Immediate Constituent Analysis: History, Trees, and a Short How-to

An immediate constituent is one of the parts you get from a single cut of a construction. The parts you would reach by continuing those cuts are smaller constituents. In Leonard Bloomfield’s terms, the last cuts land on ultimate constituents. For a sentence, those are often the words; in a morphology problem they may be morphemes.

The first cut is usually the largest grouping that still behaves as a unit. In The old man walked slowly, a standard first cut is the subject string and the predicate string:

• [The old man] [walked slowly]

Each of those strings is a construction of its own. The old man cuts into a determiner and a noun phrase headed by man, and old combines with man before the determiner does. Walked slowly cuts into the verb and the adverb. The diagram under this section is that bracketing, written as nested cuts rather than as a story about parts of speech.

Two checks keep the cut honest. A constituent should be replaceable by a shorter form of the same kind (he for the old man, did so for a verb phrase in a larger sentence). It should also coordinate with a string of the same kind. If a proposed group fails both checks, it is probably not a constituent, even if the words sit next to each other.

IC analysis is a procedure for constituency. It is not a dependency diagram, and it is not a school exercise in labeling every word’s part of speech on a baseline. The nodes in the trees further down are constituents.

Immediate constituent analysis is a way of cutting a construction into the pieces that join at one step. The name is methodological. “Immediate” means the parts of this cut, not the words you eventually reach. A sentence can be one construction; each of those parts can be a construction again. The record of the cuts is a bracketed string or a tree.

That procedure became a standard move in twentieth-century American descriptive linguistics. It is easy, and misleading, to treat it as school grammar with new labels. School parsing also divides a sentence, often into a subject and a predicate and then into word classes. The aims differ. A traditional parse classifies words and assigns grammatical roles for a reader who already knows the language. An IC analysis is a successive division, justified by whether the pieces behave as units. Historians of linguistics have argued that the twentieth-century procedure is not a straight renaming of the older one. The essay that makes that case at length, and that reading lists still cite, is W. Keith Percival’s On the Historical Source of Immediate Constituent Analysis.

Percival presented a version of the paper at the University of Wisconsin–Madison in 1967. It was published in 1976 in Notes from the Linguistic Underground, volume 7 of the Syntax and Semantics series, edited by James D. McCawley (Academic Press). For years an HTML copy sat on a University of Kansas personal page (people.ku.edu/~percival/ICanalysis.html). That host no longer serves the essay. The Internet Archive holds snapshots, including one from 27 January 2019 linked in Sources. This page does not reproduce that essay, its footnotes, or the publisher’s text. It states the bibliographic facts, the historical question, and a drawing method.

The historical question is about sources, not about who first drew a tree. Branching diagrams for language appear in late nineteenth-century psychology. Wilhelm Wundt, in the language chapters of his folk psychology, drew sentences as wholes that split into parts and then into smaller parts. Those drawings matter because they show division of a whole, not a list of word classes. They are not yet the distributional procedure of American structuralism, and they should not be dropped into a modern handout as if the labels matched.

Leonard Bloomfield’s Language (1933) is the book in which constituency becomes a working method for American linguistics. Bloomfield describes forms as composed of constituents, and he uses the immediate / ultimate distinction: the immediate constituents are the parts of this construction; the ultimate constituents are the smallest forms you reach if you keep cutting. The point of the method is to avoid a cut that ignores how the language groups the string. A noun and the adjective that modifies it can be one constituent even when a word-class list would set them side by side as two labels.

Rulon Wells’s 1947 article “Immediate Constituents,” in Language, is the paper most courses mean when they say “the IC article.” Wells asked for cuts that keep the number of constituents small at each step, that prefer patterns attested elsewhere, and that respect substitutability. Later textbooks turned related ideas into classroom routines: pronoun replacement, coordination, fronting. Those routines are descendants. They are not a transcript of the article.

Through the 1940s and 1950s the method spread in descriptive and teaching work associated with Zellig Harris, Charles Hockett, Eugene Nida, and others. Harris pushed a distributional program: classify elements by what they co-occur with. Hockett’s textbooks presented successive cuts to a broad student audience. Nida’s morphology work used similar cuts below the word. This was not one diagram style. What these authors share is a hierarchy of constituents, justified by distribution rather than by a prior grid of Latin cases.

Phrase-structure rules, from the mid-1950s, keep the hierarchy and change the job. A rule such as S → NP VP claims that a sentence has those two immediate constituents, written as a rewriting step rather than only as a cut you discovered in a text. Noam Chomsky’s Syntactic Structures (1957) is the usual citation for that shift. Later X-bar trees and Minimalist trees still draw constituents. They add bar levels, functional heads, and movement. An IC tree and an X-bar tree of the same sentence can agree on the first cut and disagree on every label under it. That is why the next section draws both.

The relation to older grammar is the point a history summary has to keep. Renaissance and school grammars divide sentences, and some of them draw structure. Division alone does not make them IC analysis. IC analysis, in the structuralist sense, is a procedure with an account of what counts as a constituent. Treating it as “the old parse, continued” hides the break. Cite Percival for that historical argument, and use the archive when you need his wording. Use this page when you need the question, the names, and a tree you can redraw.

A compact reading order for a seminar: Wundt for early branching diagrams; Bloomfield 1933 for immediate and ultimate constituents; Wells 1947 for the classic statement of the cut; a structuralist textbook for the classroom routine; Chomsky 1957 for phrase-structure rules; then a modern constituency tree, labeled as a picture of constituents rather than as a substitute for the essay.

Both drawings below use The old man walked slowly. Every node is a constituent. The words stay in order along the bottom.

The traditional tree makes one NP and one VP under S. Inside the NP, determiner, adjective, and noun are sisters. That is a flat phrase-structure analysis of the sort many introductory courses still draw. It matches the brackets in the first section: [The old man] and [walked slowly], then a further cut inside each.

The second tree keeps that first cut and adds X-bar shape. Old sits under an N′ with man, so the adjective and the noun form a constituent before the determiner joins. The adverb projects an AdvP. Intermediate courses often want exactly this extra layer; others ban it. The picture shows the difference. It does not settle your course’s rule.

These trees were typeset for this guide. They are not output copied from a journal figure. If you want the sentence redrawn in Traditional, X-bar, or Minimalist labels — including DP and TP if that is your framework — use the syntax tree generator. Add a note such as “use DP” when the menu’s defaults are not your course’s defaults. A second pass with “X-bar, adjective under N-bar” should move old off the flat NP.

Work on paper first. The generator is for the drawing, after you have decided the cuts.

1. Read the whole string as one construction.
2. Make the first cut into the immediate constituents: the largest pieces that combine at this step. For a simple clause, that is often a noun-phrase subject and a verb-phrase predicate.
3. Repeat inside each piece. Stop when the next cut would only list words you are treating as ultimate constituents.
4. Test the groups you are unsure about. Can a pro-form stand in for the string? Can you coordinate it with another string of the same type? If the group is an adverb or a prepositional phrase, say what it modifies before you attach it.
5. Write brackets or draw a tree. Read the leaves left to right. They should spell the sentence you started with.

Practice sentence: Those noisy students left early.

A defensible analysis, and not the only one a course might want:

• [Those noisy students] [left early]
• [Those] [noisy students]
• [noisy] [students]
• [left] [early]

Early is a sister of left if it modifies the verb. If your instructor wants an AdvP, say so when you draw. Nothing in the sentence forces a relative clause, a trace, or a second reading.

When the cuts are stable, draw the sentence as a syntax tree. Compare the generated attachment with yours. Where they differ, run the tests again before you change your mind.

A modern constituency tree is the picture IC analysis was reaching for: a hierarchy of groups, not a line of word classes. The intuition carries over. The labels often do not. Traditional trees use S, NP, and VP. X-bar trees add bar levels. Minimalist trees use CP, TP, and DP and mark movement with traces. Calling all of them “IC trees” without saying which label set you mean will confuse a reader who has Wells open next to a current textbook.

Use this guide for the history and the first cuts. Use the tools for the drawing:

• Syntax tree generator — type the sentence, pick Traditional, X-bar, or Minimalist, and download a tree.
• Syntax trees for linguistics students — frameworks, what to check, and how a semester of trees is priced.

If your question is the history of the method, stay on this page and then open Percival in the archive. If your question is how to label a tree for a course, open the generator and name the framework.

This page is a Vizcept rewrite, not a reprint. It does not reproduce W. Keith Percival’s essay, the footnotes, or the text published by Academic Press (now under Elsevier’s academic list). It does not present itself as an official electronic edition.

Essay to cite for the historical argument

Percival, W. Keith. “On the Historical Source of Immediate Constituent Analysis.” In Notes from the Linguistic Underground (Syntax and Semantics, vol. 7), edited by James D. McCawley, 229–242. New York: Academic Press, 1976. A version was presented in Madison in 1967. Bibliographic details should be checked against the volume you hold; this page is not that volume.

Archived HTML copy (the University of Kansas URL no longer serves the file):

Wayback Machine, 27 January 2019, people.ku.edu/~percival/ICanalysis.html

Other public landmarks named above, for a seminar shelf rather than as excerpts: Bloomfield, Language (1933); Wells, “Immediate Constituents,” Language 23 (1947); Chomsky, Syntactic Structures (1957).

View the archived original when you need Percival’s wording. Use the trees and the five steps on this page when you need a figure.