I edited this question in response to Karlsson's paper, "Constraints on Multiple Center-Embedding of Clauses" (Journal of Linguistics 43 (2), 2007, 365-392), linked here: http://www.ling.helsinki.fi/~fkarlsso/ceb5.pdf .


Given that Piraha (and Warlpiri) do not allow embedded clauses, and Piraha does not allow recursion at all, clause embedding is not universal to all languages. As Karlsson persuasively argues, full blown recursion developed in the historical period. I will assume this here, and I will not repeat Karlsson's arguments.

In the Hebrew bible, there is no more than one embedded clause in any given sentence, consistent with Karlsson's bound for pre-literate societies, and there are grammar errors which suggest that the authors had a problem with complex sentences. As Karlsson explains, Greek and Latin writers developed full blown center-recursion explicitly somewhere around the 1st and 2nd centuries BC, and there is no evidence that recursion existed before then in any form


As recursive embedding becomes common, it is possible that speakers prefer constructions which allow embedded clauses to take on different roles with no internal modifications. If you say

  • I walked home-ward quickly

and you try to modify home to the embedded noun-phrase "the wide-open field where John slaughtered the goat", you say

  • I walked wide-open field-ward where John slaughtered the goat

this requires inserting a syllable in the middle of the noun-phrase "wide-open field where John slaughtered the goat". If you want to move the noun-phrase to the subject position of a different sentence, you need to scan the interior of the noun-phrase to remove the case-modifier:

  • The wide-open field where John slaughtered the goat is pretty.

If you have a stand-alone preposition, no scan is required

  • I walked to the wide open field where John slaughtered the goat

the words past "to" are exactly the same as when the noun-phrase is in the subject role. If the preposition can appear at the beginning of the phrase, it gives a clear "push" indicator for a mechanical parser, and it allows for trivial transformation of phrases to different roles.

compare the cased:

  • I walked store-from-ly the wide open sheep grazing field-ward where John slaughtered the goat.

Store from-ly means "from the store" in a case approximation in English.

  • I walked the wide open sheep grazing field-from-ly where John slaughtered the goat store-ward.

to the uncased:

  • I walked from the store to the wide open sheep grazing field where John slaughtered the goat.

  • I walked to the store from the wide open sheep grazing field where John slaughtered the goat.

There is no internal scan of the noun-phrases required in order to change their function. the words stay the same as the noun-phrase takes on different roles, only the preposition changes.

The case-introducing/case-removing noun-phrase scan is really annoying. It is computationally taxing, and I believe that it creates a linguistic pressure to remove cases from a language, and replace these with prepositions. When there are no cases, you have effortless noun-phrase embedding--- you just change the preposition, which always appears at the beginning.

I personally witnessed case-shedding events, in modern Hebrew. In the Bible, you would always say "Halachti le-beyto", "I went to-his-house" to mean "I went to his house", using a possessive marker on the word "bayit" (house). But modern Hebrew speaker will always say "Halachti la-bayit shelo", "I went to the house of him", precisely because the word "shel" can be used to embed, as in "Halachti la-bayit shel ha-ach hagadol sheli" "I went to the house of my big brother".

In addition to a possessive marker, which was shed, Biblical Hebrew also has a "to" case, so that it says "I walked Schem-ward" ("Halachti Schem-a"). Modern Hebrew, despite prescriptivist admonitions, also dropped this case entirely, so that all modern speakers use the recursion friendly: "I walked to shchem" ("Halachti le-Schem").

From my own native speaker intuition, I know that this is a consequence of the ubiquitous embedding in modern Hebrew. You don't say "I walked city-ward by the sea" in modern Hebrew, it is ungrammatical. you say "I walked to the city by the sea", with the exact same form as in English.

The Bible doesn't embed very much, and if it wanted to do this, it would say it in a pre-literate way that suggests recursion is completely alien to the author: "Halachti Schema, zu ha-'ir le-yad ha-yam"/"I walked city-ward, this is the city by the sea."

So the hypothesis is that recursive languages shed their cases as soon as most speakers begin to produce and transform multiply embedded sentences on a regular basis. This happens at different times for different languages.


This hypothesis makes the following predictions, and these are the questions I have:

  • Cases should be most diverse and pronounced in non-embedding languages, like Piraha. This is true of Piraha. Is it true of other non-embedding languages?
  • The most cased languages today should have a recent writing system, and the least cased languages should have an ancient writing system. For Russian and English, this works. Are there counterexamples?
  • Are there opposite events where the case system was strengthened? Do these correspond to a loss of literacy?
  • 9
    No, no and no. It's not true that Warlpiri has no embedded clauses. Finnish has recursion and embedding and a complex case system. There are very many non-literate societies that have languages without case systems. Finally, it is not 'the mainstream view of linguists that case-shedding is a kind of degeneration of language with time [etc]'. Commented Mar 3, 2012 at 13:32
  • 4
    @RonMaimon: I don't think that's what Qwertie (cool name btw!) meant. It is precisely the fact that the possessive 's in English is not the most typical example of a case (if any at all) that makes it hard to embed things with 's. It is dead simple with regular cases: poetarum Graecorum ex saeculo Periclis magni opera non placent mihi. Or das Haus des guten Mannes meiner Schwester.
    – Cerberus
    Commented Mar 3, 2012 at 20:46
  • 3
    @jlovegren: You're probably right: he is looking at the possessive 's, which is a clitic that comes after the entire noun phrase, which is one reason not to consider it a case. (Note: normally, cases are not limited to the head noun: any adjective or apposition to the head noun generally agrees with the case of the head noun.)
    – Cerberus
    Commented Mar 3, 2012 at 21:48
  • 4
    @RonMaimon It may be useful to look at WALS (World Atlas of Linguistic Typology). Here's WALS feature 49A, 'Number of cases'. In the chapter text and look at the map you'll see where case-marking languages are found and how many cases they have. There are many non-literate (or only recently literate) societies with no case marking. You'll also see languages which are Western, with long histories of literacy, and much case-marking. You could also look at WALS 23A to see how grammatical relations may be marked other than by case. Commented Mar 4, 2012 at 6:18
  • 3
    BTW, Biblical Hebrew did not have a complex case system. In fact, it did not have case marking, apart from the allative (or 'to/towards') that occurs in your example, and fossilised case marking on pronouns (much like what we still have in English). Case marking occurred in Proto-Semitic but was lost by Biblical times. Commented Mar 4, 2012 at 11:30

4 Answers 4


I have read the latest edited version of your question, and I think I understand what you mean now.

It is important to make a distinction between an ordinary suffix and a case ending (which could be considered a special kind of suffix). Consider the following two examples:

Magni viri liber est in mensa.

"The large man's book is on the table."

The Latin sentence means exactly the same as the English one. The word liber ("book") is the subject of the sentence, and so it is in the nominative; viri ("of the man") is in the genitive, as indicated by the case ending -i (the nominative would be vir). The adjective magni ("large") is also in the genitive (nominative magnus). The head word (viri) of the noun phrase (magni viri) determines the case, number, and sex of the whole noun phrase; the adjective must always be in the exact same case, number, and sex.

This is a central principle in most of the case-heavy languages, including Latin, Greek, and German. Perhaps it would be better to characterise them as heavy in noun inflection, as this includes number and sex too. This principle is called concord or agreement, as you may know.

Consider the contrast with the English phrase: only the head of the noun phrase, the noun itself, is marked (with 's). Because the English suffix does not involve concord, I would not call it a case ending. The principal remaining concord in English is between subject and verb (I am v. he is). How about Biblical Hebrew? Is only the head marked, or adjectives as well?

Can you see now how easy it is to add a relative clause to an inflected noun phrase?

Pater magnum virum in urbe captum, qui pius erat, servavit.

[Father large man in city having-been-captured, who pious was, saved.]

"Father saved a large man captured in the city, who was pious."

This time virum ("man") is in the accusative, because it is object. I have marked all words agreeing in case/number/sex with virum in bold. The word pater ("father") is marked as subject by the nominative. The finite verb of the main clause is servavit ("saved").

The parsing difficulties in your the wide open sheep grazing field-from-ly are not present in the Latin, simply because the first word of the noun phrase (magnum) is already marked as object. Imagine if the word the in your field phrase were already marked as "from-ly". It would look like this:

I walked the-from-ly wide-open-from-ly sheep-grazing-from-ly field-from-ly, where John slaughtered the goat, store-ward.

See how the parsing problem disappears? You don't need to hold your breath for the function of the noun phrase to appear.

German works the same way, and Ancient Greek too. The latter does have a few suffixes that serve as case endings but are only attached to the head word; and you are right, those can normally only be used when the noun phrase consists of a single word, no dependencies or embedding. But I can only think of two such suffixes, and they are rare and/or semi-archaic.

οἶκόνδε νέεσθαι

[house-ward to return]

The word οἶκός (/oi.kos/) means "house"; the suffix δε (/de/) means "towards"; νέεσθαι (/ne.e.stai/) means "to return". (The accusative ending -όν (/on/) is used because the suffix -δε requires the accusative.) So the whole means "to return home". This suffix cannot be used with complex noun phrases, as you suggested.

As to your suggestion that embedding a clause in the middle of another clause did not happen until the Hellenistic period, that is just not correct. It took me a whole two minutes to find a counter example, in lines 11 and 12 of the Odyssey. Homer's work comprises the oldest European literature in existence, probably composed in the 8th century BC.

ἔνθ᾽ ἄλλοι μὲν πάντες, ὅσοι φύγον αἰπὺν ὄλεθρον,

οἴκοι ἔσαν, [πόλεμόν τε πεφευγότες ἠδὲ θάλασσαν]

"then all others, whoever escaped sheer destruction,

were home, [having escaped the war and the sea]"

The word ὅσοι is a relative pronoun referring back to ἄλλοι μὲν πάντες, "all the others"; it introduces a subordinate clause with the finite verb φύγον, "escaped". The main clause has ἄλλοι μὲν πάντες as its subject, and ἔσαν ("were") as its finite verb.

The Homeric poems were meant to be recited out loud in the correct rhythm (metre). In fact, nearly all literature of Antiquity was meant to be read out loud up until the emergence of novels, around the first century AD. So people were able to use and understand (centrally) embedded clauses in speech throughout European history. I agree that they are generally less easy to parse than embedding at the end of a sentence, so there were obviously some restrictions, just as now; but central embedding was used. It is probably very, very old.

One thing to note is that relative pronouns were relatively new: it can be observed in Homer that relative pronouns were still in the process of developing out of demonstrative/personal pronouns of the third person (there never was a strong distinction between demonstrative and personal pronouns of the third person in Latin and Greek throughout Antiquity). However, participles are much older, as are conjunctions; and both can be used to embed verbs with complements just as well, and were in fact so used (e.g. in Homer).

  • I am not offended. I don't speak latin, so please provide a gloss, like you did in the comments. I am interested in multiple levels of recursion only, so I prefer highly embedded artificial examples, like "I know of Mary that she knows of John that he knows of Martha that she knows that I am not dead." I disagree with you from personal experience with languages that I happen to personally know, and Hebrew has some cases, so I know how they work. The issue is that word markers that muck up clauses don't play well with transformations.
    – Ron Maimon
    Commented Mar 4, 2012 at 0:49
  • 2
    @RonMaimon That kind of side-branch embedding can repeat many times and still be pretty comprehensible. I think centre-embedding is much more interesting, as well as harder to process--it's usually said that humans can't cope with more than three levels. Eg: 'This is the bus that the car that the professor that the girl kissed drove hit'. Commented Mar 4, 2012 at 5:31
  • 2
    @RonMaimon But surely centre embedding is more relevant to your mention of limits to embedding depth as it's much harder to comprehend deep centre embedding than deep side-branching embedding. Commented Mar 4, 2012 at 8:55
  • 3
    @RonMaimon No, that's incorrect. Humans cannot cope with arbitrarily deep centre embedding. This is a much studied area, see (eg) "A usage-based approach to recursion in sentence processing" by Christiansen and MacDonald, in "Language as a complex adaptive system". Also you'll find there mention of centre-embedding in Finnish, a strongly case--marking language. Commented Mar 4, 2012 at 21:42
  • 1
    It's not just the bracketed part--- an ancient Hebrew style for this sentiment would read like this: When the ruler of the feast had tasted the water, that water which was made wine, and the ruler knew not whence it was. But the servants, those servants who drew the water knew. Then the governor of the feast called....", it would not have the long flowing complex sentence structure as it does above, with modern implicit attachment rules. The bracketed part is nothing special, the whole of John is flowing and recursive, and later Hebrew texts like Ecclesiastes are recursive in similar ways.
    – Ron Maimon
    Commented Aug 10, 2015 at 22:05

The question deals with two ideas: the possibility of recursive embedding of clauses, and the use of case systems. We are asked to entertain a hypothesis that has the following basic parts:

  1. There was some kind of early state in prehistory at which no language had embedding, and where all languages had elaborate case systems.
  2. The advent of writing encouraged the development of embedding.
  3. Embedding developed at the expense of case-marking.
  4. All languages in the world are at some point in the continuum between having an elaborate case system and having recursive embedding.

To be able to begin to entertain this idea, we need to show that case marking and embedding are strategies which can serve the same purpose. I think that this is false.

Case-marking is a strategy for marking dependency relations between heads and dependents, where the marking falls on the dependent. Alternate ways of indicating these relations are by marking the head, or by marking neither, but using certain word order patterns to indicate different types of grammatical relations. So if a language lacks case-marking, then it would be reasonable to suppose that it is either a head-marking language, or it has relatively rigid constituent order. We could also suppose that case-marking languages will have more flexible constituent order, and will be less likely to mark grammatical relations. The choice between case-marking and some parallel strategy affects the organization of the language at the clause level.

Clause embedding, however, is a strategy that is reflected primarily at the level above the basic clause. When two clauses are linked, the linking strategy could involve coordination (no embedding), subordination (one clause is embedded in the other), or cosubordination (both clauses are embedded in some higher structure).

Basically, I can't see now how this theory has any a priori plausibility right now. I don't think it should be explored further until we can get an explanation for why, on linguistic grounds, case-marking as an organizational strategy in grammar should compete with, rather than work in conjunction with, clause embedding.

  • 1
    This is reasonable, +1, but I am not saying that cases compete with embeddings, or that they serve the same purpose. Rather, I am saying that embedding noun-clauses in a language that has noun cases is difficult, because you need to know how to case clauses. The different interpretations of nesting are not particularly relevant--- just use a simple noun-phrase and try to case it, and you'll see that it is not so simple, or if it is, that it requires some scanning to do a transformation.
    – Ron Maimon
    Commented Mar 4, 2012 at 0:55
  • 3
    @RonMaimon Yes, that's non-standard terminology. What you're calling a 'noun-clause' is normally called a noun phrase (or 'NP' for short). This is the term standardly used in linguistics (and in computational linguistics). It's weird to talk about making a NP accept a case marker. Rather, a NP is used in a particular grammatical role in a clause, and therefore requires the appropriate marking. In some languages this marking of the grammatical role is done by case inflections, in others by indexing on the verb, in still others by word order. Commented Mar 4, 2012 at 8:53
  • 3
    @RonMaimon your knowledge of how languages work seems to be at variance with that of most other participants. You are going to have to interpret your insights in terms of standard vocabulary or point people to suitable references when you employ non-standard terminology, lest your insights be lost on people who cannot understand them correctly.
    – user483
    Commented Mar 4, 2012 at 15:27
  • 3
    @RonMaimon sorry, I don't want to discourage you from participating (nor do I want to inflate this comment thread). "Noun-clause" is not a big deal, but what is eluding me is your theory of sentence processing, which I cannot recognize, and which most of your claims are hinging upon.
    – user483
    Commented Mar 4, 2012 at 15:47
  • 2
    @RonMaimon Closing is not censoring. It's a pause state that you can use at your advantage to improve your own question. I can't force you to edit it, as it is in your best interest. If you decide not to do it, I don't have much choice.
    – Alenanno
    Commented Mar 6, 2012 at 11:33

I don't think you have any real evidence that this "center-marking" is diachronically unstable. It also seems that you are totally unfamiliar with things such as case prefixes and prenomial relative clauses. I suggest you read grammars of Basque or Japanese, for example. (I'm not all that familiar with any case prefix languages, so I can't help you there.)

In a hypothetical SOV language with case suffixes and prenomial relative clauses, your example of

I walked store-from-ly the wide open sheep grazing field-ward where John slaughtered the goat.

might be glossed something like this:

1SG.NOM store-ABL John-NOM goat-ACC slaughtered where wide open sheep grazing field-ALL walked.

In the comments you stated that you didn't believe kaleissin that Basque works this way. I don't know it well enough to translate this sentence, but I assure you that it works this way (and in Japanese, too).

Likewise, in a hypothetical VSO language with case prefixes and postnomial relative clauses, it might be glossed like this:

walked 1SG.NOM ABL-store ALL-field grazing sheep open wide where slaughtered John-NOM goat-ACC.

Your problem is avoided entirely when the locus of case marking is on the opposite side of the head noun as relative clauses. This is fairly common. (Few languages have case prefixes, but that is irrelevant. Case prefixes are equivalent to prepositions for this purpose, as are case suffixes to postpositions.)

The other way to solve your problem is with case clitics. Rather than give you another hypothetical example, I will use English here. The possesive 's is a clitic that attaches to whole noun phrases. For example, you say "The Queen of England's crown" instead of "The Queen's of England crown", which you wouldn't like. There are languages that build their entire case systems like this. (In fact, I'm not entirely sure about this, but I think I remember reading that full case-concord systems like we are all used to from Latin, Greek, etc. are relatively uncommon, and case is applied only once to a whole noun phrase in many more languages.)


As I was commenting above, English has a possessive "case" marker that can apply to phrases and it exhibits an "allergy" to recursion.

On the other hand, if "cases" just refer to context-sensitive form variations like "be is am are was were", I see no clash with recursion there (likewise in Spanish, which has plenty of recursion but vastly more verb forms than English.) Esperanto has a simple two-noun-case system and a very productive system of affixes, but when you're used to it, it feels very natural even though the affixes can't mark phrases.

If we're simply talking about morphology that changes word meaning, I don't see a real issue there either. For example, in Esperanto I can say manĝaĵujo, a food container, with the "container" suffix -ujo, but ruĝa fruktujo "red fruit container" should mean "a fruit container that is red", not "a container of red fruit"... basically what I'm getting at is that affixes ordinarily apply only to words, not phrases, and this is no different from English affixes. Yet this doesn't feel like an incompatibility between recursion and morphology, merely a limitation on the complexity of morphological changes. If I want to say "container of red fruit" then I simply give up on affixes and say "ujo de ruĝa frukto", "container of red fruit".

Now I can only talk about the languages I know about (English, Spanish and Esperanto, all in the same family I'm afraid) but it seems to me that word-level variations such as case markers and affixes can be thought of as providing shortcuts for standalone words like "of", "that", "female" (stewardess) etc. Yes, word-level changes don't play well with recursion but that doesn't necessarily encourage them to disappear. As long as they provide some utility that the stand-alone alternatives don't, they will tend to stay. For example, English will keep "'s" for the foreseeable future because it is more concise than "of" (and because "of" is more ambiguous, having more meanings than just the possessive); and this would probably be true even if "'s" were strictly applicable onto to words and not phrases. If Esperanto ever became really popular, the accusative case "-n" might disappear for most sentences, yet would probably remain whereever it provides information concisely, e.g. "hejmen" = homeward, to home.

Anyway, I'm a linguistic 'newb' myself. I suspect there are lots of forces acting on languages that neither of us appreciate very well, forces that probably outweigh the hypothesis of this question.

  • The reliance on Esperanto (an artificial language) makes me nervous, as it did not develop organically. It doesn't seem relevant. Commented Mar 3, 2012 at 21:30
  • +1: Your point about the ambiguity of "of" is important. If you use enough cases, I walked store window-with-ish hat-with-ly (I walked to the store with a hat with a window) loses the ambiguity of whether the "with a hat" applies to me or to the store. This is one of the central annoyances in natural language formal descriptions--- the verb argument/adverb-like-phrases and the subject modifiers/adjective-like-phrases commute and are usually all ambiguous.
    – Ron Maimon
    Commented Mar 8, 2012 at 11:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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