The split verbs mystery
« previous post | next post »
As a result of an exchange (1, 2 ,3 ,4, 5) with Alan Gunn in the comments yesterday, I was reminded that for many years, legal scholars throughout the U.S. were subjected to a peculiar form of stylistic tyranny, imposed by a curious work known as The Texas Manual on Style. According to James Lindgren ("Fear of Writing", California Law Review 78(6):1677-1702, 1990):
Unquestionably, the most dangerous advice in the old fifth edition of the Texas Manual was its disapproval of split verbs: "Avoid splitting verb phrases with adverbs. . . ." In other words, don't place an adverb between the parts of a compound verb. Yet Fowler and Follett (both praised in the Foreword to the Texas Manual) argued that the normal place for an adverb is in the midst of a multiple word verb. Thus the fifth edition of the Texas Manual seemed to have gotten the rule backwards. It prohibited what the experts recommend.
Specifically, this means that choices like "has always been" are to be suppressed, in favor of "always has been" or "has been always".
This is not the only bad advice in the book — Lindgren makes a strong case on other grounds for his view that "The Texas Manual on Style is one of the most pernicious collections of superstitions that has ever been taken seriously by educated people". But "avoid split verbs" is certainly the most eccentric piece of voodoo syntax since the prohibition of clause-final prepositions.
I read Lindgren's review when I wrote about the Texas Manual several years ago ("Grammatical indoctrination at law reviews", 10/8/2005), but I'd forgotten about the "avoid split verbs" nonsense, and I was skeptical of Alan Gunn's claim that
The phobia about avoiding "split verbs" is widespread among newspaper people–George Will does it all the time, as does the Wall Street Journal. The New York Times is one major exception.
So I did a quick quantitative check, using the ProQuest Historical Wall Street Journal (1889-1991) and Historical New York Times (1851-2005) as sources, and taking the placement of always relative to "has been" as a proxy for the split verbs question more generally:
| WSJ | NYT | |
| "always has been" | 997 (15.1%) |
7,741 (11.1%) |
| "has always been" | 5,581 (84.6%) |
61,770 (88.1%) |
| "has been always" | 22 (0.3%) |
579 (0.8%) |
(Yes, I know that there can be shades of difference in meaning between these choices; in my opinion, that doesn't invalidate the use of these proportions as an index of conformity to the "avoid split verbs" dogma.)
So it's true that the WSJ differs slightly from the NYT , in the direction prescribed by "avoid split verbs": but the difference is a rather small one,
Holding to his impression of a difference, Alan suggested that I look only at post-1980 WSJ. I did as he requested — and found a small change in the wrong direction for his hypothesis. In the years 1980-1991, the pattern "has always been" gets an 86.9% share (up from 84.6%), compared to 13% for "always has been" (down from 15.1%) and 0.1% for "has been always" (down from 0.3%).
For this morning's Breakfast Experiment™, I've decided to continue probing this particular prescriptivist sore tooth.
To start with, I wonder how the WSJ and NYT proportions compared to the estimates from other sources. So let's check Mark Davies' Corpus of Contemporary American English (COCA); the British National Corpus (BNC);, the LDC's news corpus (about 2.6 billion words); the LDC's corpus of transcribed English-language conversational telephone speech (about 26 million words); the Literature Online (LION) prose and drama collections; Google's web index;, and the Google News index as of this morning:
| COCA | BNC | LDC (news) |
LDC (CTS) |
LION (prose) |
LION (drama) |
Google (web) |
Google (news) |
|
| AHB | 529 (10.6%) |
86 (6.6%) |
2,185 (7.8%) |
27 (18.4%) |
37 (4.9%) |
16 (7%) |
2.82M (6.7%) |
2,037 (7.0%) |
| HAB | 4,445 (89.2%) |
1,214 (93.2%) |
25,568 (91.5%) |
119 (81%) |
618 (82.5%) |
181 (79%) |
38.7M (91.8%) |
27,061 (92.7%) |
| HBA | 9 (0.2%) |
3 (0.2%) |
192 (0.6%) |
1 (0.6%) |
94 (12.6%) |
33 (14%) |
0.657M (1.5%) |
80 (0.3%) |
The numbers are all fairly consistent. But several things strike me.
First, the frequency of the "has always been" (HAB) order is lower in the LION collections, which are on average from earlier time periods. Is there a historical change here, and could the "avoid split verbs" business be a misguided response to it?
Second, both the WSJ and NYT have slightly lower proportions of the HAB order than other comparable collections — could this be a sign of a more conservative or formal style?
(A side note: in the conversational speech, the proportion of HAB is also much lower — but compared to LION, the extra goes to AHB rather than HBA.)
So in order to check the idea that there's a historical process behind all of this, let's look at LION's drama collection in two time periods. The results are consistent with the idea that the HAB percentage is increasing over time:
| LION drama 1280-1800 |
LION drama 1801-1929 |
|
| AHB | 2 (3%) |
9 (10%) |
| HAB | 46 (65%) |
77 (86%) |
| HBA | 23 (32%) |
4 (4.4%) |
OK, so we turn back to ProQuest and check out the WSJ's time trend:
| WSJ 1890-1909 |
WSJ 1910-1929 |
WSJ 1930-1949 |
WSJ 1950-1969 |
WSJ 1970-1989 |
|
| AHB | 25 (7%) |
134 (11%) |
321 (25.8%) |
139 (15.7%) |
347 (13.5%) |
| HAB | 333 (92.2%) |
1068 (88%) |
916 (73.6%) |
747 (84.3%) |
2221 (86.5%) |
| HBA | 3 (0.8%) |
10 (0.8%) |
7 (0.6%) |
0 | 0 |
Oops! The 1890-1909 pattern was more "modern" than the 1970-1989 period was! No, that's inaccurate: the older pattern, according to the LION drama counts, was to put always more often between been and the following word, yielding the proportions 3%/65%/32%; whereas the mid-century WSJ pattern was put always more often before has, yielding the proportion 26%/74%/1%.
But all the same, something happened in the middle of the 20th century to push the WSJ's HAB percentage down by almost 20 points. Maybe the head of their copy desk was converted around 1925 to the Texas Manual's cult of grammatical snake-handlers?
No, it's not likely to have been anything so individual and specific, because the NYT shows a similar (though less extreme) trajectory, with the HAB proportion in the two decades before 1950 falling to a value almost 20 points lower than the proportion at the end of the century:
| NYT 1851-1869 |
NYT 1870-1889 |
NYT 1890-1909 |
NYT 1910-1929 |
NYT 1930-1949 |
NYT 1950-1969 |
NYT 1970-1989 |
NYT 1990-2005 |
|
| AHB | 172 (9.4%) |
241 (6.6%) |
736 (8.8%) |
1320 (16.1%) |
2239 (20.9%) |
1483 (16.6%) |
1051 (7.1%) |
499 (3.7%) |
| HAB | 1593 (87%) |
3305 (91%) |
7422 (89.2%) |
6791 (82.8%) |
8430 (78.5%) |
7391 (82.9%) |
13768 (92.7%) |
13059 (96.1%) |
| HBA | 66 (3.6%) |
85 (2.3%) |
167 (2%) |
95 (1.2%) |
69 (0.6%) |
38 (0.4%) |
33 (0.2%) |
26 (0.2%) |
In fact, if we zero in on the five years between 1945 and 1950, and compare it to the period from 2003 to 2008 (using the online NYT index for the last five years), the difference is more than 20% absolute, or a 28% increase in relative proportion (96.8/75.5):
| NYT 1945-1950 |
NYT 2003-2008 |
|
| AHB | 438 (24.2%) |
128 (3.0%) |
| HAB | 1,365 (75.5%) |
4,076 (96.8%) |
| HBA | 6 (0.3%) |
5 (0.1%) |
That's all the experiments that I have time to run this morning — and it's probably way too many for most of you.
But it leaves us with a mystery: what happened to English adverb placement in the middle of the 20th century? Was there some sort of natural social process, which for a while pushed always from position 2 into position 1 in the pattern <1> has <2> been <3>, and then reversed course? Was this natural process somehow perceived by the authors of the Texas Manual, who turned their stylistic intuition into a "rule" that blighted the lives of thousands?
Maybe; but my money's on the arrow of causation going the other way. I bet that the Texas Manual authors — or some other self-appointed usage authority who influenced them — made this rule up, and managed by top-down copy-editing coercion to shift the (written) language for a while, at least in some jurisdictions. I freely admit that I don't know whether this hypothesis is true, but it seems to me that either way, the history of this trivial question is an interesting laboratory for the quantitative study of social change.
Some of the many linguistic and cultural questions that remain: What happened to the order of adverbs and compound verbs during this period with other periodicals, in the U.S. and in the U.K.? Is there a difference between copular and auxiliary uses of "has been"? What about other adverbs and other compound verbs? What's the history of the "avoid split verbs" prescription in successive editions of the Texas Manual, and in other style guides?
And if it was really an author of the Texas Manual who invented "avoid split verbs", what was the motivation? Was it an attempt to extend the logic of "avoid split infinitives", as Alan thought? Speaking for myself, I'd be happier if it turned out to be another example of my favorite theory of intellectual history: the bar bet hypothesis.
[Apologies for the strange and inconsistent formatting of tables — I haven't yet figured out how to sneak table formatting past the moronic filters that WordPress applies to html on the way to storing it in the database.]
Update — see also "When zombie rules attack" (8/26/2008); "The true history of the split verbs rule" (12/23/2012); "The true history of the split verb rule" (12/23/2012).
Constance said,
August 23, 2008 @ 8:55 am
Could it, perhaps, derive from French custom? You don't split verb phrases for adverbs, I believe, in French. Likewise, it seems odd to end a French clause with a preposition. It just doesn't look right, at least, to me.
Mark Liberman said,
August 23, 2008 @ 9:52 am
Constance: Could it, perhaps, derive from French custom? You don't split verb phrases for adverbs, I believe, in French. Likewise, it seems odd to end a French clause with a preposition.
Well, the prejudice against preposition-stranding is the fault of John Dryden; and C.S. Lewis implies that this "curious taboo" was motivated by Dryden's gallicism; and I've noted at least one weird modern echo of the idea that English syntax should be improved along French lines, specifically with respect to the treatment of prepositions.
But I've never encountered any suggestion that the prejudice against splitting sequences of verbal auxiliaries has a French connection. The business about "split infinitives" is attributed to Henry Alford, a churchman whose other contributions include philological works on New Testament Greek. As far as I know, he had no particular interest in or affinity for French.
Specifically, in "A Plea for the Queen's English: Stray Notes on Speaking and Spelling" (1866), Alford wrote (p. 188):
That's all that he says about it. He bases his opinion squarely on a categorical assertion about English usage. His assertion was completely false, but he seems simply to have been a poor observer of his own language. There's no indication that we can shift any of the blame to the French.
Tom said,
August 23, 2008 @ 11:02 am
The prohibition on split infinitives thrives in legalese in the UK too: the rote definition of theft is "dishonest appropriation of property belonging to another with intent permanently to deprive".
Stephen C. Carlson said,
August 23, 2008 @ 12:58 pm
You also see the prohibition on split infinitives in the legalese of Presidential impeachment resolutions. For example:
Stephen C. Carlson said,
August 23, 2008 @ 1:28 pm
What makes "faithfully to execute" more interesting or at least somewhat more relevant is that the corresponding passage in the Oath of Office in the U.S. Constitution "splits" the verb:
JJM said,
August 23, 2008 @ 1:31 pm
"J'ai toujours été un européen convaincu"
– Jacques Chirac
"You don't split verb phrases for adverbs, I believe, in French."
I'm not sure where that idea about French originated.
David Eddyshaw said,
August 23, 2008 @ 5:30 pm
I've always supposed (indeed, I believe I was once explicitly taught) that the nonsense about "split verb phrases" originates in the supposed equation of (eg) Latin
amare = "to love" (rather than "love")
combining misanalysis of the English with the idea that the best way to analyse any language is by analysing the corresponding Latin translation.
In other words, the logic is, because one cannot separate "ama" from "re" with interpolated words, one must not separate the "equivalent" "to … love".
Yes, I know …
Rick S said,
August 23, 2008 @ 6:12 pm
It occurs to me that "always has been" may not be truly typical of split verb alternations:A: My sister moved to Omaha, the capital of Nebraska.B: Idiot! The capital of Nebraska is Lincoln, and always has been!Neither of the other word orders is grammatical in this construction, so AHB/HAB/HBA might not be typical of split verbs in general. Though if it has something to do with the mid-1900s shift, I can't imagine why this construction should have seen a special popularity during that period.
Rick S said,
August 23, 2008 @ 6:14 pm
Oops! Sorry for the awkward formatting in the preceding. It looked fine in the preview (I used an HTML unordered list, with no line breaks).
Mark Liberman said,
August 23, 2008 @ 7:21 pm
David Eddyshaw: the logic is, because one cannot separate "ama" from "re" with interpolated words, one must not separate the "equivalent" "to … love".
I've also been told this, but at least in the Henry Alford passage where the idea is said to originate, he doesn't mention any language other than English, and simply asserts that "this is a practice entirely unknown to English speakers and writers", who "ever regard the to of the infinitive as inseparable from its verb".
This seems to have accurately characterized Dean Alford's own writing, but of course it was total nonsense as a description of modern English, in the mid-19th century as well as today.
marie-lucie said,
August 23, 2008 @ 8:30 pm
The idea that French does not allow adverbs between the two parts of a complex verb phrase is nonsense if taken as a general rule. Some adverbs are unlikely to occur there, but by no means all.
e.g. Proust: Longtemps, je me suis couché de bonne heure.
= Je me suis longtemps couché de bonne heure.
(For a long time I went to bed early)
but only the following is acceptable:
Hier, je me suis couché de bonne heure.
(Yesterday I went to bed early)
Placing the word hier after the auxiliary would sound foreign, as would the equivalent sentence in English. That order was acceptable in past centuries, eg there are examples from the 17th C such as
Il fut hier présenté au Roi
(Yesterday he was introduced to the King)
but it is no longer used in modern French.
On the other hand, the opposite order would occur with an adverb like rarement where initial placement would be strange, but the following is quite straightforward:
Je me suis rarement couché de bonne heure
(I rarely went to bed early).
So there are semantic restrictions on the types of adverb allowed in certain positions, but no general rule encompassing all adverbs.
This is true also of complex verb phrases where the first verb is faire or laisser, where adverbs often occur after the first verb:
Elle fait souvent faire ses vêtements par une couturière.
= Souvent elle fait faire ses vêtements par une couturière.
(She often has her clothes made by a dressmaker)
Ils ne laissent absolument pas leurs enfants jouer avec ceux de leurs voisins.
(They absolutely forbid their children to play with the neighbours' children).
As for French sentences not ending with a preposition, this is true as a general rule: structural equivalents of "What are you looking at?" or "What did you do that for?"are not possible, because a French prepositional phrase cannot be split.
There is the occasional exception, for instance (eg in looking at a strange gadget)
Qu'est-ce qu'on fait avec?
(What do you do with it?)
Not to be confused with
Qu'est-ce qu'on fait avec ça?
(What do you do with this/that?)
Justin L said,
August 23, 2008 @ 8:41 pm
In Spanish it's extremely rare to "split the verb". I Googled the Spanish equivalent of "has always been" and found:
"siempre ha sido"-2.8 million hits
"ha sido siempre"-1.6 million hits
"ha siempre sido"-1200 hits–many of which are either literal translations of English (song lyrics, for example), or in journals/blogs of English speakers.
Matthew Stuckwisch said,
August 24, 2008 @ 5:42 am
With haber in Spanish it's not possible to split the verb when forming the perfect, but with estar and making the progressive, it most definitely is:
siempre estoy viajando – 3430 gh
estoy siempre viajando – 235 gh
estoy viajando siempre – 197 gh
The last one had a number of instances of "estoy viajando, siempre he pasado por …" or something similar, google unfortunately ignores punctuation. So in reality, position one is most common, and position two in between is used though when special emphasis is needed on certain parts of the sentence.
Another random example:
Siempre tengo que ir – 625 gh
Tengo siempre que ir – 7 gh
Tengo que siempre ir – 3 gh
Tengo que ir siempre – 2270 gh
Oddly enough, those 10 gh for placing in the middle a number were by native speakers (or presumably so by the URL and the sites they were found on).
Dan in Texas said,
August 24, 2008 @ 10:47 am
Thank you, Mark! As a longtime newspaper copy editor who has frequently been tormented by this bizarre superstition, I have always wondered where it came from; I couldn't find it in any style manual. And now, to learn that it might have come from a work bearing the name of my own state … Texas, Texas, I weep for you.
Arnold Zwicky said,
August 24, 2008 @ 12:33 pm
A note of caution: the prescription is a general one, framed to apply to all sequences
Aux Adv V
but Mark has looked at it only for one Aux (perfect have), one Adv (always), and one V (been). A dip in the frequency of this one combination might mean nothing at all.
There are also the auxiliaries be of the progressive, be of the passive, and the modals. And quite a few adverbs that are possible in slot 2 (including often, sometimes, rarely, occasionally, usually, and never), plus modified versions of these (very often, only occasionally, etc.). And of course an extraordinary variety of verbs.
It seems to me that the variation to look at is between Aux Adv and Adv Aux. Adv immediately after Aux is structurally restricted in various ways: for instance, never is restricted to preverbal occurrence, and a great many verb + complement constructions don't allow an adverbial intervening between the verb and the complement, though adverbials following the complement are often possible.
But I wouldn't put much weight on statistics from just one Aux and just one Adv.
TootsNYC said,
August 25, 2008 @ 1:07 pm
When i worked on a news magazine, the word "also" was always placed before the verb:
"The company also will release a new product…"
I started querying every editor or reporter who handed me a story w/ that in there. To a person, they said, "You are not supposed to split the infinitive."
They weren't worried about splitting the VERB. They were worried about splitting the INFINITIVE.
Marc in Chicago said,
August 25, 2008 @ 4:42 pm
What TootsNYC just said.
Over the years, I've had colleagues edit my "will generally be," "has often been" or similar constructions to "generally will be," "often has been" (or the like), which, to my ear, sound awkward and unnatural.
When questioned about what they found wrong with the original, the answer has always been, "You're not supposed to split an infinitive."
Fortunately, they usually relent when reminded (or informed) that "will be," "has been," and other compound verb forms AREN"T INFINITVES.
Interesting to see that the Texas Style Manual may also be [also may be?] behind this mischief.
.
Greg Miller-Breetz said,
October 6, 2008 @ 11:55 pm
The most ridiculous example of grammatical arbitrariness and the tyranny of untutored student editors that I encountered during my tenure on my alma mater's law review was the journal editors' flat-out prohibition of the passive voice. No exception to this rule was granted, no matter how small.
At first I laughed when a student editor noted "passive" next to the sentence that started my "note": "Children can sue and be sued." I agreed, yes, you have indeed identified the passive voice in that sentence. But what would you have me do? "Children can sue and find themselves defendants in litigation?" That tortured rendering of my simple, straightforward sentence could not be justified on the grounds of directness, simplicity, and clarity (the reasons given for the anti-passive voice rule), and the (law-student) editors exhibited a very unlawyerly aversion to reason and argument. What a little power will do.
Mike said,
November 6, 2008 @ 8:09 pm
I wonder whether the frequency of splitting verbs is related at all to the frequency of usage of contractions, not just in print but in speech? For example, could usage of "he's always been" (i.e. he has always been) influence writers to want to use "has always been" rather than "always has been"?
Sociolinguistics and the Botched Oath « Dan Scherlis said,
January 22, 2009 @ 3:44 pm
[…] Liberman, co-founder of the excellent Language Log, has cited the highly-influential Texas Law Review Manual of Style as a leading perpetrator of the split-verb […]
To boldly go where no chief justice has gone before « Mighty Red Pen said,
January 26, 2009 @ 6:15 pm
[…] culprit, according to Pinker, was that old bugaboo: the split verb. In his legal opinions, Chief Justice Roberts has altered quotations to conform to his notions of […]
An insurmountable obstacle in the way of a speeding train « Arnold Zwicky’s Blog said,
June 17, 2009 @ 10:14 am
[…] bit later, Mark Liberman recalled that at one time … legal scholars throughout the U.S. were subjected to a peculiar form of […]
Anne said,
July 21, 2009 @ 10:01 pm
This "don't split your verbs" nonsense has been driving me absolutely crazy. I have been studying for the CLA exam for two months now using the CLA Review Manual, Second Edition, by Virginia Koerselman. Those I have asked about this problem have told me to just do it the way the author tellls me to do it; otherwise, I will not pass the exam and no one will care what I think about split verbs except for me. Okay, I get it. I'm taking the test this Friday and Saturday. On Monday, I will be copying and mailing your article to Ms. Koerselman. I'll let you know what she says.
Richie Val said,
December 2, 2009 @ 10:50 am
"Also" presents interesting placement situations as well. "She has also sung at local theaters" seems to misplace the adverb and damage meaning. "She has sung at local theaters also" gets the meaning best, in my opinion, and "She also has sung at local theaters" is "also" better in my book.
Adrienne said,
July 23, 2012 @ 1:14 pm
Ahhh, I'm editing a book for a friend, and wanted to split both infinitives and verbs, so to be accurate, I checked online for authoritative voices. Coming upon this, I'm delighted with the thoughts and writing style of Mark Lieberman. I also thank those who made comments. My descent into the bowels of things online has certainly diverted me from today's tasks, but has been quite a romp!