“Clause” and “sentence” are two terms that linguists use all the time, but they have a hard time explaining what they mean. I recently posted a question about this on Facebook, and my feeling was confirmed that there is a lot of uncertainty about these two terms. They are almost never discussed, but I think that it’s worth attempting to use precise terminology in linguistics, so here are some thoughts and proposals (also about the term main clause, which also causes confusion).
First of all, here is my proposal for the definition of the term sentence:
(1) A sentence is a maximal clause, i.e. a clause that is not part of another clause.
This definition relies on the notion of clause, and it is less straightforward to give a definition of clause that makes most people happy, i.e. that conforms to our intuitions. Here are two possibilities:
(i) A clause is a combination of a predicate (full verb or nonverbal predicate) and its arguments, plus modifiers.
(ii) A clause is a syntactic unit that can be independently negated (cf. Haspelmath 2010)
But alternatively, and certainly for the purposes of this blogpost, we can take ‘clause’ as a primitive notion. As Anna Wierzbicka has reminded us (in many works, e.g. in Wierzbicka 1996), any attempt at consistent definition must ultimately work with undefinable (= primitive) words.
Getting back to the sentence, it may be surprising, but linguists rarely define it in a way that corresponds to actual usage.
Is a sentence a possibly complex clause?
One intuition that one often finds is that a clause is simple, while a sentence may be complex, consisting of multiple clauses. This is reflected by the definition in SIL’s Glossary of Linguistic Terms:
“A sentence is a grammatical unit that is composed of one or more clauses.”
However, this will not work as a definition, because a clause may itself be “complex” – as everyone agrees. For example, a relative clause may contain an adverbial clause (the pen [that I bought [after my pencil broke]]), or an adverbial clause may contain a relative clause ([when I bought the pen [that I am now using]]).
Someone might suggest that the situation is different when two clauses are conjoined, because in that case, one clause does not contain the other. However, conjoining is usually (except in Chomskyan syntax) be thought to create constituents of the same type – two conjoined nominals are still a nominal, two conjoined adpositional phrases are still an adpositional phrase, and so on. Thus, two conjoined clauses should still be a clause.
Is a clause a CP?
Many linguists think that the technical vocabulary that is taught in many syntax courses provides a solution – after all, why have this technical vocabulary if it does not add precision? I cannot answer this latter question, but it certainly does not help to say, for example, that a clause is a “CP”, while a sentence is an “S” (as is done in Larson 2010: 293). In that textbook, Larson uses the term “sentence” in the first half, but switches to CP in the second half. At some point, he replaces “S” by “TP”, but then later he keeps using “sentence” for what in his analysis is really a CP.
Larson’s book would have been less confusing if he had used only “clause” or only “sentence”, but this wouldn’t have conformed to general terminological usage (so he preferred general usage to precision). By contrast, it seems to me that the definition in (1) above corresponds quite well to standard usage (and it does not need the “CP/TP” terminology involving the controversial functional head notion).
Is a sentence a discourse unit?
In the Facebook discussion, Alexandre François made the proposal that while “clause” is a syntactic concept,
“A sentence is a unit of discourse: It is a self-contained unit of speech endowed with one speech act / illocutionary force (declarative, interrogative, imperative…).“
This definition is similar to the one given by Dictionary.com:
a grammatical unit of one or more words that expresses an independent statement, question, request, command, exclamation, etc., and that typically has a subject as well as a predicate, as in John is here. or Is John here? In print or writing, a sentence typically begins with a capital letter and ends with appropriate punctuation; in speech it displays recognizable, communicative intonation patterns and is often marked by preceding and following pauses.
This is similar to the definition that I proposed above because illocutionary force (the difference between the speech act types: statements, questions, and directives) is not usually taken as a property of a unit that can be part of a sentence. A unit that expresses a directive cannot be a part of another sentence: In the example She told me to go home, the expression to go home is part of another sentence, but it is not thought to be a speech act by itself. And embedded questions (as in He asked me where I lived) may be said to be questions, but not in the sense of speech acts. Similarly, it is not usually possible to conjoin units with different illocutionary forces, e.g. questions and statements (?*The sun is shining and are we going out?).
The main reason why I think that the definition in (1) is better is that it specifies that a sentence is a kind of clause. By contrast, a definition in terms of a discourse unit and illocutionary force would include expressions such as “hello”, or “ouch”, or “me too!” when used as complete utterances, and these are not normally treated as clauses. Likewise, they are not normally treated as sentences, so I feel that a definition of a sentence as a maximal clause corresponds best to ordinary usage.
Is a sentence a clause plus dislocated elements?
Van Valin & LaPolla (1997: 38) say that a sentence consists of a clause plus a pre-clausal position for left-dislocated elements, and a post-clausal position for right-dislocated elements (e.g. I have not seen them in two weeks, the Smiths). Again, this definition is quite similar to my proposed definition, because left dislocation and right dislocation are not normally thought to be possible in embedded clauses. But as a universally applicable definition, it faces the problem that the notions of “left dislocation” and “right dislocation” are not easily applicable to different languages. Van Valin & LaPolla say that English allows a left-dislocated element before the position of the question word (Yesterday, what did Robin show to Pat in the library?), but in German, the counterpart of this would be completely impossible. So does German have “left dislocation” in the same sense as English?
“Complex sentences” are really “complex clauses”
The idea of a sentence as a possibly complex clause is very widespread, and the term “complex sentence” has been used very widely, both in comparative studies of the world’s languages (e.g. Austin 1988; Bybee & Noonan 2002) and in psycholinguistic studies of English (e.g. Diessel 2004). It seems that in all these cases, a “complex” clause/sentence really means “clause that includes at least one other clause”. Indeed, some authors (e.g. Culicover 2009: Chapter 7) use the more accurate term complex clause.
“Sentence” has no counterpart in nominal syntax
It has often been suggested that the internal structure of clauses and the internal structure of nominal phrases show similarities, but there is no counterpart to the term “sentence” in nominal syntax: We have no special term for a maximal nominal, i.e. a nominal that is not part of another nominal. Just like clauses (which may contain other clauses), nominals may be recursive and contain other nominals. Everyone knows this, but nobody seems to miss a term for a maximal nominal.
Thus, I suspect that we wouldn’t miss the term sentence if it didn’t exist, because every sentence is a clause, and we rarely need to talk about maximal clauses in contrast to non-maximal clauses. (In fact, German does not make the distinction between “clause” and “sentence” and uses Satz for both notions. When I speak about grammar in German, I don’t miss the distinction.)
While we are at it: What is a “main clause”?
Many authors distinguish between subordinate (or embedded) clauses and main clauses, but what exactly is a main clause? Of course, we all know stereotypical examples such as:
She e-mailed me [because she didn’t have my phone number].
Here there is a causal subordinate clause, and if a student said She e-mailed me is the main clause, we wouldn’t say that this is wrong. But in fact, we also all agree that a subordinate/embedded clause is part of another clause, so there are really only two clauses here: the subordinate clause, and the full clause (containing the subordinate clause).
In the Facebook discussion, Rob Truswell noted that linguists often talk in a sloppy way, as if in such complex sentences there were two non-overlapping clauses. For example, in the following example
What1 do you think [that Mary ate _1]?
linguists often say that what has been fronted “from the subordinate clause into the main clause”. But of course, in I think that Mary ate pasta, the nominal pasta is not only part of the subordinate clause, but also of the full clause (by transitivity of the part-whole relation).
So what do we call the part of the clause that the question word what has been moved into in this example?
Before getting to an answer to this question, we need to consider yet another term: matrix clause.
Main clauses and matrix clauses
What is the difference between a “main clause” and a “matrix clause”? In Trasks’s Dictionary of English grammar (2000), we read:
“A matrix clause is often a main clause . . ., but it need not be: it can itself be a subordinate clause. In the sentence The victim told the police [that the man [who attacked her] had had a beard], the subordinate clause who attacked her is contained within the subordinate clause that the man . . . had had a beard.”
I am not sure how frequently the distinction between main clause and matrix clause is made, but it could be made, and in fact it should be made unless one wants to allow the possibility of a main clause that is itself a subordinate clause – and this does sound contradictory.
Thus, in What do you think [that Mary ate]?, the question word what has been fronted out of the subordinate clause and is now included the part of the matrix clause that is not the subordinate clause. For this part, it is useful to resort to the special term matrix-clause fragment (= the part of the matrix clause that is not the subordinate clause).
This may sound like a strange term, but if we want terminological consistency (as we should if we aspire to rigorous science), then I see no way around it. Note that I did not invent the term: The IDS’s grammatical glossary has the term Hauptsatzfragment (Haupt-satz-fragment [main-clause-fragment]), which is defined in roughly this sense.
So we now have a definition of matrix clause (and matrix-clause fragment), but we still have no definition of main clause.
Is a main clause the same thing as “a matrix clause that is not part of another clause”? Is it an “independent matrix clause”, i.e. a sentence that contains another clause? I don’t think so.
The current Wikipedia entry https://en.wikipedia.org/wiki/Independent_clause equates main clause with independent clause, and it seems to say that an independent clause is not part of another clause (“can stand by itself”). The SIL Glossary’s definition of main clause is similar (“may stand alone as a complete sentence”).
I think this is correct, and I would simply define a main clause as “a clause that is not a subordinate clause”. (This ignores the distinction between subordinate and dependent clauses that I made in Haspelmath 1995, but this blogpost is already long enough…)
The definitions proposed here are summarized as follows:
clause: A clause is a combination of a predicate (full verb or nonverbal predicate) and its arguments, plus modifiers.
sentence: A sentence is a maximal clause, i.e. a clause that is not part of another clause.
complex clause (= “complex sentence”): A complex clause is a clause that contains at least one other clause.
subordinate/embedded clause: A subordinate/embedded clause is a clause that is an argument (“complement clause”), and adnominal modifier (“relative clause”), or an adverbial modifier (“adverbial clause”) (but see Haspelmath 1995).
matrix clause (of subordinate clause S): The matrix clause of subordinate clause S is the minimal clause that contains it.
matrix-clause fragment: A matrix-clause fragment is the part of a matrix clause that is not the subordinate clause.
main clause: A main clause is a clause that is not a subordinate clause.
I struggled quite a bit to come up with these definitions (and I find it sobering to realize that it is so hard to define even these rather simple, everyday terms….) – so I’d be curious to hear from my colleagues what they think. Do they correspond largely to actual usage, as I intend them to do? Can they be improved upon? Did I omit some crucial considerations?
Haspelmath, Martin. 2010. The interplay between comparative concepts and descriptive categories (Reply to Newmeyer). Language 86(3). 696-699