two analytical solutions are, I should like to suggest, logically incompatible. Either there is a relation of modification between the single word beer and the phrase very cold, or there is a similar relation between the single word beer and the single word cold. Both analyses cannot be true, one of them (or possibly both of them) must be false. I conclude, therefore, that immediate-constituent analysis was not a more explicit version of traditional syntactic theory, but that the two were logically separate and distinct. Hence, immediate-constituent analysis, when it was first thought of, must have constituted a break with the grammatical tradition. I said above that immediate-constituent analysis was formulated by Leonard Bloomfield in his 1933 monograph Language. But was Bloomfield really the earliest spokesman for this type of analysis?
If we examine his earlier book An Introduction to the Study of Language (1914), what we notice first of all is that instead of giving the sentence its traditional synthetic definition as a combination of words, Bloomfield defines it analytically as "an utterance analyzing an experience into elements." The relation between the various sentence elements has, according to Bloomfield, a special "psychological tone called the logical or discursive relation." This special psychological tone "consists of a transition of the attention from the total experience, which throughout remains in consciousness, to the successive elements, which are one after another focused by it."
The picture is then the following: A sentence corresponds psychologically to a total experience, and each word in it to a constituent element of this total experience. The experience remains in consciousness as a totality while the sentence is being uttered, but the attention is focused on each successive |[p. 235] constituent word as it is uttered. Bloomfield proceeds as follows: "We can attend to but one thing at a time. Consequently the analysis of a total experience always proceeds by single binary divisions into a part for the time being focused and a remainder. In the primary division of an experience into two parts, the one focused is called the subject and the one left for later attention the predicate ... If after this first division, either subject or predicate or both receive further analysis, the elements in each case first singled out are again called subjects and the elements in relation to them, attributes" (Introduction, p. 60).
I hardly need to point out that the notion of binary splits, the bipartite sentence, the hierarchical principle, the analytical approach from the sentence down to the individual syntactic units are all clearly expressed in this passage. It is also noteworthy that the terms "subject," "predicate," and "attribute" are used in ways at variance with their traditional meanings. This idiosyncratic use of traditional terms raises a further question. Where did Bloomfield get this analysis from, if not from the grammatical tradition?
The answer is clear. The immediate source of Bloomfield's syntactic theory was Wilhelm Wundt. In the preface to his 1914 monograph Bloomfield writes: "It will be apparent that I depend for my psychology, general and linguistic, entirely on Wundt; I can only hope that I have not misrepresented his doctrine. The day is past when students of mental sciences could draw on their own fancy or on popular psychology for their views of mental experience" (Introduction, p. VI).
Let us accordingly turn to Wundt's monumental Völkerpsychologie. The section we are concerned with bears the title Die Sprache and first appeared in two parts in 1900. In the second of these two parts, Wundt defines the sentence as "den sprachlichen Ausdruck für die willkürliche Gliederung einer Gesammtvorstellung in ihre in logische Beziehung zueinander gesetzten Bestandtheile," that is "the linguistic expression for the arbitrary division of a total idea into its constituent parts placed in logical relation to one another." Here, then, we have an explicit analytical definition of the sentence, in contrast to the traditional synthetic one.
But one may ask at this point whether Wundt was aware that he was breaking with traditional grammatical theory in this regard. The answer again is clear. He devotes several pages to an extensive critique of what he calls the definitions of traditional grammar "die Definitionen der alten Grammatik" (Die Sprache, 2nd part, pp. 222-226). Here he argues against the traditional definition of the sentence as a combination of words on the following grounds. First, a sentence may |[p. 236] consist of a single word. Second, not just any combination of words is a sentence. For example, an enumeration of the signs of the zodiac is a combination of words, but can hardly be said to constitute a sentence. Whether a combination of words qualifies as a sentence depends on how they are put together: "es müsste also hinzugefügt werden, wie das Ganze beschaffen sein muss, um als Satz zu gelten" "it would be necessary to specify in addition what the totality must be like to qualify as a sentence" (Die Sprache, 2nd part, p. 224). Third, the notion of "thought" that is normally invoked in traditional definitions (e.g., "A sentence is a combination of words that expresses a complete thought") is inherently vague (Die Sprache, 2nd part, p. 225).
Nor does it help to replace the notion "word" by "concept" and define the sentence as a combination of concepts. Hermann Paul in his Principien der sprachgeschichte (1880) had offered such a definition, which reads as follows: "Das wesen des satzes besteht darin, dass mehrere vorstellungcomplexe in beziehung zu einander gesetzt werden durch nebeneinanderstellung der wörter, an die sie sich angeschlossen haben" ("The nature of the sentence consists in the fact that several conceptual complexes are placed in relation to one another by juxtaposition of the words with which they have become associated"). According to Wundt, this definition suffers from the first two defects mentioned in connection with the traditional definitions and from a number of additional ones. One of these is the fact that the definition is worded in terms of the thought that generates the sentence. But the thought and the sentence are separate phenomena, since what one person thinks while he is uttering a particular sentence may be quite different from what another person might think when uttering the same sentence. Wundt draws the following conclusion (Die Sprache, 2nd part, p. 230): "Der Satz ist in erster Linie ein sprachliches Gebilde, ein psychologischer Vorstellungsverlauf nur insofern, als dieser wirklich im Satze ausgedrückt wird, und vollends ein logisches Urteil nur unter der Bedingung, dass er direct eine Aussage enthält" ("The sentence is first and foremost a linguistic formation; it is a sequence of psychological concepts only insofar as such a sequence is in fact expressed in the sentence; and finally, it is a logical judgment only if it directly contains an assertion").
Indeed, Wundt prefers the traditional definitions to the pychological one suggested by Hermann Paul, arguing that while in most cases a sentence is a combination of words, it is false to call a sentence a combination of concepts. For while a person is uttering a sentence the constituent concepts do not patiently wait to turn up in his consciousness until the corresponding words are about to be uttered. On the contrary, the sentence is a psychological unit present in the speaker's mind throughout the time he is uttering it. It is true that the word actually being uttered is usually in the |[p. 235] focus of attention (im Blickpunkt des Bewusstseins), but the other word-concepts, or at least those essential for the meaning of the whole sentence, are in the middle ground of attention (in dem weiteren Umfang des Bewusstseins). Moreover, the principal constituent concepts of a sentence are already in consciousness the minute one starts uttering it.
Wundt, therefore, feels justified in asserting that the sentence has a twofold character: it is both simultaneous and successive. It is simultaneous in the sense that throughout its production the speaker is conscious of it in its entirety. It is successive in that the state of consciousness (der Bewusstseinszustand) varies from moment to moment as particular constituent concepts pass into the focus of attention and others pass out (see Die Sprache, 2nd part, p. 236).
According to Wundt, Hermann Paul's mistake was to carry over the outer grammatical form (die äussere grammatische Form) into the area of consciousness, to assert that each particular grammatical form (jede äussere Form) is a true reflection of the underlying psychological processes as they take place (von Moment zu Moment ein treues Abbild der zugrunde liegenden psychischen Vorgänge). When a sentence is uttered, it is, of course, the result of a set of psychological processes, but the latter are quite distinct from the sentence itself. Thus, Wundt advocates distinguishing between the psychological factors that bring about a particular concrete utterance and the external grammatical form (die äussere grammatische Form) of the sentence in question, and he accuses Paul of confusing the latter with the inner psychological structure (das innere psychische Gebilde). Wundt's linguistic theory is, therefore, one of those in which the correspondence between inner form (in his case the psychological factors) and outer form is not one to one.
To recapitulate the argument so far, Wundt was the immediate source of Bloomfield's untraditional definition of the sentence, and he was fully aware that in this regard he was breaking with the long-standing synthetic definitions of sentencehood. Moreover, he presented psychological arguments to support his rejection of the grammatical tradition on this vital issue. But one may ask whether he was responsible for Bloomfield's notion of the binary sentence split and the idea of a single analytical hierarchy of sentence constituents. If we examine the further discussion in Die Sprache we find unmistakable evidence that he in fact was.
Having established the simultaneous character of the sentence, Wundt goes on to discuss its character of being a sequence of successive elements. Here he emphasizes that a sentence is a set of elements among which certain relations obtain; it is not merely an aggregate of individual items.|[p. 238] The basic relations among the words of a sentence are the ones familiar to the grammarian, i.e., subject, predicate, attribute, adverbial, direct and indirect object. If these relations are divested of their logical character and regarded from a purely formal point of view, it becomes obvious, according to Wundt, that they invariably obtain between two words: subject and predicate, verbal predicate and object, nominal subject or object and its attribute, and finally verb and adverbial modifier. These combinations occur in their most typical form when the subject, or predicate, or object, etc. is a single concept. Then other concepts can be associated with these primary ones and be related to them by co-ordination or subordination.
Co-ordination introduces a new element into sentence formation in that it can be extended over indefinitely many members. All other grammatical relations are exclusively binary. If one asks why this is so, Wundt provides the following rationale. First, a logical relation is by its very nature restricted to the two concepts between which it obtains. Similar relations may, of course, obtain between some third concept and one of the two members of a logical relation. However, the process of establishing such a relation requires an extra act of thought that in turn is of the binary variety. Therefore, any analytical relation is a mental act that embraces two and only two members. This again follows from the duality of the principal syntactic constituents.
For example, a sentence such as German Ein redlich denkender Mensch verschmäht die Täuschung 'a sincerely thinking person scorns deception' can be regarded in all its parts as nothing more than binary combinations whose members are in turn binary combinations (Die Sprache, 2nd part, pp. 318-319). The two major constituents are related as subject to predicate, all other pairs exemplify the same relation in compressed form (in verdichteter Form). Thus the subject includes the assertion (Aussage) ein Mensch denkt redlich 'a person thinks sincerely,' and the predicate the assertion die Täuschung wird verschmäht 'deception is scorned.'
Wundt then goes on to provide a geometrical diagraming system to mirror these relations. The formula represents the close subject-predicate combination, and G stands for Gesammtvorstellung, the total concept embodied in the whole sentence. Thus, a sentence like A sincerely thinking person scorns deception could be diagramed thus:
A stands for a sincerely thinking person, B for scorns deception, A1 for a person, B1 for thinks sincerely, A2 for deception, B2 for is scorned, A3 for thought, and B3 for is sincere.
It may be pointed out that Wundt thinks of the relation between the two members of a combination as logical rather than grammatical, for as he points out, it is necessary to subject the sentence to a variety of grammatical changes in order to bring out these underlying relations (for instance, the change of scorns deception to deception is scorned). The grammatical form of a particular combination depends on whether it is the principal assertion (die Hauptaussage), and if not, in what relation it stands to the principal assertion.
That Wundt's linguistic theory is the source of Bloomfield's theory of immediate constituents is, I believe, too obvious to require further demonstration. Bloomfield may be complimented for relaying Wundt's theory so faithfully in his early book. The difference between the position he adopted in 1914 and the more familiar approach of the 1933 monograph deserves more study. Whether there are antecedents to Wundt's theory is another interesting question. It may be recalled that the primacy of the sentence over the word was emphasized by Wilhelm von Humboldt. But it is not known at present whether Humboldt's ideas were in any way influential in Wundt's abandonment of the traditional synthetic definitions of the sentence. It is clear, however, that Wundt did not develop his linguistic theory specially for his book on folk psychology. The same theory, expressed in almost identical terms, can be found in the first volume of his Logik (1880). It would seem, therefore, that the gestation of the theory may have taken place around the time |[p. 240] when Baudouin de Courtenay was active in Kazan. If this is the case, we have one more reason to regard the 1870s as an unusually creative period in the history of linguistics.
FURTHER THOUGHTS (2007)
The strict dichotomy between traditional grammar and 'scientific' linguistics posited in this article cannot be maintained. First, traditional grammar did not precede linguistics: the two coexisted, albeit uneasily, at least from the beginning of the nineteenth century. Moreover, we professional linguists are now aware, much more than we were thirty or forty years ago, that traditional grammar is not a monolithic entity. That having been said, I still maintain that some of the most cherished beliefs held by immediate-constituent theorists of the 1950s were incompatible with fundamental tenets of syntactic analysis as it had been widely practised in the near and distant past. The crucial influence of Wundt on Bloomfield, moreover, seems reasonably clear, although much in Wundt's scholarly ideology is far from clear, and Bloomfield's theoretical outlook undoubtedly changed when he repudiated Wundtian psychology and embraced the theories of A. P. Weiss (1879-1931), who like Wundt was both a psychologist and a philosopher of science. As I pointed out in the 1976 version of this article, Wundt's criticism of his linguistic colleagues was aimed at what he called the definitions of traditional grammar.
To understand this we must remember that a fundamental characteristic of the Western grammatical tradition was that it was built on definitions of key terms. Grammatical textbooks were full of definitions. This was partly because grammar was taught to youngsters, who need to be given thumbnail definitions. From antiquity right up to the mid-twentieth century, moreover, grammarians and linguists concentrated much of their fire on their opponents' definitions and terminology or on definitions and terminology that they categorized as traditional. As an eminent twentieth-century linguist once expressed it: "We are a pugnacious profession"! The so-called structuralists of the first half of the twentieth century were, if anything, even more obsessed with rigorous definitions than their predecessors had been. What was peculiar in the American structuralist toolkit, however, was the notion of operational definitions. With the vogue for so-called distributional procedures came a requirement that for a new term to be scientifically valid it had to support an infallible discovery procedure. This fact helps to explain much that might otherwise seem bizarre in the procedures of the immediate-constituent analysts of the 1940s and 1950s.
Another peculiarity of immediate-constituent analysis and of much linguistic theorizing of the 1940s and 1950s was the preference for binary analyses. (As we shall see in a moment, this idea goes a long way back, but linguists did not know this at the time.) Breaking down sentences into two and only two immediate constituents was one example of this obsession at work. Obviously, in morphological analysis, many words can also be divided in two. Thus, English unfriendliness consists of the suffix -ness and unfriendly, which in turn consists of the prefix un- and friendly, which breaks down into friend and the suffix -ly. It is interesting to note that Bloomfield most often applied the notion of immediate constituent in morphological analysis (see Language, pp. 209ff., p. 221f.). In that area, we should recall, binary analysis had been extensively practised by the traditional grammarians of Sanskrit in India, with whose work Bloomfield was well acquainted. (Hence, we have yet another source for his binary procedure.) Later theorists in the U.S. also continued to apply immediate-constituent analysis below the word level. For reasons that are unclear to me, this fact tends to be forgotten nowadays.
Strictly speaking, however, by the time he wrote his monograph Language Bloomfield did not divide sentences into "subjects" and "predicates," because by then he disapproved of those two quintessentially traditional terms, replacing them with his own terms "actor" and "action." Rulon Wells did not use those two traditional terms either, but perhaps for a different reason. He may have been loath to create terminology to refer to phrasal categories. If he divided an English sentence such as Poor John ran away into two halves, he was reluctant to label the resulting pieces. The aim was to discover each layer of immediate constituents inductively by means of scientific, i.e., reproducible, procedures, and ideally the analyst should eschew the traditional phrasal terminology in order to underline his emancipation from the errors of the past. (See Wells's own reference to the terms "subject" and "predicate" in his article "Immediate Constituents," §14.)
Dividing sentences binarily into subjects and predicates was clearly not Wundt's invention either, nor had it been a recent innovation. I was guilty of exaggeration when I suggested in endnote 5 of the 1976 article (footnote 8 in this electronic version) that the binary subject-predicate analysis of the sentence that I had seen in A New English Grammar by the English grammarian E. A. Sonnenschein (Oxford, 1916) "does not reach very far back." In fact, this procedure was already common practice among grammarians in the nineteenth century and as far back as the High Middle Ages and early Renaissance, at which point it had been introduced from the analysis of the proposition into subject and predicate by logicians. For the most part, however, grammarians did not subject entire sentences to analysis. Moreover, when they used the terms "subject" and "predicate" they tended to cite cases in which subjects or predicates were single words, like, for instance, English Smoke rises. This was to a large extent a distant legacy of traditional logic, which tended to deal with basic propositions consisting of just two categorematic terms, namely a one-word subject accompanied by a one-word predicate. I might cite a couple of typical examples of this practice from the Logica magna, a late medieval textbook of logic by Paul of Venice (died 1428): Socrates currit 'Socrates runs' and Plato est bonus 'Plato is good.' See Logica magna Pauli Veneti (Venice: Albertino Vercellese, 1499; HC *12505), f. 102ra.
If grammarians were aware of logic, logicians were necessarily also aware of grammar throughout this period. But the fact that close relations existed does not mean that grammarians and logicians analysed sentences and propositions in the same way. This was because their perspectives were different. Thus, a logician, well aware that for his colleague the grammarian a logical constant such as necessario 'necessarily' was simply an adverb, could submit to elaborate logical analysis a proposition containing it, such as anima Antichristi necessario erit 'the soul of the Antichrist will necessarily exist.'
Originally, grammar and logic had been separate disciplines that were regarded in the Middle Ages, Renaissance, and the early modern period as part of the so-called 'liberal arts' (within the liberal arts, to be precise, they were components of the trivium). However, cross-fertilization between those two curricular traditions increased as time went on. The High Middle Ages, a period responsible for the explosive growth of university education, brought grammar and logic closer together. Close relations also existed in the Middle Ages and the early Renaissance, especially in southern Europe, between grammar and rhetoric, which was the third component of the trivium. Individual scholarly disciplines never live in a vacuum, however. In so many cases the same students would sit in grammar classes as sat in the classes that covered logic and/or rhetoric. Moreover, in many cases the same instructor taught one subject as taught the other. Then as now, teachers were forced to be jacks-of-all-trades.
At first, the advent of Renaissance humanism in early fifteenth-century Italy had the effect of severing the connections that had existed in the immediately preceding period between grammar and logic, but in the following century these connections were re-established. The earliest example of this rapprochement that I have seen so far is in the works of the Protestant humanist Philipp Melanchthon (1497-1560). Like many educators, Melanchthon taught both grammar and logic. Indeed, he undertook to set up a special Protestant curriculum covering all the basic scholarly disciplines.
The fact that logic was an integral part of the curriculum in the sixteenth century meant that the analysis of propositions into subjects and predicates was familiar to students and could be appealed to on occasion in grammar classes. It is interesting to observe Melanchthon injecting elements of logical syntax into his grammatical textbook Syntaxis and defending this practice on pedagogical grounds. Specifically, he argues that unless students know how to divide a sentence into subject and predicate they cannot understand its structure, and he suggests that such an analysis is based on natural principles (on what he calls natura sermonis 'the nature of speech'). Logicians are the people who point the way here, says Melanchthon. It is useful for youngsters beginning the study of Latin to be taught the terms subject and predicate so that they can understand how sentences are put together.
A century later, we witness a similar kind of importation of logical syntax into grammar on the part of the Port-Royal gentlemen in France: as is well known, they wrote both a treatise on logic and a number of grammatical textbooks, the most famous of which is the Grammaire générale et raisonnée which first appeared in Paris in 1660. The eighteenth century saw the appearance of a number of rational grammars that further developed this kind of mixed logico-grammatical analysis. A good example is the Grammaire générale, ou exposition raisonnée des éléments nécessaires du langage by Nicolas Beauzée (Paris, 1767).
"For, although thought is indivisible," says Beauzée in that book, "logic succeeds in breaking it down after a fashion by focusing separately on the various ideas that make up its content, so to speak, and on the relations (French relations) that unite them in thought (dans l'esprit). Any relation (relation) presupposes a first term and then a second term. Moreover, any idea that is the second term of a relation (rapport) is at the same time the first term of another relation (rapport)" (Beauzée, Grammaire générale, v. 2, pp. 1-2).
Regardless of Beauzée's priority in this development, about which I make no claim here, the fact remains that he expressed the basic concept of a hierarchical chain of relations uniting the constituents of a sentence. Syntax, therefore, could be reduced to a chain of relations. Almost two centuries later this kind of hierarchy was to underlie immediate-constituent analysis as formulated by Rulon Wells in his seminal 1947 article. The many intermediate steps are still not entirely clear. As is well known, Leonard Bloomfield set the ball rolling in American linguistics in his monograph Language of 1933, but as I showed in my 1976 article Bloomfield was in part building on ideas that he had personally inherited from Wundt. What complicates the picture, however, is that by trade Wundt was a psychologist, i.e. neither a linguist nor a grammarian. He could not reconcile his view of sentence structure with what he claimed were the definitions of traditional grammar. That he would make such a blanket assertion is undeniably puzzling because we know now that in the mid-nineteenth century some pedagogical grammars of German, notably the Ausführliche deutsche Grammatik by Karl Ferdinand Becker (two editions dating from 1836-1843), had already made the transition from a synthetic to an analytical approach to the sentence on roughly the same logical lines that we see indicated in Beauzée's grammar.
What further complicates the picture is that Wundt himself was a philosopher with wide interests in many branches of that field, including logic: after all, he wrote a major monograph entitled Logik (Stuttgart, 1880). Logicians, of course, with their sophisticated predicate calculus, had been breaking down propositions into subjects and predicates since classical antiquity. As we have seen, Wundt saw an antithesis between his own way of analysing sentences and one characteristic of what he conceived of as traditional grammar. I suggest that this theoretical move on his part may still reflect the old disciplinary split between logic and grammar, at least in part. At the same time, his analysis of the sentence was firmly anchored in his own psychological theory of "apperception." He was not merely a grammatical theorist.
In the meantime, however, the old non-logical approach to sentence structure, which also goes back to antiquity (take Priscian's Institutiones grammaticae as the prime example), had survived into the nineteenth and twentieth centuries. By the non-logical approach I am referring to the view that the sentence is not a basic grammatical unit, but merely results from combinations of words, and therefore that the only truly basic grammatical unit is the word. A language, viewed from this perspective, is a collection of words and ways of using them in word-groups, i.e., expressions of varying length.
I might suggest in passing that this non-modern perspective may in part have been what led Ferdinand de Saussure to relegate much of sentence formation to parole 'speech,' not to langue 'language,' a decision that shocked some linguists who followed in his wake in the twentieth century. It is interesting to note that in the posthumous Cours de linguistique générale nothing is said about the sentence as a syntactic unit and in general attention is focused away from syntax to what the author termed the "syntagmatic" and "associative" relations among words and affixes. It is as if Saussure regarded the sentence as merely one of several types of word-groupings. In this connection, it is a curious, almost ironic, fact that Rulon Wells, the prime initiator of post-Bloomfieldian immediate-constituent analysis, was a major interpreter of Saussure's theoretical system. The widespread focus on the sentence as a syntactic unit by many schools of twentieth-century linguistics may reflect the continuing influence of the logical tradition. Wells himself was philosophically trained.
This all illustrates the fact that types of syntactic analysis inculcated in schools in recent times have varied greatly, and different systems have even competed with each other within the same country and more or less at the same time. When I myself studied English grammar in England in the early 1940s, I was not introduced to the type of subject-predicate analysis that had been promoted by Sonnenschein and the Joint Committee on Grammatical Terminology half a century earlier (see footnote 8 in this electronic version). In the United States, on the other hand, a sophisticated system of sentence-diagraming, still popular in the mid-twentieth century, was propounded in textbooks written in the 1870s by Alonzo Reed and Brainerd Kellogg, two English teachers at the Brooklyn Collegiate and Polytechnic Institute in New York. (At this time, Reed was described as an "instructor" in English Grammar, while Kellogg was a "professor of the English Language and Literature.") The Reed & Kellogg diagrams were similar to Wundt's tree diagrams (although the two systems are not known to be directly connected historically). It is interesting to note that Reed & Kellogg's approach to grammar rested on the traditional subject-predicate analysis of the sentence with a heavy reliance on the notion of dependency, which also had deep roots in the grammatical tradition. As far as I am aware, the vogue of the Reed & Kellogg system did not spread beyond the United States.
The fact that such analytical systems as the one developed by Reed and Kellogg in the nineteenth century were still popular in the United States in the twentieth century may partly explain the statement made by Zellig Harris that I cited at the beginning of the 1976 version of this paper. Much more relevant here, however, is the curious fact that in the course of the nineteenth century the term parse changed its meaning from describing individual words in terms of their respective parts of speech to analysing entire sentences in terms of their various sub-parts and specifying their syntactic interrelations. (See the two definitions of the term parse in recent electronic editions of The Oxford English Dictionary.) Most probably, John Lyons had this more recent meaning of parse in mind when he made the statement that I referred to at the beginning of my 1976 article. It is much less likely that his statement was an allusion to the British school tradition inaugurated by the English grammarian E. A. Sonnenschein, as I imagined earlier.
As I see it now, the creation of a terminology to refer to sentence parts and word-groupings of various kinds has had a long and complicated history. In traditional grammar, as we see it at the end of the classical period in the Institutiones grammaticae of Priscian (ca. 500 A.D.), sentence parts and word-groups were virtually ignored. This attitude, which seems peculiar to us nowadays, first began to change towards the end of the first millennium. What caused this development, we think, was the need that many teachers felt to convey basic grammatical notions to students whose native language was no longer Latin or any variety thereof. Vivien Law wrote insightfully on this development. See her Grammar and Grammarians in the Early Middle Ages (London: Longman, 1997). The valuable work of Roger Wright, Bengt Löfstedt and others may also be consulted in this context.
Naturally, various systems of syntactic terminology arose during the many centuries that we perversely lump together under the rubric of the Middle Ages. One discovery that grammarians made quite early was that much of sentence structure can be illuminated once the main verb in a nuclear clause has been identified. This insight is reflected in rules compiled by grammarians of that period which called for the verb to be regarded as "governing" the case of all the major nominal elements in a simple sentence (not just the constituent that we nowadays call the object or complement). The notion of government had not been completely absent in Priscian's grammar, but it was not especially emphasized. Teachers of grammar in the Middle Ages grappled with the problem of formulating rules on how to analyse and construct sentences from the verb outwards, as it were.
Dividing the sentence into two principal constituents was a move that was made about 1300 in the treatises on the modes of meaning (de modis significandi), in what we nowadays call the "modistic" tradition, and these two sentence halves were called the suppositum and the appositum, corresponding to the logicians' terms subiectum and praedicatum respectively. Since the modistic grammars have been extensively studied by scholars in recent decades, I shall not say anything about them at this point. One additional comment suggests itself, however. In roughly contemporary southern Europe (ca. 1200-1400), as I have pointed out above, an analysis of the sentence into three parts was prevalent. These three parts were the two I have just mentioned, plus the verb, so that for grammarians in this tradition their term appositum referred to the predicate minus the verb. This tripartite division meant that verbs could be classified in terms of the nominal elements they "governed" on both sides, to the left and as well as the right. (The terms left and right were used in relation to the verb, which was pictured as occurring between them in the natural order Subject-Verb-Predicate.) Oversimplifying, one might say that transitive verbs governed a noun in the nominative case to the left of the verb, and a noun in the accusative case to the right of the verb, while intransitive verbs governed a nominative on the left and no accusative on the right. Although the resulting elaborate classification of verbs must have been an extremely practical tool for students composing in Latin, it cannot be said to have yielded a general theory of sentence structure, which is perhaps unfortunate. Beyond the nuclear clause this system handled subordinate clauses with difficulty, if at all.
One might have expected this system to be discarded as soon as Renaissance humanism arrived, but this was not the case. In that regard, the fifteenth-century Italian humanists were still more or less completely medieval, and that is even true of an outstanding figure such as Lorenzo Valla (1407-1457), whose influential Elegantiae linguae Latinae circulated widely from the mid-1440s on. Its vogue soon spread to northern Europe, mediated by such eminent scholars as Erasmus of Rotterdam.
As I have pointed out above, the logical analysis of nuclear clauses into subjects and predicates is certainly present in a treatise on syntax written by Philipp Melanchthon in the mid-sixteenth century. Future research will doubtless fill out this picture. It would be interesting to discover whether Melanchthon had any immediate precursors. Sufficee it to say that after the sixteenth century a blend of logical and grammatical analysis steadily gained popularity, as many scholars have shown. The landmarks are the famous Grammaire générale et raisonnée of 1660 and a series of "general" grammars that appeared in France during the period of the Enlightenment. However, the traditional kind of grammatical analysis inherited from late antiquity, the kind that considered the word to be the basic grammatical unit, coexisted with that newer tradition and survived into the twentieth century in some quarters. Historically speaking, this vacillation in basic orientation may even be a distinctive feature of traditional grammatical theorizing of recent date.
We need to distinguish several theoretical issues here. First, there is the question of whether the sentence is regarded as more basic, more "real," than the words composing it. In such a framework, sentences are broken down into words, and words only exist as components of sentences. Alternatively, words are given primary status, and then sentences are viewed as resulting when words have been correctly concatenated. (We must bear in mind at all times that neither sentence nor word are pre-theoretical givens.) Second, there is the issue of whether there are intermediate units between the sentence and the word. At what point, one may ask, does the phrase, in the English sense of that term, come into the picture? (It was certainly already being used in English grammars of the mid-nineteenth century.) Third, there is the vogue among grammarians of the logical terms subject and predicate. Fourth, there is the question of how many parts the sentence should be primarily analysed into, which is connected with but not the same as the procedure of analysing sentences into subjects and predicates. Fifth, there is the issue of the sorts of relations assumed to hold between sentence parts. Finally, there is the use in the classroom of graphic devices to mirror the analysis of sentences into parts and sub-parts and all their interrelations.
As for the diagraming system that we see in the works of Alonzo Reed and Brainerd Kellogg, it was copyrighted in 1868 by Reed and O. H. Hall. The copyright date of Graded Lessons in English was 1876, and Higher Lessons in English in 1877. As regards the two authors, Reed died in 1899, according to the "Online Computer Library Catalogue." I know nothing about Kellogg other than the fact that he was a full professor at the Brooklyn Collegiate and Polytechnic Institute at the same time as Reed was an instructor at that institution. I know nothing about O. H. Hall either.
An interesting fact is that Reed and Kellogg had a worthy predecessor, namely Stephen W. Clark (1810-1901), who published a number of popular grammar textbooks, one of which is entitled The Science of the English Language: A Practical Grammar, in which Words, Phrases, and Sentences are Classified according to their Offices, and Their Various Relations to one another, Illustrated by a Complete System of Diagrams. The first edition (but can we be sure that it was the very first?) dates from 1847 (New York: A. S. Barnes & Cincinnati: Derby, Bradley & Co). I have access to an 1863 edition bearing the copyright date of 1855. It is conceivable, therefore, that Clark's example stimulated Reed and Kellogg. Superficially, however, their respective diagraming systems look quite different. Clark drew balloons round all the words in a sentence, and then the relative position of the balloons mirrored the various grammatical relationships between the words. Reed and Kellogg, on the other hand, used what one might call a stick notation: the sentence was symbolized by a long horizontal line with a small vertical line separating the subject from the predicate. From the horizontal line other lines sloped off to the right to symbolize the various modifiers of the major constituents and modifiers of modifiers.
An even more interesting question is whether S. W. Clark also had predecessors in this matter of diagraming. I do not know the answer to this question. The development of grammar writing in the United States is discussed by Charlotte Downey in her stimulating article "Trends that Shaped the Development of 19th Century American Grammar Writing," in English Traditional Grammars: An International Perspective, edited by Gerhard Leitner (Amsterdam/Philadelphia: John Benjamins, 1991), pp. 27-38. Downey divides nineteenth-century English grammarians in the United States into two groups: traditionalists and innovators. One characteristic of the traditionalists was to conceive of the sentence as divided into three major parts: subject, predicate, and object. The innovators, on the other hand, divided the sentence binarily into subjects and predicates. She does not mention W. S. Clark, but I can report that he falls squarely into the traditional camp in that while he uses the terms "subject" and "predicate," he divides the sentence into subject, predicate, and object. For instance, he makes the following statement: "The Object of a Sentence, being distinct from the Grammatical Predicate, is properly regarded as a distinct Element in the structure of such Sentences as contain Objects" (Practical Grammar, p. 27). He distinguishes between the logical and the grammatical predicate thus: "The Logical Predicate includes the Grammatical Predicate and its Object" (ibid.). Hence, among grammarians of English active in the United States the binary division of the sentence into a subject half and a predicate half was an innovation of the nineteenth century, but it was not shared by all grammarians. In England, on the other hand, J. C. Nesfield, author of English Grammar Past and Present (1898), was still a traditionalist in this respect, and Nesfield's influence was still felt in England in the mid-twentieth century.
As regards the development of syntax in the United States in the nineteenth century, one grammarian of that period, namely William Swinton, makes the following revealing statement: "The introduction, some thirty years ago, of the method of Sentential ANALYSIS, devised by the German philologist Becker, and adapted to American school use in the meritorious works of Professor Greene and others, marks the only considerable innovation, in this country, on the Murray system" (William Swinton, A Progressive Grammar of the English Tongue [New York: Harper, 1878], pp. iii-iv). Swinton is referring here to the famous grammarian Lindley Murray, whose English Grammar was popular in the United States in the early part of the century. The Professor Greene whom Swinton mentions here, was Samuel S. Greene (1810-1883), who published a number of highly successful textbooks, one of which has the suggestive title A Treatise on the Structure of the English Language: or the Analysis and Classification of Sentences and their Component Parts (Philadephia: Cowperthwait, 1846).
A decade earlier, a successful textbook writer called Boswell C. Smith published a series of grammars dubbed Smith's New Grammar. An edition to which I have access has the suggestive title English Grammar on the Productive System: A Method of Instruction Recently Adopted in Germany and Switzerland (Hartford: Spalding & Storrs, 1838; copyright filed at the District Court of Massachusetts, 1831). In the preface to that work (pp. 3-6), Boswell Smith explains at some length that his own system incorporates the best features of two different contemporary European educational systems, namely what he refers to as the "Philanthropist" School and the School of Jean Henri Pestalozzi. The significance of Pestalozzi in the history of education is well known, the Philanthropists less so. They were followers of Johann Bernhard Basedow (1723-1790), who preceded Pestalozzi by a generation. At this point, I should perhaps add an interesting fact not mentioned by Boswell Smith, namely that both these schools had also created new pedagogical methods for arithmetic that were specifically designed for teaching elementary school children. Grammar was another basic discipline like arithmetic that had to be covered in elementary schools in the New World.
In addition to the educational aspect, it seems that familiarity with the early nineteenth-century German grammarian Karl Ferdinand Becker (1775-1849) was crucial: Becker's system was what caused English grammarians in the United States to abandon the traditional synthetic approach to sentence analysis and adopt a binary division of sentences into subjects and predicates. Note, moreover, that we encounter in this period the notion that grammar was the study of language in general and also the novel idea that it had already become a "science." Accordingly, scientific-sounding terms like "analysis," "synthesis,""element," and "structure" enter the picture. At the same time, the focus on relations (rapports) that we have seen in Nicolas Beauzée was added to this intoxicating terminological mixture.
That Becker's ideas were already well known among teachers of grammar in the United States by the middle of the nineteenth century is proved by the appearance of an explicit attempt to apply them to English, namely in a book by Josiah W. Gibbs entitled Philological Studies with English Illustrations (New Haven: Durrie & Peck, 1857). In his preamble, Gibbs states that the principles advocated by him "have been derived for the most part from distinguished German philologians, particularly from the writings of Dr. Karl Ferdinand Becker, and are here illustrated from our own language." Moreover, he also makes it clear there that in this regard he is not the first to advocate these ideas: "Although many of these principles are now current in our schools of learning, the writer hopes that the publication may not be amiss" (p. iii). Gibbs's ideas are relevant to the history of IC analysis in that he insists on applying the principle of phrase structure to the analysis of sentences. To quote him verbatim: "According to the view of modern philologists, a sentence or proposition is made up immediately, not of words, but of syntactical groups or combinations of words" (p. 39).
Various mnemonic devices widely used by twentieth-century professional linguists in the course of their work on syntax, such as diagraming and arithmetical notation, likewise go back to the nineteenth century, at least in the United States. Whether they go back farther than that I do not know at this stage. As we have seen, sentence diagraming had already appeared in S. W. Clark's grammar in the 1840s, and in its further development at the hands of Brainerd Kellogg and his colleague Alonzo Reed two decades later it enjoyed immense popularity in the United States.
Another graphic device which has an undeniably scientific cachet is the use of capital and small letters to represent syntactic categories and sub-categories (e.g., N standing for Noun, VP standing for Verb Phrase), etc. This useful procedure is already found in Samuel Greene's An Analysis of the English Language (Philadelphia: Cowperthwait, 1874), pp. 23-24. Greene introduces the idea as follows (a short extract will suffice to give an idea of what he had in mind):
"The entire sentence and its several elements may be represented by a system of significant symbols.
"(a.) No system can be devised to represent all the subtle distinctions of thought, and consequently no attempt is made to meet every emergency. It is proposed to show in a condensed form, the general features of any sentence.
"1. With the exception of V, taking the initial letters of the elements, capitals for the principal elements and small letters for the subordinate, we have S for subject, P for predicate, a, o, v, respectively for the adjective, objective, and adverbial elements" (Samuel Greene, Analysis of the English Language, p. 23).
Accordingly, a simple sentence like "Ice melts" is represented in Samuel Greene's system with the letters S P; and "Red squirrels eat nuts" becomes a S P o. For sentences containing more than one member of the same class one after the other, he resorts to coefficients. A sentence such as "Those large red cherries tempt us" is represented 3 a S P o, the 3 indicating that three elements of the adjective class occur before the subject. (These examples all appear on p. 23 of the 1874 edition.)
Greene also uses exponents, e.g., S1 P2 represents a sentence such as "Birds are bipeds," which contains a subject of class 1 followed by a predicate of class 2. Here, 1 is the default value, so a sentence like "She walked a queen" is represented S P P, without exponents (Samuel Greene, Analysis of the English Language (1874), p. 24). There are a number of further notational refinements that I cannot go into here. Suffice it to say that Samuel Greene could handle complex sentences of some length and complexity, as for instance "Wit is a shining quality that everybody admires; most people aim at it, all people fear it, and few love it unless in themselves" (see the same page).
Even the expression "immediate constituent" was used by one of these nineteenth-century grammarians. S. W. Clark, namely, distinguishes between what he calls the "proximate" and the "ultimate" analysis of a sentence: "[...] the Proximate Analysis of a Sentence consists in resolving it into its immediate Constituent Elements (S. W. Clark, A Practical Grammar, [New York: American Book Company, copyrighted 1875] p. 69). On the other hand, Clark explains, "the Ultimate Analysis of a Sentence consists in reducing its Proximate Elements to the WORDS which compose them." Thus, words were regarded as the ultimate constituents of a sentence, not morphemes, since the notion of the morpheme and Baudouin de Courtenay's technical term "morpheme" were unknown at this time. While Clark's use of the expression immediate Constituent Elements does not appear to signal the conscious creation of "immediate constituent" as a technical term, the underlying notion of an analysis of the components of a sentence into their ultimate and immediate constituents is clearly expressed.
Let me now return to the two main topics of my original paper, namely (1) the historical source of the theory of immediate constituents, and (2) the role of the phrase in the historical development of grammatical analysis. As for the first question, we do not know and may never know what system of grammatical analysis Bloomfield was exposed to as a schoolboy, but it is clear that some of the basic conceptual and terminological ingredients of the system that he was to present in his 1914 and 1933 books were already in use in school grammars of English current in the United States in the nineteenth century. Above all, the notion of sentence "analysis," whether diagramable or not, had been applied in those grammars. That Bloomfield was at first powerfully attracted to Wilhelm Wundt's linguistic and grammatical theories must also be regarded as an important additional factor, for Wundt had endorsed a type of syntactic analysis that originated in early nineteenth-century German pedagogical grammars of the type found in the works of Karl Ferdinand Becker. Wundt even mentions Becker's Ausführliche deutsche Grammatik more than once (see, for instance, Die Sprache, 2nd part, p. 223fn. and p. 320fn.). It is possible, therefore, that in Bloomfield's theory we see a confluence of the German and the American types of syntactic analysis. Obviously, however, more research needs to be done in this area. Here, I have merely outlined some of the salient features of the development, as I see them at the present time.
As regards my second main topic, the role of the phrase in traditional grammar, the use of the term varied. As I have mentioned, the English term first came into use in the mid-nineteenth century, but some grammarians defined it narrowly, as did Allen and Greenough in their New Latin Grammar (first published in 1888). However, a fact that I failed to mention in my original article, because I was not aware of it, is that some nineteenth-century grammarians of English, such as Josiah W. Gibbs (1790-1861), were already operating in terms of word-groups. There is, for instance, a clear example of phrasal constituents in Reed and Kellogg's Higher Lessons in English, in which the expression that old wooden house is analysed thus: 'Here wooden modifies house , old modifies wooden house, and that modifies old wooden house ([New York: Clark & Maynard, 1877], p. 30, fn.). In other words, according to Reed and Kellogg, the phrase wooden house is modified by a single word old, and the single word that modifies the three-word phrase old wooden house. We must conclude therefore that not all so-called traditional grammarians failed to recognize sentence constituents consisting of more than one word. Not only did some pedagogical grammarians in the nineteenth century already divide the sentence into a subject half and a predicate half (regardless of the length of those two halves), but some of them even recognized grammatical relations within these two halves that obtain between single-word and multi-word constituents. In this way, the conceptual foundations of what was later to be called immediate-constituent analysis were laid in the nineteenth century.
In connection with phrasal terminology, it is interesting to contrast the situation in the non-English speaking world. In Saussure's influential Cours de linguistique générale, we find the term syntagme, which Saussure defined as referring to a sequence of two or more consecutive signifiants 'signifiers.' Saussure's examples include the words re-lire 're-read' and contremaître 'foreman,' which both clearly consist of two signifiants. In the manuscript sources, this principle is clearly spelled out: "Cette notion de syntagme peut s'appliquer à des unités de n'importe quelle grandeur, de n'importe quelle espèce" (Rudolf Engler, Lexique de la terminologie sausurienne [Utrecht/Antwerp: Spectrum, 1968], p. 50). On the other hand, many of Saussure's other examples of the syntagme were what we in English would call phrases (or even sentences), e.g. désireux and Que vous dit-il?. It is interesting to note, however, that after Saussure the meaning of the term seems to have been narrowed down to 'word-group.' Hence, re-lire and contremaître are now no longer syntagmes. It appears that the neologism syntagme now functions as a technical term performing pretty much the same function in French as the English word "phrase." In 1967, the French dictionary Le Petit Robert defined the term as 'a group of words forming a unit within a sentence.' (Emphasis mine.)
The earliest sign that I have ever seen that grammarians were beginning to think in terms of phrasal constituents occurs in the Grammatica nova 'New Grammar' by Bernard Perger, which seems to have been in existence by 1479. The passage in question reads as follows: "Quoniam adiectivum et suum substantivum tamquam pro una dictione in oratione habentur, ideo inter se in genere, numero, et in casu convenire debent, ut doctus vir, honesta mulier, crudele animal." Rough English translation: "Inasmuch as an adjective and its substantive are considered to stand, as it were, for a single word in a sentence, they must therefore agree with each other in gender, number, and case, e.g. 'learned [masc.] man,' 'honest [fem.] woman,' 'cruel [neuter] animal.'" Needless to say, it is difficult to be sure what is behind this statement, but it seems likely that grammarians at that time thought of a sequence consisting of an adjective and a noun as some sort of a unit. Supposedly, that fact was sufficient to explain the grammatical rule requiring the grammatical agreement of an adjective with its noun.
A considerably older strand in the grammatical tradition was the notion of dependency, on which I commented in my article "Reflections on the History of Dependency Notions in Linguistics," Historiographia Linguistica, 17 (1990), 29–47. It is interesting to note that in the twentieth century a dependency framework underlay the successful Éléments de syntaxe structurale by Lucien Tesnière (1893-1954), which was published posthumously in 1959, although the author mentions that he had already begun working on his system in 1932 (see fn. 26 in this document). Note also Tesnière's earlier brief Esquisse d'une syntaxe structurale (Paris: Klincksieck, 1953). On Tesnière, see the article about him by Kjell-Åke Forsgren in the Encyclopedia of Language & Linguistics, edited by Keith Brown et al. (Amsterdam & London: Elsevier, 2006), vol. 12, pp. 593-594.
Since this is a Web document it can be and is being constantly revised. That being so, readers should regard the present version as part of an open-ended project. If you have comments of any kind, please do not hesitate to get in touch with me. Let me at this point express my deep appreciation for having had stimulating e-mail correspondence with a number of colleagues in different countries, among whom I should like to single out for special mention Kjell-Åke Forsgren, Kenneth L. Miner, Jan Noordegraaf, Pierre Swiggers, and John Walmsley..