July 10, 2005

Still more Declaration of Independence

On the occasion of America's national day, the Fourth of July, Geoff Pullum has pointed out a dangling modifier in the Declaration of Independence.  In follow-up e-mail, a correspondent notes a restrictive relative which in this document -- not a surprise, really, but still gratifying.  And now I supply a striking failure of parallelism, in one of the senses of parallelism that appears in the advice manuals.

Close to the end of the Declaration comes the bold

(1)  We must... hold them, as we hold the rest of mankind, Enemies in War, in Peace Friends.

Yes, yes, this was deliberate.  It's an instance of CHIASMUS, a rhetorical figure in which elements are inverted in the second of two matched phrases, so as to foreground, and emphasize, these elements.  (It's also an example of ASYNDETIC COORDINATION, lacking an explicit conjunction, but that's not my topic for today.)  Still, it falls foul of Strunk & White's 19th principle in their Elements of Style, "Express coordinate ideas in parallel form"; in fact, it's a much bolder violation of this principle than the example that Strunk & White begin their discussion with:

(2)  Formerly, science was taught by the textbook method, while now the laboratory method is employed.

Strunk & White continue their exposition on maxim 19 with various types of questionable coordinations, mostly of the sort we've been calling "WTF coordination" here at Language Log Plaza (most recently, in Eric Bakovic's report of a possible Bushism and in my discussion of two specific cases, one involving recipe register features, the other coordinate questions).  That is, they lump together rhetorical parallelism and the requirements of syntax.  As it turns out, they also work with an implicit, unexamined theory of coordination that's seriously confused.  And they cast their advice in very general terms, without seeming to realize that their rules actually make predictions about what's acceptable English, many of which they would surely not welcome.

Strunk & White aren't alone in these respects.  As I'll illustrate briefly from two recent manuals, the advice literature on parallelism exhibits all three of these problematic features: a fuzzy notion of parallelism (more generally, a failure to distinguish grammar, usage, and rhetoric), a seat-of-the-pants syntactic theory, and wildly overgeneralized prescriptions.

But first, restrictive relative which.  Language Log reader Mike Jacovides reminded me, on July 5th, that the very first sentence of the Declaration of Independence has one:

When in the course of human events, it becomes necessary for one people to dissolve the political bands which have connected them with another,...

This is scarcely a surprise to scholars of English grammar.  Occurrences of restrictive relative which are all over the place.  Still, it's nice to see one right up front in the Declaration.

On to parallelism.  The Declaration's striking chiasmus, in (1), has contrasting phrase orders (with the syntactic functions of these phrases held constant).  Strunk & White's example, in (2), is subtler, involving contrasting syntactic functions for the phrase X methods -- oblique object in the first clause, subject in the second (both clauses being passive, and so in some sense entirely parallel in form); the main verb is also varied (taught in the first clause, employed in the second), but that pretty much comes with the territory.

Let me remind you: (2) (from the 4th edition of The Elements of Style (1979), p. 26) is supposed to be BAD.  Those of you who thought this was an effective deployment of NPs within the matched passive clauses are just wrong, according to S&W.  (Well, I thought it was pretty good, and so did my sophomore seminar students who were assigned S&W in writing classes at Stanford.  But what do WE know?  I'm a linguist, and linguists are widely believed to be enemies of all that is good, true, and beautiful about language.  And they're just kids.)  S&W want you to rewrite (2) as (3), which strikes me, and my students, as clunky and flat-footed:

(3)  Formerly, science was taught by the textbook method; now it is taught by the laboratory method.

S&W explain their maxim 19, "Express coordinate ideas in similar form", as follows:

    This, principle, that of parallel construction, requires that expressions similar in content and function be outwardly similar.  The likeness of form enables the reader to recognize more readily the likeness of content and function...

    The unskilled writer often violates this principle, mistakenly believing in the value of constantly varying the form of expression.  When repeating a statement to emphasize it, the writer may need to vary its form.  Otherwise, the writer should follow the principle of parallel construction.

At least they leave some room for chiasmus and for lexical variation, in cases of extreme exigency.  Like the Declaration of Independence, maybe.  But (2) won't do.

Note that S&W don't just deprecate (2).  They also attribute motives to the writer, who, they suggest, is guilty of "mistakenly believing in the value of constantly varying the form of expression".  Now, pointless variation is in fact a sin of unpracticed writers, but lots of variation has a point, so there's no justification for viewing all variation with suspicion.

The attribution of motives gets worse.  On pp. 26-7, S&W, quite extraordinarily, tell us:
    The [first] version [(2)] gives the impression that the writer is undecided or timid, apparently unable or afraid to choose one form of the expression and hold to it,.  The [second] version ([3]) shows that the writer has at least made a choice and abided by it.
My student who wrote his course paper on faulty parallelism, Jerry Zee, was offended by this bit of psychoanalyzing.  So was I when he drew my attention to it.  I mean, S&W, where do you guys get off?

But let's get specific.  We're supposed to express coordinate ideas in parallel form.  In particular, no shifting phrases from one syntactic function to another.  What about  the ordering of constituents?  Is the following bad writing?
They first decide by vote that -orum is to be the plural ending of the 'genitive' case ('of the cactuses'), and then they start arguing about the plural ending for the 'dative' case ('to the cactuses').  (Guy Deutscher, The Unfolding of Language (NY: Metropolitan Books, 2005), p. 4.  I'll have some quotes from Deutscher just because I happen to be reading the book right now.)
The non-parallel elements are they first decide... and then they start..., with varying orders of adverb and subject.  Is this bad?  I think not.

What about conjoining a passive with an active?
... in the latter days of the language, when the endings on nouns were worn away and disappeared.  (Deutscher, p. 6)
I wouldn't have noticed this "non-parallelism" if I hadn't been looking for such things.

Once you start taking S&W's bald notion of "parallel form" seriously, you realize just how silly it is.  Are transitives conjoined with intransitives bad?  Main verb VPs conjoined with auxiliary verb VPs?  Positive constituents with negative ones?  NPs with a determiner conjoined with ones without?  NPs with adjectives conjoined with NPs that have none?  As in the following examples, most of them invented:
They drank the poison and died.
They knew the answer and couldn't wait to show off.
They could see the creature but couldn't identify it.
I sing of arms and the man.
She craved boys and strong drink.

What about conjuncts with ellipses?  Are they parallel to conjuncts with explicit elements?  Why or why not?  Consider the following:
The language machine allows just about everybody... to tie these meaningless sounds together into an infinite variety of subtle senses, and all apparently without the slightest exertion.  (Deutscher, p. 2)
Two expressions connected by and.  They sure don't look parallel in form.  Is this bad?

I could go on like this forever.  Most kinds of variation in the internal form of matched phrases are entirely innocuous.  S&W haven't thought for a minute about what counts as relevant parallelism in syntactic structure and when it's relevant to rhetorical grace and effectiveness.  What they're doing is arguing from particular cases: identifying examples they find unacceptable in some way. assigning a label to the "error" in question, and then proscribing all instances of that sort.  In so doing, they overgeneralize drastically, barring all sorts of unexceptional examples; their "rules" make predictions, many of which are just wrong.

So far this is all about rhetoric.  No one would, I hope, claim that any of the examples I've given so far are UNGRAMMATICAL.  But at this point S&W pass on to a collection of examples where many people, including several Language Log posters, would entertain that possibility.  We have reached the borders of WTF-land.  S&W see no boundary here; for them it's all problems with "parallelism" in some sense or another, who cares which.  This is a disservice to their readers.

And they have no vocabulary at all for talking about the errors they exhibit, beyond saying that expressions are non-parallel because words in them are misplaced or missing or incorrectly chosen.  They have no vocabulary because they have no conceptual apparatus for syntactic structure, in particular the structure of coordinate expressions.  I'm going to have to take the first steps towards remedying these lacks, just so we can talk about the data.

What we're looking at, mostly, is "factorable coordination", in which conjuncts Y and Z (I'll restrict myself to two conjuncts for simplicity) are associated with a factor X, as in the following:

Mary and her mother shouted. 
    (conjuncts Mary, her mother; factor shouted)
Mary shouted and got her mother's attention.
    (conjuncts shouted, got her mother's attention; factor Mary)
Mary was tired but happy.
    (conjuncts tired, happy; factor was)
Mary bought a screwdriver and hammer.
    (conjuncts screwdriver, hammer; factor a)

There's a huge literature on factorable coordination and the conditions it must satisfy.  (Here I tell you that I'm not going to refer to this literature, or explain where my somewhat idiosyncratic terminology comes from.  This is a Language Log posting, not a journal article.)  Among these are a category likeness condition, a distributivity condition, and a factor constancy condition, all of which could be seen as requirements on "parallelism":

Category Likeness: the factors must belong to the same syntactic category.

Distributivity: the factor must be well-formed in combination with each of the conjuncts.  (Crudely, X (Y + Z) = (XY) + (XZ).)

Factor Constancy: the factor must have the same semantics in combination with each of the conjuncts.

According to these conditions, the following coordinations are ill-formed (factors are in small caps, conjuncts in bold face):

*Mary BECAME enamored of the law and a judge at age 30.
    (Category Likeness violated: AP vs. NP)

*Analysts say that's evidence Microsoft should -- and likely is -- TAKING GOOGLE MUCH MORE SERIOUSLY.
    (Distributivity violated, because *should taking Google... is ill-formed)

*He PUT OUT his cigar and the cat.
    (Factor Constancy violated: put out 'extinguished' vs. put out 'expelled')

It's long been known that these formulations appear to be too strong; Category Likeness, in particular, is apparently violable in certain contexts, as in an example I caught on NPR this morning:
... no proof that the man is dead, or even a Taliban prisoner.
(In such cases, it's reasonable to ask -- as a great many investigators have done -- whether the problem lies with the formulation of Category Likeness or with the notion of SYNTACTIC CATEGORY itself.)

The formulations might be too weak; Distributivity, for instance, might be strengthened by requiring Relational Constancy, that is, by requiring that the factor bear the same syntactic relation to each of the factors.

And there may well be further conditions that don't have to do with "parallelism" -- a Constituency condition, for instance, requiring that each conjunct be a single syntactic constituent.

In addition, there are plenty of uses of coordinating conjunctions that are not, or at least are not obviously, instances of factorable coordination (as in tags: Mary is a lawyer, and so is Emma), and there are several types of factorable coordination, not all subject to the conditions in the same way.

Back to S&W.  What's supposed to do all the work for S&W is not the conditions above, but, so far as I can make out, a condition of Structural Likeness, requiring not identity of category for the conjuncts -- that is, identity with respect to their external distribution -- but identity of internal structure.  As we saw above, Structural Likeness (as a principle of syntax) massively makes the wrong predictions about acceptability (as well as rhetorical effectiveness).  There are a few cases, which I'll look at in a little while, where it seems to do some work that Category Likeness doesn't do, but on the whole it's a bad idea.  Once you bring out S&W's implicit assumptions, you can see that they're not very plausible, and that S&W's discussions are likely to confuse and mislead students.

S&W's successors have not improved on things.  Here I'll look briefly at the way two recent manuals treat parallelism: Diana Hacker's Rules for Writers (5th ed., Bedford/St. Martin's, 2004) and Andrea Lunsford & Robert Connors's The New St. Martin's Handbook (4th ed., Bedford/St. Martin's, 1999).  (S&W, Hacker, and Lunsford are not entirely random choices on my part.  These were among the authors my Stanford students consulted in their writing classes.)

Hacker's main instruction is justified on rhetorical grounds:
9  Balance parallel ideas.  If two or more ideas are parallel, they are easier to grasp when expressed in parallel grammatical form.  (p. 87)

This is familiar enough, and not very useful.  But then Hacker introduces a principle not in S&W:
Single words should be balanced with single words, phrases with phrases, clauses with clauses. (p. 87)

Oh dear, this is seriously bizarre advice.  Not joining phrases and clauses (*I asked a few questions and if I could be allowed to leave early) is generally good advice, which would follow from Category Likeness, or possibly Factor Constancy, as well as from Structural Likeness.  But not joining single words with multi-word phrases is terrible advice; it would, however, follow from Structural Likeness, though not from any of the more credible conditions above.

Hacker seems willing to stick to her (largely unarticulated) principles, though.  But at what cost?

Oddly, there are no bad examples of word-phrase mixing given.  There is, in fact, no example at this point in the text of single words joined with single words.  Instead we get (p. 87):
A kiss can be a comma, a question mark, or an exclamation point. (Mistinguett)
This is a fine sentence, but is she telling us that there would be something wrong, even just rhetorically, with a version in which the indefinite article is factored out?
(4) A kiss can be a comma, question mark, or exclamation point.

This is a subtle point, subtler than Hacker realizes, I suspect.  To start with, question mark and exclamation point are written as two-word sequences, but as far as English structure goes, they're just compound nouns.  That is, they're single words.  In which case, (4) involves a coordination of three single words -- comma, question mark, and exclamation point -- and should be fine by the word/phrase/clause rule. 

Well, fine only if the structure of the predicative in (4) has a in combination with a coordination of comma, question mark, and exclamation point.  But maybe Hacker is thinking of the conjuncts as the things that are separated by commas and/or conjunctions in writing, so that the three conjuncts are a comma and then question mark and then exclamation point.  If so, then (4) is not so good rhetorically.

But Hacker probably thinks that question mark and exclamation point are two-word phrases, not single words.  Then if (4) has a structure with a factored out, it's not so good rhetorically, because it would have the single word comma coordinated with the phrases question mark and exclamation point.  On the other hand, if a comma is a conjunct in (4), then (4) is rhetorically fine, since it would be a coordination of three two-word phrases.

So, according to what you think about the structure of (4) and what you think about the word status of compound nouns, (4) is either rhetorically fine (in two cases) or not so fine (in two cases).

I'm going on at such length about this one poor example because we -- and the students trying to use Hacker's handbook (let's not forget about them) -- don't know what her assumptions are on these matters, so that as soon as we try to go beyond the Mistinguett example, even to a closely related one, we don't know what to do.

Back in the real world, people do, of course, coordinate single words with multi-word phrases, all the time.  Here's an example from Deutscher, his second sentence in fact:
Other inventions -- the wheel, agriculture, sliced bread -- may have transformed our material existence, but... (p. 1)
There it is: a single word conjoined with a determiner-noun phrase and an adjective-noun phrase.  Sounds great to me.

As with S&W, Hacker starts out with purely rhetorical matters but then, still marching under the banner of parallelism, advances into the neighborhood of syntax.  By page 89 we're up to the sub-principle:
9a  Balance parallel ideas in a series.  Readers expect items in a series to appear in parallel grammatical form.  When one or more of the items violate readers' expectations, a sentence will be needlessly awkward.

It's stated as rhetorical advice, but the examples slide into some very dubious syntax, in particular a coordination of nominal gerunds with an infinitival VP:
Hooked on romance novels, I learned that there is nothing more important than being rich, looking good, and to have a good time.  [Hacker tells you to alter to have to having.]

The gerund-infinitive combination is way down on the acceptability scale for me, whether as object (above) or subject:
*Winning the race and to get the prize are my goals.
*To win the race and getting the prize are my goals.
The combination of an -ing-form VP and an infinitival VP in verb complements is equally bad:
*It started raining and then to snow.
*It started to rain and then snowing.

This is not some lack of rhetorical finesse.  This is rotten syntax.  The question is whether Category Likeness (or some other credible condition) would bar such combinations.  At first glance, nominal gerunds and infinitival VPs in subject and object positions look like they should just be NPs from the point of view of their external syntax, and -ing-form and infinitival complements look like they should just be VPs from the point of view of their external syntax, but these guys just don't like to coordinate.  Structural Likeness would predict this, but it's pretty much a dead loss otherwise, so the obvious tack to take is to examine the assignment of syntactic categories to these phrases.  Another technical question of syntax -- technical, but very important.

As it happens, (bad) coordinations of -ing-form and infinitival phrases turn up in discussions of parallelism in virtually every handbook, S&W being a rare exception.  Here's one from Lunsford & Connors (p. 266), an exercise to correct:
To need a new pair of shoes and not being able to afford them is sad.

I cannot believe that such coordinations are ubiquitous in the manuals because they occur with great frequency in student (or any) writing.  I've been editing other people's writing for about fifty years now, and I don't recall coming across any such examples.  The ones in the manuals all look invented; certainly, they're not attributed, though failures of Distributivity (a very frequent type of "faulty parallelism", and not only in student writing) sometimes are.  I can't help thinking that the coordinations of -ing-form and infinitival phrases found their way into the handbooks BECAUSE they were clear violations of Structural Likeness.

However, Structural Likeness is a blunt instrument, and Hacker wields it to bash some relatively innocent types of examples.  On page 90, under "parallel ideas linked with coordinating conjunctions", she disapproves of an -ing-form phrase coordinated with an ordinary NP:
At Lincoln High School, vandalism can result in suspension or even being expelled from school.  [Hacker recommends altering being expelled to expulsion.]
And from an exercise on p. 92:
Activities on Wednesday afternoons include fishing trips, computer training, and learning to dance.
Similarly, from Lunsford & Connors (p. 263):
The duties of the job include baby-sitting, house-cleaning, and preparation of meals.  [L&C recommend altering preparation of to preparing]

And on page 91, under "comparisons linked with than or as", she similarly disparages matching of nominal gerunds with infinitivals, in either order:
It is easier to speak in abstractions than grounding one's thoughts in reality.  [Hacker says grounding should be altered to to ground.]

Mother could not persuade me that giving is as much joy as to receive.  [Hacker says to receive should be altered to receiving.]

The first three of these sound just fine to me.  The other two are a bit more awkward, but not unacceptable, I think, and the problem with the Mother's persuasion sentence has as much to do, I think, with the enormous formality of infinitival VP subjects (To receive is a great joy) as with any failure of parallelism. 

The larger point is that it's a mistake to treat these examples together with the really awful coordinations of -ing-form and infinitival phrases. 

Then Hacker turns to the repetition (or omission) of function words in parallel constructions, instructing the student to:
9c  Repeat function words to clarify parallels.
...Although they can sometimes be omitted, include them whenever they signal parallel structures that otherwise might be missed by readers.

Think of this as the Mistinguett Rule: Go for a comma, a question mark, or an exclamation point, not for a comma, question mark, or exclamation point. The problem here is that the student is supposed to be able to judge when parallel structures might be missed by readers.  Well, if the students knew that, they wouldn't need Hacker's advice.  In the absence of such knowledge, they are in effect being urged by Hacker to always repeat function words, just to be safe.  This can make for some mighty clunky writing.

Granted, sometimes the repeated function word could be helpful, as in Hacker's example from page 91:
Many smokers try switching to a brand they find distasteful or a low tar and nicotine cigarette. [Hacker says to insert to after or.]
But on other occasions, it's not so clear that repetition is is the right strategy, as in Hacker's exercise on page 92:
Jan wanted to drive to the wine country or at least [to] Sausalito.
(Surely a comma after wine country would be enough to fix this one.)

In any case, prepositions omitted in second conjuncts are routine:
There is no longer fierce debate, for instance, about whether the earth is round or flat, and [about] whether it revolves around the sun or vice versa.  (Deutscher, p. 16)

In one case, infinitivals, Hacker's advice just goes against the grain.  Omission of to in second conjuncts is incredibly common.  Two examples from Deutscher:
... we'll have to dig beneath the surface of language and [to] expose some of its familiar aspects in an unfamiliar light.  (p. 10)

... it will be possible to synthesize all these findings into one ambitious thought-experiment, and [to] project them on to the remote past.  (p. 11)
(I hope you'll have noticed that all of the Deutscher examples come from the first few pages of this 358-page book.  This stuff is gut-easy to find.)

Finally, there are Distributivity violations of several types, mostly of the sort discussed in the faulty parallelism article in Merriam Webster's Dictionary of English Usage, where they are viewed as venial sins, if sins they are.  For instance, placement of matched coordinators like either... or, both... and, and (Hacker, p. 92) not only... but also:
During basic training, I was not only told what to do but also what to think.

The streets were not only too steep but also were too narrow for anything other than pedestrian traffic.

Some Distributivity violations arise from letting the second conjunct determine the form of the factor that follows, when the first conjunct would require a different form in the factor; that is, these violations are a species of "determination by the nearest", related to the very common phenomenon of agreement with the nearest (... my admiration for her versatility and artistry have continued to grow -- Bay Area Reporter, 5/26/05, p. 38).  A very common subtype (in real life, not just advice manuals) involves government of verb forms by auxiliaries:
Mayor Davis never has and never will accept a bribe.  (Hacker, p. 93)

I had never before and would never see such a sight.  (Lunsford & Connors, p. 263)
Another common subtype involves selection of prepositions by verbs or other head words:
Many South Pacific Islanders still believe and live by ancient rules.  (Hacker, p. 93)
I mostly judge instances of determination by the nearest to be straightforwardly ungrammatical, in contrast to misplaced matched coordinators, many examples of which I accept without question.  In any case, determination by the nearest is, in general, a more serious violation than misplaced matched coordinators, so that it's a mistake for advice manuals to just lump them together.

Both handbooks identify the problem in determination by the nearest as omission of words: "Add words needed to complete compound structures" (Hacker, p. 93); "including all necessary words" (Lunsford & Connors, p. 263).  Now, adding words is the fix for the problem, but the advice to add words as necessary is deeply unhelpful.  Students have to figure out when these extra words are required, what they are, and where they go, and for this they need much more specific advice.

In any case, Distributivity is a principle of grammar (concerning well-formedness), not of rhetoric (concerning effectiveness), while the Mistinguett Rule is a principle of rhetoric, not grammar.  In both cases, you can argue about when the principles apply and when they do not, but they're principles of very different sorts, and it's a mistake to think of them simply as species of "faulty parallelism".

zwicky at-sign csli period stanford period edu

Posted by Arnold Zwicky at July 10, 2005 01:18 PM