I am interested in understanding these structures work. My question is: Why do some diagrams start with S, while others begin with TP, IP, etc... Is there a way to stick these all of these phrases under S? Please use simple feedback, I haven't taken any linguistic courses in 10 years.

  • In Joseph Emonds' theory about structure preservation, there are root sentences with distinctive syntactic behavior, such as (for English) those containing inverted finite auxiliaries and certain performative adverbs. In that case, choosing S as the symbol for root is more a matter of fact than it is a matter of theory.
    – Greg Lee
    Jan 22, 2017 at 19:45

3 Answers 3


Edit: This answer focusses on mainstream generative grammar approaches which are heaily based on English (or Indo-European languages') syntax, assuming that this is what you are primarily interested in. However, as WavesWashSands pointed out in their comment, things might look completely differently for other languages where the assumption of something like a category I is of not much avail and S might indeed be approriate.

S as the root of a sentence is a rather old assumption that is rejected my most modern linguists and is only used for simplification when a detailled analysis of what a sentence constitutes is not relevant (e.g. when explaining how phrase structure grammars work in general, or in very simple "toy grammars" to demonstrate parsing algorithms in computer science/NLP).
When not immediately related to natural language syntax, but for the description of formal languages (such as {anbn : n ∈ ℕ}, i.e. the language consisting of all possible sequences of aaaaa...bbbbb... where there are just as many a's as there are b's) with phrase structure grammars, which has its use mostly in computer science and mathematics, S is still the convention for the root label, but usually thought of as an abbreviation for "start symbol" rather than "sentence".

The reason why S is not so nice is that it doesn't really give any information of what it is built from, and doesn't fit the general scheme of constituent trees.
In X-bar-theory, variants of which pretty much any major syntax theory is based on, you want to build up your trees systematically starting from lexical items (words) below and making constituents larger and larger, going over to phrases until everything is combined to one maximal structure, thereby assigning every node a combination of a category (such as nominal, verbal, prepositional, inflectional, ..) and a so-called projection level (roughly, it tells us how complete our tree is up to this point: is it a single word, a phrase (= a group of words which somehow belong together) or something in between) encoded in the label name.
Labelling the root of your syntax tree with an S doesn't fit into this scheme now, because you do have "S" as an abbreviation for "sentence" and thus a kind of category, but it doesn't tell us anything about the mentioned projection level and isn't build up of anything S-like either (a CP (= complementizer phrase) has a complementizer in it, an IP (= inflectional phrase) contains an inflection node, ...); for all the other constituents further below in your tree, you have a so-called head which determines the category of the phrase, dependents which complement the head and so on, but the S is just standing there as a single S-thing and branches into, e.g., some NP and VP which violates the assumption that sentence structures are built up in sysetmatic and preditable patterns.
TP, IP, CP etc. are different here; they do fit the widely accepted X-bar scheme and give us the information we want, namely of what category a sentence is (something tensed/inflected/...) and the projection level (namely a complete phrase that can stand on its own). That's the reason why most modern syntax theories rely on these labels as the top nodes of sentences, because they don't break the general explanatory pattern of how constituent trees are built up.
Additionally, IP and TP are labelled so because they are headed by a category "inflection"/"tense" which makes sure that the sentence is properly inflected w.r.t. tense, subject-verb agreement and so on, and it is assumed that this happens at a stage higher up in the hierarchy after all verbs have started off as uninflected. If you don't include an I or T in your tree, but simply say that an NP and a VP combine to an S, the question is where the inflection comes from, because it must be reflected in the tree somewhere.
So, no, you can't simply put everything under S without violating basic assumptions shared by major syntax trends. But you can, if you want, make the generalization for yourself that "CP", "IP" etc. are more or less equivalent ot "sentence".


  • old syntax theories or dummy grammars without a need for linguistic adequacy --> S
  • more recent syntax theories that in some way make use of X-bar theory and want to describe syntax in a more accurate and more generic way --> TP/IP/CP/...

Edit: Two more comments:

  1. You write in your title "the head label of trees". The term "head" is not quite right in this context; it usually refers to the head of a phrase as assumed in X-bar theory, but apart from that it is not a term that is defined for trees as mathematical objects independently of a syntactic theory. What you mean is the root of the tree.
  2. If you choose to label your root with IP or similar, you can of course not just go and replace S by IP, but as soon as you decide to stick to consitent X-bar labelling and an account for inflection (which would be the motiviations for chosing IP, or TP or whatever), this labelling must be licensed, i.e., the category I must occur in your tree as the head of the inflectional phrase, so you'd need to restructure your whole tree accordingly. A tree of the form [IP [NP VP]] is not better (in fact even worse) than a tree [S [NP VP]] if your IP isn't made up of anything that involves the category I, similarly with T and C; so you actually get different trees depending on the choice of the root label.
  • Thank you. Now I have a better understanding of this stuff. When I was in school everything was placed under the S. However, your explanation makes more sense. Dec 7, 2016 at 0:16
  • @Rachel Sullivan I've edited two more comments into my answer. Dec 7, 2016 at 14:49
  • 1
    I respectfully disagree with the notion that S is used in old syntax theories without a need for linguistic adequacy. That may indeed be the case for English, but there are many languages - most famously the Australian languages - where constituency is expressed by word order to a limited extent, or even not at all. In such cases it would seem preferable not to force the IP, which originates from the grammar of English and Romance languages, onto them. Dec 8, 2016 at 2:18
  • In addition, it is possible for 'slightly more configurational' languages to have both IP and S. Austin and Bresnan (1996) analysed Warlpiri as having both IP and S, and Jiwarli as simply having an S. Dec 8, 2016 at 2:18
  • @WavesWashSands Thanks for the feedback - I must admit that I have no idea of syntax when it comes to data other than IE langugages; my post was based on the mainstream generative approaches which heavily focus on English. I can't elaborate much on where S might indeed be useful, but I'll include for clarification in my post that the answer I've given mostly applies to English. Dec 8, 2016 at 9:12

For one thing, modern grammar has been strongly influenced by the study of logic and the foundations of mathematics. In the early 20th century, paradoxes were discovered, for which the most popular resolution was to prevent the paradoxical propositions from being expressed in a rigorous, logical language. To resolve the paradoxes in this way requires an explicit and restrictive grammar to be devised which prohibits the expression of the troublesome paradoxes.

In logic and mathematics, linguistic interest is focused on theorems and other statements corresponding most closely in natural language, probably, to simple declarative statements. For this type of language expression, linguists generally use "S". It corresponds to "well formed formula", or wff, in some presentations of logic. There is not a great deal of uniformity among logicians, but sometimes this part of logic is called "morphology", rather than the linguists' term "grammar".

So that is one place "S" comes from. A related usage is in applications of CFG, context free phrase structure grammar, where by general convention, when a grammar has a single initial non-terminal symbol, one calls it "S". It's just a shorthand to avoid tedious explanation.

If the intent of a CFG or other form of generative grammar is to generate strings for language expressions other than simple declaratives, such as questions, imperatives, or other expression types recognized in traditional grammar, then the sense of "S" is usually extended to those other types. Then "S" means, in effect, complete expression generated by a grammar.

In CFG, the names of the non-terminal symbols, including "S", are of no intrinsic significance. This probably extends to transformational grammar and other theories, as well. If for stylistic reasons, you prefer other names for initial symbols, you should feel free to use them. What's in a name?


The use of the symbol ‘S’, standing for ‘sentence’ - originally, since Roman times, ‘a coherent phrase expressing a complete ‘thought’ or ‘proposition’’ - has largely been discontinued in technical accounts of syntactic structure written within Chomskian generative grammar because, at one point in its development, 'S', and its associated rule, stopped satisfying the requirements of a new theory that Chomsky considered it necessary to adopt in order to explain, instead of merely describe, the syntactic constructions of English and Human Language in general.

The long story is really long and complicated, but the gist thereof is about this:

Whereas early Chomskian analyses of S were trivially transparent 'symbolizations' of the long established view of the 'sentence' as a combination of a 'subject' and a 'predicate' (categorially: an NP and a VP, respectively), by the late 1960's Chomsky had realized that (context-free/sensitive) 'phrase structure rules' of the form X > Y (for Y = any non-null sequence of symbols) could describe, and therefore predicted the existence of, just about any sequence of syntactic elements. As that included infinite sequences never attested in any human language, such rules were obviously too powerful and failed to offer satisfactory explanations of what was or was not possible in human syntax, and Chomsky decided to first constrain their Y side and then replace them altogether with a set of simple principles meant to predict and formally explain what was actually found in the syntax of human languages.

As a result, in 'Remarks on Nominalization' (1970) he proposed the 'X-Bar Theory' of syntax, a set of ‘principles’ (not called ‘rules’ anymore) according to which all syntactic constructions must be 'headed' (= 'endocentric', in Bloomfield's jargon), with additional constraints on what the non-head could and could not be and on how headed structures can recur inside other headed structures that we can ignore for current purposes.

According to the X-Bar 'principles', then, any phrase of category XP is ultimately a (possibly recursive) 'expansion' of a head category X - where X ranges over the set of minimal syntactic categories, which, at the time, largely reduced to the 'parts of speech' already established by traditional grammarians in the preceding two millennia, i.e., N, V, A, Adv, P, Art, etc. Thus, an NP was ultimately an 'expansion' of N, a VP an expansion of V, an AP an expansion of A, and so on.

The ‘endocentricity’ idea, of course, was not new (Bloomfieldian doctrine aside, it is occasionally found much earlier, e.g. in the treatment of adjectival modification in John Wallis’ Grammatica Linguae Anglicanae, published in 1653), but X-Bar Theory as a whole and its ambitious intended scope were new, and very important insights bound to stay in Chomskian linguistics for more than thirty years.

X-bar theory worked reasonably well for core categories like NP/N, VP/V, or AP/A, but ran into serious obstacles with coordinate constructions, and, above all, with the syntactic structures par excellence, i.e., 'sentences': obviously, under the early TGG analysis of 'S' expressed by the popular rule S > NP + VP, S was non-headed ('exocentric'), and a flagrant exception to the XP = [...X...] pattern that X-Bar Theory predicted all syntactic constructions should comply with.

Therefore, if X-Bar Theory was to stand, as desired, it was necessary to find a way to reanalyse the 'sentence' as a 'headed' construction, and, since S consisted of only two 'immediate constituents', the NP (functioning as the 'subject') and the VP (functioning as the 'predicate'), only two possibilities seemed available: either S was an expansion of the NP (and ultimately of N, according to the X-Bar principles), or the head of S was the VP and S was an expansion thereof, and ultimately of the finite verb.

The former view (although defended, in the 1920’s and 30’s by no less a grammarian than Jespersen) was considered absolutely untenable for many syntactic and semantic reasons I cannot review here. There, thus, remained the latter option, i.e., that S was simply an expansion of the VP (and, ultimately, of the finite verb).

In spite of certain semantic obstacles (VP, the predicate, names a ‘function’ in Frege’s sense, whereas S, like NP, names an ‘object’ = 1/0, again in Frege’s terms), that the finite verb was the 'heart' of the sentence had been the preferred view in the preceding two and a half centuries of ‘traditional grammar’ and remained so among late 19th century German linguists and early 20th century pre-Chomskian European 'structuralist' and 'functionalist' grammarians like Guillaume, Tesniére, etc.

Unsurprisingly, therefore, between 1970 and 1979, TG grammarians, too, seriously considered adopting the V-centered view of the sentence as the obvious way to make S comply with X-Bar Theory, and, as a matter of fact, one of the most important monographs on X-Bar Theory ever published, Jackendoff's X' Syntax (MIT Press, 1977), did analyse S as an extended VP phrase (V''', in his notation). The fact that in many languages the subject seemed dispensable, whereas the finite verb was not, seemed to support that traditional view, and until about 1979 several other influential Chomskian linguists proposed similar, V-centered, analyses of the sentence. Or rather, of the ‘core sentence’ (= the 'predication'), since, as early as 1972, Joan Bresnan had already offered compelling reasons to analyze real speakers’ sentences as structures containing S preceded by a ‘complementizer’ category C lodging certain subordinators and expressing the sentence’s crucial ‘illocutionary’ properties (assertion, interrogation, command, etc.), a view that, under X-Bar theory, first suggested treating sentences, rather sloppily, as ‘S-bar’ phrases with the internal structure [Comp + S] (e.g., in Chomsky’s Lectures on Government and Binding 1981), or as Comp + V’’’ structures (in Jackendoff’s earlier work), and eventually as CPs, i.e., as projections of the C category (e.g., in Chomsky’s later, and extremely influential book Barriers, 1986).

That is the source of the sentential tree diagrams dominated by CP that figure in most introductory syntax textbooks even nowadays. However, as Comp is invisible in independent declarative sentences, and those are the obvious sentential structures to start explaining syntax with to beginners, such textbooks often omit representing the CP segment of the projection and offer informal sentence trees dominated by the old symbol S or by the categories that eventually replaced it under the pressure of the X-Bar principles, either an uncompromising IP, as in Chomsky’s Barriers, or the more informative, and controversial, symbols (TP, AGR-P, etc.) that eventually replaced IP.

To justify the original IP idea a bit more, though, by 1980 the study of verbless predications in English and other languages had progressed enough to lead Chomsky to claim that, at the UG level, the grammatically key element of a 'predication' was neither the subject nor the predicate, but something more abstract connecting those two elements, a view that logicians and philosophers of language could not but sympathise with (due to the existence of the 'predication paradox', a problem we need not go into here). Interestingly, in the Port Royal Grammar, that 'linking' element had been claimed to be the 'copula' be, but that ‘logicist’ analysis was arbitrary, at bottom, and never caught on in mainstream grammatical theory, so, for a long time, no alternatives were proposed. When the study of ‘universal grammar’ was seriously resumed by Chomsky and his disciples, though, the problem of the ‘linking’ category re-emerged, and in certain cross-linguistic TGG analyses published between 1979 and 1981, the possibility was seriously contemplated that the crucial element of 'predication' might be the abstract 'Aux' node that Chomsky himself had introduced in his earliest work on TGG (notably, in Aspects, but, in a less structured way, already in his 1955 LSLT thesis). As, within 'Aux', (English) auxiliaries all seemed optional, the key ‘linking’ component could only be 'tense' ('c', in Chomsky's earliest work), which, at the time, pointed to tense as a likely candidate to sentence-head status.

That hypothesis, however, raised the obvious problem that many predications were non-finite (or even non-verbal) and could hardly be analysed as 'projections' of a syntactic element they did not have. The solution that Chomsky eventually adopted from his 1979 Pisa lectures (later Lectures on Government and Binding, 1981) onwards within the framework of the then new 'GB' theory was to analyse sentences as 'projections' of an abstract 'functional' head INFL instantiated in different ways, i.e., mainly as a tense morpheme (in finite clauses) or as 'zero' or other 'verbal' inflections (in non-finite ones), but also as other 'functional' morphemes (e.g., certain preposition-like words, etc.) in other constructions attested in diverse languages. That much will hopefully suffice as to the origin of the IP analysis of the predicational core of sentences.

Analysing S as a projection of Infl according to the 'rules' IP/I''> NP + I' and I'> I + VP (alternating with other 'predicates' like PP, AP, NP, etc.) solved the 'endocentricity' issue that sentential constructions had initially raised, kept and reinforced the X-Bar principles as one of the key 'modules' of the ensuing 'Principles and Parameters Theory' of the mid 1980's, and, at the same time, allowed for great flexibility in analysing predications, which is why many introductory textbooks of the 1980's and early 1990's adopted the IP analysis of the core sentence, if only as a pedagogical solution.

However, as P&P theory developed during the mid and late 1980's and the number of languages analysed within that framework increased, INFL inevitably revealed itself as a simplification, a cover term for an aggregate of different inflectional features, in particular, 'tense' (or lack thereof) and 'agreement', and, as such features could be neatly distinguished in 'phonetic form' in many languages, and could even appear separately in some, it soon became inevitable to insert them in syntactic trees under different category nodes. By 1988, largely due to Mark Baker’s influential work on the correspondence between verb morphology and syntactic projection, the Tense and AGR components of Infl were already distinct 'nodes' in all serious work on P&P syntax, and, to the extent that, in view of the morphological structure of verbs, AGR-S arguably is structurally higher than Tense, later P&P work by Chomsky and others between 1989 and 1991 started including analyses of the predicational core of the sentence as an AGR-sP(hrase), although eventually Chomsky himself retrenched into a simpler view that reduced IP to TP, for technical reasons that need not concern us here.

That was only the beginning of the ‘split-INFL’ analysis, though. At about the same time or soon afterwards, evidence was offered that certain languages had not only subject, but also object agreement, and, for parallel reasons, the Infl 'segment' of the sentential projection quickly gained an 'object-AGR(eement)' node. Also, from about 1989 on, the sheer internal logic of late P&P Theory gradually led to the establishment of many other such 'functional' heads inside the 'core sentence', both within the 'INFL'/AUX area (e.g., 'polarity' and several 'modality'-related heads) and above it, in the area traditionally associated with Joan Bresnan's (1972) Complementizer and the expression of a sentence's 'illocutionary' features.

In Chomsky’s writings on GB and P&PT, e.g., in Barriers (1986), Comp was still a single head projecting a CP than contained IP as a complement, but, in work by Rizzi, Cinque and others about a decade later, the original Comp node had itself split into a C in the strict sense, plus, perhaps, an additional Force head (with its specifier), a Topic ‘shell’ (= a Top head plus a suitable specifier), a Focus ‘shell’ (again, a Foc head with its own specifier), and, in Cinque’s influential monograph on Adverbs and Functional Heads (1999), several attitudinal and epistemic adverbial heads and specifiers. [Parallel developments took place in the analysis of NP-structure starting with Abney’s (1987) influential 'DP Hypothesis' soon followed by very delicate DP internal structure developed in Cinque’s cartography programme, among others].

By the late 1990’s, therefore, CP was clearly just a convenient cover term for a number of hierarchically ordered syntactic projections - including ForceP, TopP, FocusP, etc. - that syntax textbooks chose to specify or not depending on their pedagogical policies and on their authors' expository strategy in each textbook section, which explains why the syntax beginner may still easily encounter apparently conflicting analyses of ‘the sentence’ - and, of course, different symbols, like CP, ForceP, IP, TP, etc. - at the top of sentence trees even in contiguous sections of the same textbook.

The story of the structure assumed for what we must now informally call 'the sentence' since the early 1990’s, in sum, is so complicated that it is virtually impossible to summarise, particularly in non-technical terms, but, in response to the OP's specific question, I would like to add, by way of summary and conclusion, that the increasingly detailed hierarchical structure that research has revealed within what we used to call the ‘sentence' has converted that term into an informal, pre-theoretical, label, useful enough to the extent it invokes the traditional idea of a ‘sententia’ (= a ‘complete’ and coherent expression that ‘makes sense’) and the layman’s common sense view thereof, but with no real significance in contemporary syntactic analysis of a non-pedagogical nature.

That, of course, is a problem for beginning syntax students, who must remain alert to detect when their textbook is just omitting irrelevant structure its author accepts and assumes to be there, if hidden, and when it is opting for an alternative analysis that denies the existence of that extra structure, but, in case the OP feels she is being irresponsibly played with, let me quickly add that the possible uncertainties this state of affairs may induce are far from mere ‘collateral damages’ caused by a teacher’s or a textbook author’s pedagogical strategy.

The dire fact is that, at the time being, even professional syntacticians do not really agree as yet as to the fine details of ‘sentential’ structure, including such key aspects thereof as the number, labels, and hierarchical disposition of sub-sentential phrases, the syntactic and semantic properties associated to each of the by now innumerable heads and specifiers postulated, or crucial semantic issues like what level(s) within current ‘sentence’ structure are ‘propositional’ (the ‘saturated’ VP?, the higher vP node?, PolarityP?, TP?, Agr-S-P?, CP?...), to name but a few and avoid going into the movement-related issues that follow from such uncertainties.

Assuming that a syntax textbook author is competent and well informed, in sum, what label he chooses to use at the root of a ‘sentential’ tree is often just a matter of pedagogical convenience and need not worry learners once they are aware of that trivial fact. What must trouble both learners and syntax teachers or professional syntacticians, but particularly the latter, is that too much of the highly articulated hierarchical structure currently assumed for sentential phrases has never yet been properly substantiated in either syntactic or semantic terms. That, of course, makes life very hard for those of us who teach the same students full-length courses (as opposed to mere lists of selected 'topics') in both syntax and semantics and like to start our teaching by promising them not to consider any structure or machinery that cannot be set in a transparent, compositional, relation with empirically testable aspects of the meaning of sentences and their parts. At present, such promissory notes are, to the best of my understanding, impossible to redeem unless one retreats into a much simpler view of sentence structure than is currently assumed in professional journals, and, to that extent, the decision of many textbook authors' to simplify things and analyse sentences in terms of just CPs and IPs, for example, is understandable.

  • You resumably are referring to "Remarks in Nominalization", but 1967 is a bit early for that as far as I know.
    – user6726
    Jan 22, 2017 at 14:54
  • You are right. I should have referred to the first published version in Jacobs & Rosenbaum's 'Readings...', which appeared in 1970, but, as is often the case with Chomsky's work, the first version of that article was circulating at MIT much earlier, and I recall a personal conversation with Chomsky in 1986 in which, if my memory does not fail me, he mentioned that the actual idea of replacing PSRs with a set of (future X-bar) principles dates back to 1967. I will correct the text at that point, though, since most people cannot know that. Thanks!
    – user6814
    Jan 22, 2017 at 18:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.