Consider the following extremely melodramatic story fragment (the sentences in the story, and its melodramatic character, set up context and motivation for the brain to parse an otherwise hard-to-parse sentence).

"Linda was my mother's partner, who my mother, God bless her, admired so much. Linda, along with her husband, a ruggedly handsome adventurer, would jaunt around the world, writing travel pieces. In Ecuador, as he and Linda lay in a sun-bleached hammock sweating with malaria, he sent me a letter by mail. Inside was a photo and a short note explaining their situation, and a second envelope, with the strictest instruction not to open, except upon news of their death.

They were supposed to return today, but he was alone at the Airport. Oh Linda! Through his tears, he assured me that he would read that secret letter aloud to me tonight, at my bedside. You must have already guessed that I am in love with him. He is the man that Linda, the woman my mother admired, married.

"But you took the letter, took it upstairs. You didn't let the man that Linda, the woman my mother, God bless her, admired married read it out loud to me. Why?

"You know you always stutter when you lie. I am not accusing you of anything, but if while you tell me what you took the letter that I wanted the man that Linda, the woman, my mother, God bless her, admired, married to read to me upstairs for in your own words, you start stuttering, I will never forgive you!"

The only point of this ridiculous tale is to set up the final sentence. The final sentence has 7 separate layers of center embeddings.

  • "I am not accusing you of anything, but only if SENTENCE I will never forgive you" is the outer context of the relevant recursion.

  • "while SENTENCE you start stuttering" is embedded.

  • "you tell me SENTENCE in your own words".

  • The middle sentence is "what you took the letter ... upstairs for." This is a variation on a well known embedded construction observed in a 7 or 8 year old child: "What did you take the book that I wanted to be read to out of up for?".

  • "that I wanted the man ... to read to me" is another level

  • "Linda ... married" is another level

  • "the woman my mother ... admired"

  • God bless her (that's the deepest)


I am curious about the psychological tests determining the depth of center embedding. The limit of 3 seems to be for homogenous constructions, where the repeated embedding is of the same kind in all the deepening examples.

Here is a simple sentence that violates the 3 center-embedding limit:

  • If you tell me what you took the letter that I wanted you to read from upstairs for in your own words I will forgive you.

But the sentence from the melodrama is:

  • I am not accusing you of anything, but if while you tell me what you took the letter that I wanted the man that the woman my mother, God bless her, admired married to read from upstairs for in your own words you start to stutter, I will never forgive you!

which is center embedded at least 7 times, so in terms of embedding depth, it is analogous to:

  • The car that the bus that the man that the woman that the child that the bunny that the carrot that the top was snipped off of, was fed to, hopped at, talked to, smiled at, drove, hit.

But I don't think the preceding sentence is anywhere near as difficult to parse. Are the center-embedding limits trustworthy? Am I fooling myself by reading an incomprehensible sentence too many times? Or is it only a certain type of center embedding that is limited to 3 levels?


Karlsson's 2007 paper [PDF] "Constraints on multiple center-embedding of clauses" I believe may be the most relevant reference here. His key thesis from that paper is:

The maximal degree of center-embedding in written language is three. In spoken language, multiple center-embedding is practically absent.

But his entire analysis is interesting and I recommend reading the whole thing.

(Most of Karlsson's work centers around constraint theory; e.g. "Constraints on multiple initial embedding of clauses.")

Wikipedia even has a couple reasonable sentences on this topic:

...center embedding refers to the process of embedding a phrase in the middle of another phrase of the same type. This often leads to difficulty with parsing which would be difficult to explain on grammatical grounds alone. [...] An interesting theoretical point is that sentences with multiple center embedding are grammatical, but unacceptable.


There probably is some sort of practice effect here. But I suspect it goes beyond that.

I think it's telling that you had to set that godawful multiply-embedded sentence up in a story. And what's more, you had to build it up throughout the story — starting with the deepest two clauses, "who my mother, God bless her, admired," and adding on more layers from there. Even still, I find it only vaguely comprehensible. You say you find it fully comprehensible, and you speculate that it's because you've read the full sentence too many times. I think there may be another reason — I think you've read the inner clauses too many times in unembedded form, and you've gotten yourself to a point where you can feel like you understand those inner clauses without actually having to parse them or pay attention to their structure.

Here's a hypothesis: the constraint against quadruple center embedding only applies to constituents that the hearer (or reader) actually tries to parse. If the hearer treats a constituent as an unanalyzable unit, then that unit can then be embedded two or three times without confusion, even if it actually has inner structure that the hearer is ignoring in treating it that way.

Suppose I tell you the following:

(1) [Sentences [that bop-squee-fim-blork-pings utter] are easier to understand.]

I'm guessing you find the English part of that sentence to be perfectly parseable — though, of course, you'd have to know what bop-squee-fim-blork-pings meant in order to fully interpret it. Now suppose I were to tell you that bop-squee-fim-blork-pings was a doubly center-embedded structure in some foreign language; maybe it means "women that people you've met dislike." Would you take that as evidence that you had just managed to parse a quadruply center embedded sentence? Probably not — since, after all, you didn't actually have to parse bop-squee-fim-blork-pings when you read (1). Indeed, you couldn't have parsed it if you'd tried, so your brain didn't expend any effort at all on trying. That particular rabbit hole wasn't even available for you to fall down.

Now suppose that you've heard the phrase women that people you've met dislike often enough that you've just started treating it as a single unit instead of actually parsing it on hearing it. Maybe you've even created a lexical entry for womenthatpeopleyouvemetdislike as if it were a single compound noun of some sort. In that case, my suspicion is that (2) would be just as parseable as (1) for you. But crucially, at that point, you wouldn't be parsing the quadruply-embedded structure in (2b), you'd be parsing the much simpler structure in (2a).

(2) Sentences that womenthatpeopleyouvemetdislike utter are easier to understand.

(2a) [Sentences [that womenthatpeopleyouvemetdislike utter] are easier to understand.]

(2b) [Sentences [that women [that people [you've met] dislike] utter] are easier to understand.]

I think this is what you're doing, incrementally, when you work yourself up to reading the final sentence in your story. You convince yourself to treat mymothergodblessher as a single unit; then you convince yourself to treat thewomanmymothergodblessheradmires as a single unit; and so on up. If you hadn't been rehearsing those smaller units and learning to read them without really parsing them, I'm betting you'd be entirely baffled by that seven-layer sentence when you encountered it.

  • Of course you are right, but you aren't telling me anything new--- I purposefully used the story to get you used to the constituents, to see if you would be able to parse the sentence when they were set up like that. The embedding of the constituents was word-by-word without any modification required, something which is a special property of case-free English. I wouldn't have been able to do the set-up as clearly in latin, because the various phrases would not be repeated verbatim. The point is that I can scan and read the sentence without effort, once it's set up.
    – Ron Maimon
    Mar 7 '12 at 20:34
  • +1: On second reading, I think you have a point--- but no amount of set up helps with the "bus,child, bunny" example. I came up with the sentence first, I made up the story to make sure the sentence would be parseable. Further, when I insert a grammar error into the sentence in one of the levels, I can intuitively catch it, with perhaps the exception of take ... upstairs for (which is the most difficult embedding structure for me psychologically). If I replace "upstairs for" by "to upstairs for" I sometimes can't catch the error, because the brain is too busy with the other recursion levels.
    – Ron Maimon
    Mar 7 '12 at 21:09
  • I wasn't at all baffled by the 6 layer sentence "If while you tell me what you took the letter that the man the woman my mother admired married upstairs for in your own words you start to stutter I won't forgive you", but I wanted a 7th layer. I read a bunch of 19th century books, so I am used to this type of recursion. Your final example has only 3 levels of recursion (calling toplevel 0, as I did), mine has seven, which is more than twice as deep, but only about equally hard to parse.
    – Ron Maimon
    Mar 8 '12 at 0:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.