The Microsoft Manual of Style tells us not to use once as a subordinator because it could lead to ambiguity. Specifically, from the 1995 edition:
To avoid ambiguity, do not use as a synonym for after.
After you save the document, you can quit the program.
Once you save the document, you can quit the program.
(That's the entire entry.) Ambiguity? Stop a moment and try to concoct an English sentence where subordinator once could be taken to be something else. It can be done — see below — but it's not easy, and I doubt that examples like the ones below were what the MMoS folks had in mind.
Instead, I think they were concerned about the fact that once has three primary uses and that readers (especially readers who are not native speakers of English) might not understand which one was intended until partway into the sentence: TEMPORARY POTENTIAL AMBIGUITY (TPA). (I'll take up the concern for non-native speakers in a later posting.)
(Hat tips: to Tyler Schnoebelen, who first brought up the MMoS advice to me; to Thomas Grano, who first collected data on the written use of once; and to Jason Grafmiller, who's been working on the syntax and semantics of subordinator once for a while now.) Even though I've extracted some subtopics for future postings, this one is going to be fairly long. You've been warned.
Basic facts: once has three principal uses —
(I) subordinator once, as in Once you're read the instructions, press AGREE;
(II) past adverbial once 'at some time in the past', as in Once there were dinosaurs that roamed the earth;
(III) single-shot adverbial once 'on a single occasion (not multiple occasions)', as in I did it once, but I didn't enjoy it, so I didn't do it again.
Single-shot once can occur in clause-initial position (which is of course where subordinator once occurs) only in very special circumstances — for example, as an adverbial subject, as in Once is enough; and as a contrastively focused fronted adverbial, as in ONCE I would do it, but not twice — so it's largely irrelevant here.
Past adverbial once can indeed occur clause-initially, but
(a) if what follows once is not compatible with a past reading, then clause-initial once pretty much has to be the subordinator, and
(b) if the sentence goes on with another clause, we're looking at subordinator once, not past once.
On (a): a sentence that begins Once I see the Great Pyramid… has to have subordinator once, as do sentences beginning Once I've seen the Great Pyramid… and Once I am gazing at the Great Pyramid…
On (b): when you read (or hear) Once I saw the Great Pyramid, that could be the end of the sentence, with the meaning 'I once saw the Great Pyramid' — past adverbial once. Or it could be a subordinate clause preceding a main clause, with the meaning 'Once/when/after/as soon as I saw the Great Pyramid,…' — subordinator once. (These subordinators are not exact equivalents, but that's a topic for another posting, some day.)
The only issue, in fact, is whether the WRITTEN sentence should be punctuated to show the difference (the SPOKEN sentence will allow either interpretation):
Once I saw the Great Pyramid, I realized it was a trimph of engineering. [subordinator once]
Once I saw the Great Pyramid, it was a triumph of engineering. [past adverbial once: in writing, a "comma-spliced" sence (which normally would be punctuated with a semicolon or some stronger separator, though entirely kosher in speech]
The big point here is that the written clause with once is only potentially ambiguous, and then only during a brief period of processing time. Why should (some) people be so concerned about such things?
[Digression, to get back to potentially ambiguous (and complete) sentences beginning with once; then I'll return to TPA. Here's one way to concoct some. Start with a sentence-initial subjectless predicational adjunct (SPAR) like:
(1) Once chair of the committee,…
Now the once in (1) can be either a subordinator or a past adverbial; there are two different types of SPARs here. Only a few subordinators participate in SPARs with predicative NPs like chair of the committee — after, in particular, does not — but once (conveying something like 'upon becoming') does. Meanwhile, a rather different type of SPAR with predicative NPs can be introduced by (certain) time adverbials, as in
(1') Twice/Long/Never chair of the committee, Kim doesn't seek power.
Past adverbial once can occur in this construction, so that (1) is potentially ambiguous. Usually real-world plausibility will strongly favor one or the other reading when continuations are supplied for (1). So the continuation in (2') favors the subordinator interpretation, while the continuation in (2") favors the past adverbial interpretation:
(2') Once chair of the committee, Kim will revise its procedures.
(2") Once chair of the committee, Kim is no longer influential in its deliberations.
But in principle, both interpretations are available for both sentences.
So, yes, you can concoct sentences in which clause-initial once contributes to a potential ambiguity for the sentences as wholes. But you won't come across them very often, and they simply can't be taken seriously as providing a rational basis for objecting to subordinator once.]
Clearly, the perceived problem is that if a clause begins with once, this word cannot be interpreted uniquely at the point at which it is read or heard. But seeing this as a problem is profoundly silly. It would be wrong to demand that the interpretation of the first word of a sentence be uniquely determined, or even that the unfolding interpretation must be determinable within a few word. This is occasionally so, but in general that's just not the way sentence processing works.
First, a case where it seems that the interpretation of the first word is in fact uniquely determined: if the first word is predestination, you pretty much know what's going on so far. But only up to a point: predestination is a noun, but it's not necessarily the subject of the sentence; it could be the first part of a noun-noun compound: Predestination theories abounded in the 17th century.
[Digression: a very attractive — because simple — idea about how sentence processing works is that it proceeds word by word, entertaining all possibilities for the interpretation of each word and all possibilities for incorporating each new word into syntactic constituents (partial or full, each with its own interpretation) with preceding material. This is a total train wreck: the number of partial parsings/interpretations explodes exponentially within a few words. This cannot be what people do, nor is it a remotely practicable approach to computational modeling of natural language processing.
There is now a gigantic literature on the subject, building on many ideas: that alternative parsings/interpretations could be "bundled" into higher-level constructs, that some decisions could simply be deferred, that predictions/expectations about interpretations could guide understanding, that discourse organization and contextual factors could powerfully bias understanding, and so on.
Of course sentence understanding proceeds, in some sense, word by word, but it can't proceed in such a simple way.]
Yet this simple incremental view of sentence understanding probably lies behind objections to subordinator once: if the first word of a sentence has several distinct interpretations, the entire sentence is problematic; that is, until the interpretation is resolved later in the sentence, the whole thing is trouble.
Now, there's a famous phenomenon known as GARDEN-PATHING, in which almost everyone's understanding of a sentence as it unfolds leads them to a wrong analysis, from which it is incredibly difficult to uncover, as in The horse raced past the barn fell (which has come by on Language Log most recently here). And there are intermediate cases, in which some readers might be temporarily misled, though they can recover with some effort. But temporary indeterminacy is the way of the world and is problematic only in special cases.
I know, you're going to tell me that if a sentence begins with a or the (as so many do) you know you're dealing with an article.
But each of the English articles has many different uses, so that recognizing a or the at the beginning of a sentence doesn't tell you a lot about what's happening in that sentence:
A lion is a ferocious beast. [generic]
A lion mauled me. [specific]
The lion is a ferocious beast. [generic]
The lion mauled me. [specific, given in context]
The best route to San Jose is 101. [unique]
The bigger they are, the harder they fall. [??]
(This is the tiny tip of a really big iceberg.)
Most sentences, no surprise, begin with really common words, and "really common words" in fact (a) are actually instances of several clearly distinct lexical items, not "the same word" at all, or (b) have a wide variety of uses, not easily subsumable under a single definition. So when you're looking at (or hearing) the first word of a sentence, most of the time you really don't know what's going. Understanding language is a gigantic exercise in living with temporary indeterminacy.
Human beings are highly skilled at this task. Not only do we have very complex systems of form-meaning pairings, but we also have extraordinary abilities to calculate the likely pairings in context (linguistic, social, cultural). We are exquisitely social animals, and this fact is central to the way we produce and understand language. (I'll post later about views that deny this generalization.)
In any case, we are, all the time, deferring interpretations and also hypothesizing likely interpretations on the basis of context. Yes, we can sometimes be temporarily misled, and occasionally disastrously misled (and it's important to do research into when and how misleadings arise, and how serious they are), but it would be lunatic to insist that
If it can possibly lead to trouble, it should not be permitted. [a "zero tolerance" policy]
If you take that advice to heart, you cannot speak or write anything.
Back to subordinator once. First thing to say: it's entirely standard — dictionaries list it without comment — and has been around for centuries. The OED has cites (including cites from literary sources) from the 15th century on, for once with full clauses and once in SPARs (single-shot and past adverbial once go back to Old English). It's not at all uncommon; in a recent search for once in the New York Times archives, I found 7 instances of subordinator once in the first 30 hits. Other searches of corpora (by Grano and Grafmiller) indicate that single-shot once is more frequent than subordinator once, though in spoken corpora the disparity between them is not great.
The history of subordinator once in the advice literature is thin indeed. There's MMoS. And Evans & Evans (A Dictionary of Contemporary American Usage, 1957) say that "in current English [once] is often used as a conjunction" and though it "condemned by some grammarians … it is acceptable in the United States." I haven't yet found the grammarians who condemn it, and all the evidence I have seen indicates that the usage is venerable and widespread in the English-speaking world.
Then there's Fowler (Dictionary of Modern English Usage). Fowler says that subordinator once equals if once or when once and suggests replacing this once by one of these, or by simple if or when. (Language change has proceeded so far that I find if once simply unacceptable and when once dubious, because it sounds redundant to me!)
Fowler's doubts about this once stem from his perception that it has a "vigorous abruptness" that makes it "suitable … for highly literary expression" or in conversation, but not for anything in between. I don't know where he got this idea — it's probably a highly personal and idiosyncratic judgment on stylistic matters — but no one else seems to share his opinion. Not even Burchfield in the New Fowler's; Burchfield just says that once can be used as a conjunction.
[Since I will be misunderstood if I don't say this: I am not saying that everything in Fowler is an expression of a personal crotchet of his, only that some things, in particular this thing, are.]
As a last twist on the subordinator once story, I have to point out that the suggested replacements (most often after, sometimes when) are themselves at least as temporarily potentially ambiguous as once.
Start with after. First of all, it's either a preposition (P) or a subordinator (Sub). (I'm skirting all sorts of issues here, both substantive and terminological, in an attempt to find a way of making the distinction clear to most readers without saying things that I think are confused.) Compare:
After [P] the earthquake, we had no electricity.
After [Sub] the tremors ended, we were still terrified.
In both cases, after will be followed by a NP, and most of the time you won't be able to tell whether that NP is the object of the P after or the subject of the clause introduced by the Sub after — until you see what happens later in the sentence.
And then there's the fact that the P after can denote succession in time or space or some metaphorical domain (especially in an idiom):
After breakfast, we have coffee. [temporal]
After Kim in the reception line came Sandy. [spatial]
After Formatz, Gribiche is Terry's favorite composer. [metaphorical]
After some acceptable food, Jan ransacked the refrigerator. [metaphorical, in the idiom be after 'seek']
(Entertainingly, the last of these examples is, out of context, potentially ambiguous all the way through, not just temporarily.)
Again, this is the tiny tip of a gigantic iceberg; the OED entry for after provides many uses, senses, and subsenses, dwarfing the OED entry for once.
When is similar. Here, the first big cut is between interrogative when (Int) and adverbial subordinator when (Sub) (roughly, 'at what time' vs. 'at the/that time'):
When [Int] Kim first saw Sandy is a mystery.
When [Sub] Kim first saw Sandy, they fell in love.
Much of the time, these can't be distinguished until the subordinate clause is finished, much as with the temporary potential ambiguity of subordinator once vs. past adverbial once.
Sub when also has a variety of uses, as you can appreciate by looking at the OED entry. For instance, there's a distinction between point-denoting when (the use people probably think of first), as in
When I finish an article, I'm relieved.
and interval-denoting when 'while', as in
When I'm writing an article, I'm intensely focused.
In these two cases, the potential ambiguity is resolved by the aspect and aspectual type (Aktionsart) of the verb in the dependent clause, but examples that, out of context, are not so resolved are not hard to construct:
When I write an article, I'm exhilarated.
(either point when 'when I finish writing an article' or interval when 'during the time when I am writing an article').
[Interestingly, some semanticists have tried to unify these two senses, and some others, under a single semantic interpretation that isn't just temporal but also has some non-temporal components. This is a very attractive idea, but I'm not at all sure that it tells us much about what people do when they produce and understand sentences. This might be a case where Semanticists' Truth and Speakers' Truth part ways. That is, though various "senses" of subordinator when might be analyzable as just special cases of a single semantic interpretation, it might be that speakers distinguish a number of them, the way dictionaries do, in which case decisions have to be made (eventually) during processing as to which one you're confronted with.]
In any case, the idea that subordinator once is especially problematic because of TPA founders as soon as you compare it to subordinator after (and when). But why do people not see that what they say about subordinator once should apply equally to subordinator after and when? That's a question for another posting.