display | more...
I have no idea how you would define a sentence. There's no good reason to expect it to be possible to, actually. Language just doesn't break up into neat fragments: sentences, clauses, phrases, words, morphemes, phonemes, they're all convenient fictions as rough approximations to units of a certain size; but if you push them too hard they break apart in your hands.

For practical purposes I take a sentence to be defined (i.e. delimited) by beginning with a capital letter and ending with a full stop. This is absurd, from a linguistic point of view. The written form of a language should be a completely superficial, irrelevant overlay on the actual language, and should have no bearing on its grammar. Most languages have no written form; we're still left with the task of picking out something like 'sentences' in them.

We don't pause in speech at punctuation marks. Fluent language is a continuous gabble, and disfluencies are likely to occur right in the middle of constituents at the, um, heart of phrases. In the punctuation system of the modern English language we can write the same thing in four different ways (and a fifth conventionally regarded as unacceptable):

  1. "That's very annoying. Please don't do it again."
  2. "That's very annoying: please don't do it again."
  3. "That's very annoying; please don't do it again."
  4. "That's very annoying, please don't do it again."
  5. "That's very annoying please don't do it again."
For reasons mysterious to me, (1) is considered to be two sentences, while the rest are considered to be one. The dots don't change the grammar, the structure, but do invoke a more or less arbitrary convention on how we divide text up. Okay, I haven't got any better ideas. But in the past most written languages didn't have these subtleties. These days we consider (5) totally unacceptable, we call (4) a comma splice and try to stem the flood (even I squirm at it, except in short colloquial conjunctions like this), and some people even invoke imaginary rules about when (2) is acceptable. But they're just dots. We don't actually make discrete and distinct grades of pause or intonation corresponding to different weights of connexion. We can say the two halves faster or slower, with more or less pause, but it's all continuous. Nowhere does the 'sentence' emerge.

And that's the preliminary complaints done. Now let's look at some structural and semantic matters.

Subject and predicate

While there's no definition that will cover all sentences, we can talk about a canonical sentence type. That is, a typical one that covers most examples, and from which non-canonical types can be derived. In this sense, the essentials of a sentence are a subject and a finite verb. A finite verb is a verb that is inflected for tense, person, and number. (I have to restrict myself to English and related languages for now, as some languages have no such inflection at all, but in principle it's very widely applicable.) In "John comes here" the "-s" indicates third person, singular, present tense. English hasn't got much inflection but it does show this, and this enables us to distinguish it from "for John to come here", where the form of "come" has no such inflections.

Look at any of the sentences (capital letter... full stop) in a written text, and virtually all of them contain a subject and an inflected verb agreeing with the subject. Most, of course, contain other material as well. Traditionally the sentence is divided between subject and 'predicate', which would be a useless term if it merely meant "anything that isn't subject". But there is good evidence that most of the syntax of language is binary. An element of a sentence is divisible into two other elements. Sometimes one is empty. But in principle you can start with a sentence and break it up into bifurcations of increasingly smaller elements until you get down to the word (a shadowy concept, like all else, but let's for the sake of argument just use the written spaces in English to determine 'words').

There is in fact good evidence that the meat of a sentence consists of a subject and a verb phrase, that is a complex structure containing a verb and possibly some other material. The traditional notion 'predicate' can either be the same as 'verb phrase' or it can mean the other stuff beyond the verb. I really don't care which: it's just terminology. Normally the subject is a noun phrase, that is a noun together with possibly some other stuff dependent on it.

Here we have to move away from 'traditional' definitions (which aren't worth much: burn your Strunk and White or equivalent, it's largely ignorant rubbish) and accept the results of scientific study in the last fifty-odd years. Linguistic entities are recursive. The most general concept is probably the 'phrase'. This is a word of some class (the 'head') together with some dependent words, and the whole group behaves in some grammatical way like the head. For example, 'this black cat' is an elaboration of 'cat' (but not of 'black'), and 'ate a chocolate cake on Tuesday' is an elaboration of 'ate' (but not of 'cake' or of 'on').

So what does this mean for the sentence? Normally the subject is a noun phrase and the predicate is a verb phrase. Minimally, it can be a noun and a verb ("John ate"). Usually the noun phrase is a noun plus dependent stuff ("my friend John") and likewise with the verb phrase.


Language is recursive. Pointed out by Noam Chomsky, this is fundamental to human language. Noun phrases can contain other noun phrases ("my friend John's younger sister"), verb phrases can contain other verb phrases ("ate the sandwich I was making for her brother"). There is no real limit to the recursion (except in particular places: see central embedding for a curious exception).

So one verb phrase can contain various extra bits (we call them complements if they're essential and adjuncts or modifiers if they're omissible without gross loss of meaning), of which some are other verb phrases... or indeed, some of which are other things which themselves contain verb phrases. So we get complex sentences like "My friend John ate the sandwich which I had made for my sister Mary to enjoy when she got back from the seaside where she had been camping after her new job fell through." Inside this there are numerous subject/verb pairs ("I... had made", "my sister Mary... enjoy", "she... got", "she... had been camping", "her new job... fell"). Generally these are called subordinate in traditional grammar. One subject/predicate pair at the top of the recursive hierarchy is the main clause, and all others are subordinate clauses.


Some complex constructions aren't amenable to that analysis. "I went home, and she was waiting for me" is quite similar to "I went home because she was waiting for me". Traditional grammar says that 'because' is subordinating but 'and' isn't -- it's coordinating, it's linking two equal clauses.

Have I even mentioned clauses yet? A clause is the basic subject/predicate pair. It's the thing that can be stacked up repeatedly, coordinated, subordinated: a clause is real. I don't know that a 'sentence' is, defined somehow as a top-of-the-tree clause that might have inferior clauses inside it but doesn't have anything above it.

There is justification for the distinction between coordination and subordination. A subordinate clause is part of the higher-level element it's attached to: e.g. "the brown book I was waiting for" is a kind of book, and "brown" and "that I was waiting for" are qualifiers. In "I went home and she was waiting for me" neither clause is inside the other. That is, subordinate clause fits with the general X-Bar Theory of phrases based on one head and a number of dependents attached to it. However, in a coordinated structure no one of the coordinates is in a head role (of the coordination, though the whole coordination may be the head of some other phrase).


A further method by which material is included in a sentence may be called supplementation. In this, the material is just there, attached linearly at a convenient point, and semantically related to its point of attachment, but not grammatically integrated into it. It forms neither a head/dependency structure nor a coordination. Such material is set off by some kind of pause intonation, and in writing by punctuation such as dashes, brackets, or some uses of commas.
  • Undeterred, Michael continued to explore the corridors.
  • He explained -- not that he needed to -- that he was unafraid.
  • The caretaker, a Mr Blenkinsop, nodded gravely.

Fragments and completeness

One commonly touted idea is that a 'complete' sentence has subject and inflected verb. Well. Well, I don't know what to say to that, but it's begging the question on a massive scale. Virtually any constituent of a sentence can occur on its own as a complete utterance. You ask me a question, I answer it by saying "Yes" or "On Wednesday" or "She did" or "Under the bed" or "As usual" or "Because I wanted to" or "For Mary to eat". And it doesn't have to be a constituent even: there's no way that "Reverend Green, with the candlestick, in the billiard room" occupies any node in the hierarchical description of any possible longer sentence, but it's a complex and articulate utterance in plain English. I know of no reason why it shouldn't be called a 'sentence'.

In many languages (Italian, Spanish, Chinese, Japanese, to name but a few) you routinely don't need a subject at all. In English this is much rarer, but is true of diary style ("Woke up. Went to work.") and the imperative ("Come here."). The inflected verb requirement is a bit different: subordinate clauses can have non-inflected or more accurately non-finite verbs ("by eating", "to eat", "for John to eat"), but virtually always there is, must be, some top-level main-clause with a finite verb. One unique exception in English is colloquial but pretty clearly standard grammar: "better" as an alternative to "had better" (as in "You better go now"). A small class of exceptions is the ragbag of frozen optative main clauses, such as "So be it", "Come what may", "God save/bless ...", where the non-finite plain form is used. (Frozen means they have to be learned individually like words: the grammar doesn't allow you to do this with just any verb).

Sometimes the problem is moved out to semantics. A sentence is described as the expression of a "complete thought". I can't make any sense of the words "complete thought". Do you start thinking new thoughts, then stop when they're finished? If you do, does it correspond to the bits between a capital letter and a full stop? If you think it does, can you provide any evidence for this?

Fragmentary sentences don't just occur in answers. ("Where did you leave the keys?" -- "On the dresser.") We can begin a discourse by all sorts of fragments out of the blue. "Silly idea," we say, after a long silence, watching someone trying to do something. Or "Oh, ugly house!", or "Tuesday!", or "Now...", or "What a fool!".