Another question

[casey61694]

Is case an inevitable part of natural language? Has its evolution ever been traced? Does its "emergence" usually follow the "emergence" of a pronominal system?

[MalFet]

Welcome to the forum!

Why would case be inevitable? There are lots of languages without any case marking.

[casey61694]

It just seems to me that the idea underlying " case " is always present in language, explicit in some languages and implicit in others. So it is just a matter of time before this idea is inevitably expressed explicitly ( to clarify relationships between verb or action participants ). Are the languages without case marking extremely well-structured to erase any possible ambiguity? 

[freknu]

Just look at English — aside from pronouns and the possessive (could be argued to be a suffix rather than a case, similar to Swedish) there is virtually no cases in English. While languages such as Icelandic and Russian have several cases, Yue and Japanese, AFAIK, have no cases at all, and certain African languages have a metric tonne of cases.

[Daniel]

Case is not found in every language. Sometimes it appears (grammaticalizes) and sometimes it is lost. In that sense, it is not universal as a morphological category so it is not "inevitable". (It is possible that in the long history of languages they'll eventually cycle through it-- it's a relatively common phenomenon, so a language might, every 5,000 or 20,000 years, end up cycling back through having case. In that sense it might be "inevitable", but I don't think that's what you mean.)

On the other hand, if you're asking if it is "not a weird part of language" then that's probably true-- it's found in a lot of places around the world.

However, case is just one instance of marking participant roles in a sentence. Another option is word order as in English. And another is agreement on the verb (where, for example, Hungarian verbs agree with subjects and objects). Case is a sort of dependent marking (rather than head marking, as agreement would be) and, obviously, an overt morphological indication as opposed to just syntax.

Historically speaking, case always (almost always?) originates from post-positions. Interestingly, prepositions rarely become case markers (I don't know of any examples, but there may be some). So it often occurs in SOV languages (verb-final, head-final). In Japanese, for example, the "particles" could be considered case markers or postpositions. Over time, the postpositions become morphologically integrated into the nouns, and you get case. Then it may also extend to adjectives and so forth through agreement. This is just one aspect of a major cycle in which languages go from isolating (no/little morphology, as in Chinese, or, in some ways, English today) to agglutinative (regular morphology, just "glued" together, like Turkish, Swahili) to inflectional/fusional (irregular and non-isomorphic [many-meanings-to-one-form] morphology, like Latin, German, etc.). Then the inflections are lost, as happened from Old English to Modern English for the most part, and you end up cycling again and again over tens of thousands of years.

That's the answer from a typological/descriptive/morphological point of view.

But another answer would come from syntactic theory (specifically Generativism), which I think might be what you're thinking about. In many versions of syntactic theory, case is considered an inherent part of language. Case checking, the assignment of theta roles, and so forth are natural and ubiquitous parts of the compositional/computational system. Some languages overtly manifest morphological encoding for case, but all of them are believed to use case in the derivation of sentences-- for example, structural nominative case must be assigned to the subject or the sentence is ungrammatical. This can explain things like the subject moving "up" to get to where case can be assigned, and so forth. In this sense, case is a structural relationship. (Note that structural case is different from inherent case-- many cases, basically other than nominative and accusative, are not structural. They're just inherent, part of the semantics. That's true in a language like Finnish with 15+ cases.)

But that's just one kind of theory.



So in general the word "case" is ambiguous between structural case in Generative theories, and overt morphological case marking.

[casey61694]

Wow. Thank you all for making this an awesome resource for amateur linguists! I'll definitely be using this site regularly! Hopefully, in time, I'll be able to help out others and start some interesting talks.

[casey61694]

I hope it's okay that I post more random questions here. My latest question is: why do embedded/indirect questions seem to break down grammatically after two "embeddings"? For example, I cannot think of an case in which a question is embedded inside of another embedded question which is in turn embedded inside of another question. I cannot get beyond a "double" embedded sentence like "I wonder how you know what I'm bringing" or "I asked why you wondered whether we would be late". Why is this? Why only two?
I appreciate any help. Thank you!

[freknu]

I hope it's okay that I post more random questions here. My latest question is: why do embedded/indirect questions seem to break down grammatically after two "embeddings"? For example, I cannot think of an case in which a question is embedded inside of another embedded question which is in turn embedded inside of another question. I cannot get beyond a "double" embedded sentence like "I wonder how you know what I'm bringing" or "I asked why you wondered whether we would be late". Why is this? Why only two?
I appreciate any help. Thank you!

"I asked why you wondered whether they knew if they had foreknowledge of whether it had any probability of explaining whether what happened was in any way indicative of whether the recent events where the result of earlier problems or whether it begs the question of whether there was a problem to begin with?"

Hm... how many would that be?

[casey61694]

Where is your respect for the linguist noobie?  This does help me though.  Thank you freknu!

[Daniel]

But freknu, do you understand that sentence? I don't. As a syntactician I can analyze it as grammatical and maybe jot down some notes to keep up then eventually translate it into real English... but I can't process it. It's possible that if you pronounced it with just the right intonation I'd be able to follow, but I think that's going to be tricky

The distinction is between competence (where deep embeddings like that are grammatical) and performance (where they are unacceptable and/or incomprehensible). (Many linguists make that distinction, on the assumption that there is a difference between what we know and how that translates into usage/behavior. Some would say that any sentence we can't use is ungrammatical, though and that everything is just performance.)

So the explanation is that while syntactic structure is infinitely recursive, after a few levels, in some cases, we lose track of what's going on and get confused.

The classic example of this is center embedding as exemplified by:
The cat ran.
The cat the dog chase ran.
?The cat the dog the rat bit chased ran.
???The cat the dog the rat the bird saw bit chased ran.
...

In fact, I think I saw experimental evidence that speakers prefer this sentence on first glance:
The cat the dog the rat bit ran.
It's actually ungrammatical, because there aren't enough verbs for the subjects, but we're terrible at parsing deep center embeddings.

The question then is: why is there a limit? Is there a limit to grammatical structure? You can only embed twice? three times? Or is there some limit to our processing ability. Generally it's attributed to the latter, although some would argue it's strange to say that we could potentially parse 100 levels of embedding "in theory", saying there's some actual limit to grammar. Interesting questions there.

One argument I've heard (recently mentioned to me by a colleague) is that this may be working memory-- we can only remember so many things, and center embedding makes you remember more things-- the cat, the dog, the rat, the bird...-- and at some point we will get lost. But other kinds of sentences allow you to package the information together (such as conjunctions-- "the cat and the dog and the rat and the bird...") so you won't get as confused, just using one unit of memory for the whole thing. So the argument would be that the grammatical structure "forks" sometimes and that means you're using more memory, and sometimes merges back together, at which point you're using less. If you reach the maximum (perhaps 4 units of memory or so) you'll get confused.


Now, on the other hand, there are also some specific restrictions, sometimes called or including "island constraints", on what can be questioned one. So that isn't about "only two" but about the specific meaning of the adverbial being questioned. Consider this odd sentence:
How did you wonder [whether] I would fix the car?
Can "how" refer to the embedded clause? vs:
What did you wonder whether I would fix?

In that case, it depends on the structure, not levels of embedding (though embedding, at all [but not counting levels], could have an effect).

[casey61694]

Forgive me if I ask or talk about something you already covered. I tried to get as much out of what you said as I could. 

In exploring "multiple" embedded questions, I found that you have to be mindful of the order of the interrogative [spec, CP]'s you use since the first has the widest impact on the rest. The second interrogative [spec, CP] has the second-widest impact on the rest and so on. The first interrogative [spec, CP] should therefore be general and not have a specific subject/object role in the following clause (like "why", "whether" ...). After I discovered that, I got "I asked if she wonders why I memorize what day people are born on".

Even if we could parse 100 levels of compounded embeddings, I think most would choose to be economical and break it up. Would it even be possible to create a program to extract meaning out of something so complicated? How would you go about it?

Also, could you expand a little on island constraints?

Sorry if this is really disorganized.  I'm ready for sleep.

[Daniel]

I'm not sure what you mean by "widest impact". Formalizing this might help to answer your own question, actually.
According to generative grammar (and in general most approaches which assume infinite recursion) there is no inherent difference between something at the first level of embedding and the fifth or 100th. Sentences are either grammatical or not. The only explanation that could be compatible with what you're saying would be something about processing-- certainly the first word we hear in a sentence could bias us against the rest. If that's all it is, I don't think it's a very deep issue. But if it's something more, it might be important.

Even if we could parse 100 levels of compounded embeddings, I think most would choose to be economical and break it up. Would it even be possible to create a program to extract meaning out of something so complicated? How would you go about it?

If you could write a program that parses a few levels (say, up to 5) recursively, and you gave it enough memory and processing power, then yes, certainly it could do 100 or 1,000,000, or whatever.

Also, could you expand a little on island constraints?

Lots to read about that, with a very long history. Some of the early work that might interest you can be found in Ross 1967:
Ross, J. R. (1967). Constraints on Variables in Syntax (Ph.D dissertation). Massachusetts Institute of Technology, Cambridge, MA. [You can find a free-to-view PDF on the MIT website.]
There will also be a ton of information online.
https://www.google.com/search?q=island+constraints

[freknu]

But freknu, do you understand that sentence?

Of course, I can, silly... :\ *scratches head*

There is also nothing wrong with "the cat that the dog chased that the rat bit, ran", then again, maybe native English speakers don't use structures like that. I might just be using my native logic in speaking non-native English.

[jkpate]

Even if we could parse 100 levels of compounded embeddings, I think most would choose to be economical and break it up. Would it even be possible to create a program to extract meaning out of something so complicated? How would you go about it?

If you could write a program that parses a few levels (say, up to 5) recursively, and you gave it enough memory and processing power, then yes, certainly it could do 100 or 1,000,000, or whatever.

Just to follow up on this, one reason center-embedding has attracted so much attention is that it seems like it should be easy from a computational perspective. We know how to write grammars and parsers that in principle handle unlimited center embedding, but our computer programs struggle with other things that are easy for humans (like co-reference resolution).

djr33 is right that the general feeling among computational people is that it has something to do with working memory and incremental processing. Computational work on incremental parsers (parsers that build structure word-by-word, left-to-right) has found that some parsing strategies involve extra work for center-embedded structures, but not others. For example, if you apply a left-corner transform to a CFG so it can be parsed incrementally by a top-down parser, the only structures that increase stack depth are (pre-transform) structures with the zig-zag pattern of center-embeddings. If there is some kind of hard limit on stack depth, such a parser fails only on sentences with center embeddings. You may also be interested in reading a more recent paper on this topic that uses the connection between left-corner parsing and transition-based parsing (probably the dominant paradigm for incremental parsing) to build parsing technology for exploring memory limitations and center-embedding cross-linguistically. I can provide a couple more references along these general lines if you're interested in a probabilistic perspective, although they focus more on memory limitations and garden paths rather than center embedding specifically.

[casey61694]

Thank you jkpate! I really appreciate your input, but it might be a while before I'm ready to read and actually understand the papers you posted. I'm not there yet.

This is what I mean by "widest impact":
A sentence like "I know [whom you will ask] or [whom to ask]" can, I feel, be replaced by "I know [him/her]". This tells me that the CP "whom you will ask" or "whom to ask" is basically expressed in terms of the [spec, CP]. If something is embedded in this CP (like another CP), it too will be expressed in terms of the first [spec, CP] (and so on). Is that basically right?