English Verb-Particle Constructions

« previous post | next post »

Lately I've been thinking about "optionality" as it relates to syntactic alternations. (In)famous cases include complementizer deletion ("I know that he is here" vs. "I know he is here") or embedded V2 in Scandinavian. For now let's consider the English verb-particle construction. The relative order of the particle and the object is "optional" in cases such as the following:

1a) "John picked up the book"
1b) "John picked the book up"

Either order is usually acceptable (with the exception of pronoun objects — although those too become acceptable under a focus reading…)

1c) "John put it back"
1d) *"John put back it"

For something like (1a) and (1b) the semantic interpretation seems largely the same, and so the "optionality" refers to the grammar allowing the generation of more than one syntactic variant. In practice however, even if multiple syntactic arrangements are permitted only one can actually be produced at a given time in a given context. Acceptability judgments tend to be more delicate or varied than would be desired here. So if we'd like to investigate what factors govern the production of one form (particle-first) over another (object-first) we may examine the overall rates of use of either variant in a corpus under different conditions. Much has been written about these sorts of phenomena, including particle placement in particular (Stefan Gries has written a whole book on the topic), yet technical constraints often limit the scope of such investigations.

For instance, two factors which have been found to correlate with/against particle-first order are the heaviness of the DP-object (heavy objects tend to follow the particle), and whether or not the object had been recently referred to in context (discourse familiar objects tend to precede the particle). Stefan finds these effects over a few hundred sentences, but because the space of lexical combinations is so large there's simply not way to control for word-level effects which may be co-variate to NP-heaviness of discourse familiarity.

To get around this I wrote a script which extracts instances of verb-particle constructions from the spoken portion of COCA and tags them for particle-order. This requires a few hand-written heuristics so as not to erroneously include prepositional phrases whose order is in fact not option (e.g. "Walk down the path" is possible but not *"Walk the path down"), but nothing too technically involved. Overall, I find a particle-first rate of approximately 60% over a very large sample of roughly 50,000 such sentences. This is in line with previous work dating all the way back to the late 1970's on this topic. However, if we zoom into rates within various predicates, things appear far more varied on a lexical level. Below is a plot showing the rates of particle-first order for the twelve most frequent verbs (each verb appearing in a few thousand sentences in my sample.) The red vs. blue colors simply represent the particle-first ratio being below/above 50%.

Some verbs (Pick, set) show nearly categorical particle-first order, while others (help, get) are majority object-first.

Subsequently zooming in to look at the behavior under "bring" (since it shows a good split around 70/30), the picture remains varied. For instance, there is a near categorical gap in ordering for "bring about" compared to "bring over".

Notice that "bring back" is roughly 50/50, so conditioning on that and splitting over the head of the object DP there is again frequently categorical split in particle ordering. There are between 10 and 100 sentences for each condition below.

None of this of course explains what's driving these large, lexically conditioned gaps, but it would be interesting to keep digging into it.


  1. Robin Melnick said,

    July 26, 2017 @ 10:41 am

    There is considerable work on this beyond Stefan's large-scale 2003 investigation, and in particular exploring additional factors with a large bearing on the lexically conditioned gaps you mention. Consider the 2004 Language article by Lohse, Hawkins, and Wasow, in which the authors show that semantic dependencies among verb, particle, and NP object head noun play a greater role in the alternation than a number of other factors previously explored.

    These dependencies vary idiosyncratically with the particular verb-particle-noun combinations. For example, in something like "Let's pick up the discussion (where we left off)", no one "picks" anything, nor is anything "up" in the process, suggesting that both V and PRT critically depend on one another for interpretation in this combination, which Lohse and colleagues suggest tends to promote particle placement adjacent to verb. This contrasts with something like "He shaved his beard off", where this does entail that the subject has shaved, and something does end up "off" in the process, suggesting that V and PRT are more independent of one other in interpreting meaning than in the "pick up" combination, which here the authors would suggest tends to promote–ceteris paribus–the distal particle placement option.

  2. Michael Watts said,

    July 26, 2017 @ 11:20 am

    In practice however, even if multiple syntactic arrangements are permitted only one can actually be produced at a given time in a given context.

    I was under the impression that productions analogous to "John picked up the book up" weren't especially rare as speech errors go.

  3. boynamedsue said,

    July 26, 2017 @ 11:51 am

    Am I misunderstanding here? There are 27,000+ ghits for "brought a change about" alone, surely that contradicts the dat on bring?

    BTW, this is not a rhetorical question, I'm sure I must be misunderstanding.

  4. boynamedsue said,

    July 26, 2017 @ 11:51 am

    Am I misunderstanding here? There are 27,000+ ghits for "brought a change about" alone, surely that contradicts the dat on bring?

    BTW, this is not a rhetorical question, I'm sure I must be misunderstanding.

  5. Robin Melnick said,

    July 26, 2017 @ 11:58 am

    I should've noted in my first comment above that semantic dependencies are similarly involved in at least a couple other post-verbal constituent-ordering alternations, as discussed with regard to heavy-NP shift in Wasow 1997 (in Language Variation and Change) and with regard to PP ordering in Hawkins 1999 (also Language Variation and Change).

  6. NW said,

    July 26, 2017 @ 12:06 pm

    If you scroll down to the bottom of the 27 700 Ghit page, there are actually only four pages. The raw number at the top is never remotely reliable.

  7. Guy said,

    July 26, 2017 @ 12:06 pm


    Google hits aren't really a rigorous way of measuring these things, and without a number for "brought about a change" there isn't even enough to conclude that there might be something fishy about the COCA data, especially since the proportions are correlated with the object. This does make me wonder what variation might exist between written and spoken English on this issue, though.

  8. Orin Hargraves said,

    July 26, 2017 @ 1:06 pm

    Looking especially at your second bar graph I wonder if indeed the "heaviness of the DP object" is the explanation in most cases. Intuitively, the most frequent uses of "bring about/in/out/up" are the somewhat figurative senses where the object is an abstraction, often a multiword NP, rather than a concrete noun or something easily substituted for by a pronoun. I don’t think this applies to "bring through" but I don’t think it’s a verb-particle construction in most cases; it’s “bring” followed by a through-PP.

    In the third graph it strikes me that usage and syntactic constraints drive the ratios. Constructions like "bring back lots/a lot/the lot" and "bring back memories/a memory/the memory" are so preferred that reversing the DP/particle order might sound odd—and not only odd but impossible if the objects are qualified with an of-PP.

    Really interesting stuff!

  9. Guy said,

    July 26, 2017 @ 2:33 pm

    A bit off-topic, but the display here of some people using "DP" and "NP" interchangeably really calls attention to the need for a term for these phrases that doesn't seem to commit the speaker to a theoretical position on the NP/DP debate.

  10. Yuval said,

    July 26, 2017 @ 3:39 pm

    What's the effect of the object being a pronoun?

  11. Guy said,

    July 26, 2017 @ 3:47 pm


    personal pronouns ("it", "him", "her", "them", "you", "me", "us") are very "light", as well as discourse-old, and generally must come before the particle absent very special circumstances such as contrastive emphasis.

    I looked it up in the dictionary
    *I looked up it in the dictionary

    I gave them back
    *I gave back them

    He told me off
    *he told off me

    I looked up "syzygy" in the dictionary
    I gave back the pens I borrowed
    He told off the rude people that were bothering us

  12. Frank Y. Gladney said,

    July 27, 2017 @ 12:36 am

    Postparticle position for the object sometimes emphasizes it, suggesting it is new information. The answer to the question, "What do you do when you find a horse in your bathtub?" is spoiled with the object after the particle ("You pull out the plug"). Preparticle position ("You pull the plug out") preserves the "horse" = "plug' equation.

  13. Mr Punch said,

    July 27, 2017 @ 10:53 am

    What strikes me is that "pick up" can be a phrasal verb, with multiple meanings. "I picked up the book from the table" means something quite specific; "I picked up the book in Istanbul" is pretty clear; "I picked up the book at the library" may be a little more ambiguous — but I'd say these are three different meanings.

  14. Timo said,

    July 27, 2017 @ 5:18 pm

    I'll say object > particle is marked (in comparison to particle > object):

    He gave up the attempt
    [where an attempt is referenced that isn't there afterwards]

    It brings about a change
    [where there is no change at first, but there is one afterwards]

    It brought the economy back
    [where the economy is backreferenced *and* it's there afterwards]

    This is rudimentary, of course, and I might be very wrong. If one was especially daring though, one could try and work on the criteria so that they account for this and the pronoun placement as well.

  15. David Scott Marley said,

    July 28, 2017 @ 11:24 am

    It seems to me that the reason "John put back it" sounds wrong is largely because the rhythm of the sentence leads us to put a stress on "it", and we usually don't like stressing pronouns. "John put back the book" sounds fine.

    Inventing a context where stressing the pronoun conveys additional information makes it sound acceptable, too, at least to my ear. "I'm confused, did John call back Marcia, or did he call back Marcia's brother?" "John called back *her*."

  16. Michael Watts said,

    July 28, 2017 @ 11:43 am

    In high school, I heard a classmate produce "put back on your skirt". I believe this was a production error; to me, the ordering "put your skirt back on" is required.

    This may be more of a constraint on how back interacts with put on than on phrasal verbs and their objects generally.

  17. Lai Ka Yau said,

    July 28, 2017 @ 2:06 pm

    @Guy: There was a discussion of this on Martin Haspelmath's Facebook profile a while ago. He suggested using 'nominal' in lieu of NP/DP and many linguists in the comments seemed to agree, though not necessarily for the same reason that Haspelmath suggested.

  18. Lai Ka Yau said,

    July 28, 2017 @ 2:16 pm

    I wonder if frequency has something to do with the lexical effects presented here. In the graphs here, judging by the widths of the CIs, it seems that the forms that prefer particle first are more frequent, and those preferring object first are less frequent. Of course there are other reasons for the widths of the CIs – verbs preferring object first tend to be nearer to 0.5, which means the standard error is greater – but even e.g. bring along, bring around, etc., had wider CIs than bring up, bring out, etc., even though the former set of phrasal verbs is farther away from 0.5.

  19. Robin Melnick said,

    July 28, 2017 @ 2:17 pm

    @Lai Ka: I suppose the challenge with 'nominal', at least in this context, is that for many that category also includes adjectives.

  20. DCP said,

    August 3, 2017 @ 11:49 am


    What about "look over", in the sense of "examine" as in: "I looked over the proposal and it seemed legit."

    Unlike "look it up" vs. "*look up it", placement of "it" after the particle in "look over" sounds fine to me:

    I looked it over, and it seems legit.
    I looked over it, and it seems legit.

    But something still seems a little funny about these; for example, a specifically human pronoun after the particle sounds odd:

    After the accident, I looked her over, and she seemed okay.
    *After the accident, I looked over her, and she seemed okay.

    (Again, adding contrastive emphasis seems to make these acceptable.)

RSS feed for comments on this post