Do Modern Grammar Theories fall short in explaining Free Word Order?

Question

Here's my childish challenge to generative grammar:

Could anyone give me an analysis of Russian sentence Мама мыла раму. (Mom washed the (window) frame.) from the point of view of modern grammar theories (Generative Grammar, Head-First Grammar, Construction Grammar, etc.)?

The background: The sentence is learned by the Russians in their elementary school as an example of the free word order in Russian. Indeed, since the subject and the direct object are morphologically marked (Мама мыла раму.) the sentence is understandable for all the possible word permutations: SVO, SOV, OVS, OSV, VOS, VSO.

SVO may look like a more natural word order in the stand alone phrase, but one could easily think of the contexts where any of the alternative word orders will be more appropriate (topicalization?).

I assume that the analysis in terms of Transformational Generative Grammar (TGG) would postulate a derivation tree based on the SVO order [S [NP Мама] [VP мыла [NP раму]]], and then add the permutation rules to accommodate the alternative word orders. Hence is my critique of TGG:

• Occam's razor: the explanation seems by far more complicated than the intuitively obvious relations between the verb, the nominative subject and the accusative object.
• There is no obvious reason to postulate one word order in comparison to others.
• Postulating an initial word order breaks the "universal" nature of TGG, as it is more suitable for the analysis of languages with a fixed word order, such as English, than for synthetic languages like Russian (German seems like an intermediate case, where the subject and the object can be swapped, but the verb is always second.)

As I mentioned previously I am interested in seeing the analysis done by somebody knowledgeable. If my reasoning seems naïve - the more I would like to learn why/where.

Note Thank you for the interesting feedback. However the question itself is a request for actual analysis of a typical SVO sentence in an inflected language, such as Russian. And not necessarily from the point of view of TGG.

Further challenge Just to spice the things up, let us add a word: "Мама мыла раму мылом." (Mom washed the frame with a soap.)

Answer 1

I base my analysis loosely on the approaches of Croft (2001) and Diessel (2019).

Under a constructionist approach, a construction is sign, i.e. a form-meaning pairing, and within Diessel's approach specifically, a sign is an association network. There are three relations that can hold within a sign:

• Symbolic relations - associations between form and meaning
• Sequential relations - associations between elements that appear sequentially.
• Taxonomic relations - associations between constructional schemas at different levels of concreteness

In Croft's approach, there are additionally semantic relations, which are between elements different components of a meaning. There are situations where these are absolutely necessary, though I think we can get away with omitting them for the current set of sentences.

I will also note that this analysis is very unsatisfying, as we only have six sentences available, which is not nearly enough for a satisfying analysis (see my comments to the OP). Fortunately, CxG approaches allow us to focus on the present construction and gloss over some of the stuff that cannot be gleaned from the present data (e.g. the presence of a cross-construction Subject, or the parts of speech of Nouns and Verbs) which are thought to be emergent from facts about particular constructions. In most other approaches, these notions are thought to be basic and necessary for the description of specific constructions (which is why I'm not using them).

Here is the network I've drawn:

At the top level is the transitive clause. You should notice that there are no sequential links, so the three components' orders are not yet known. At the second level, I have put all six permutations available. You should notice that the link from the second to the third component is stronger than from the first to the second. This is because it should be easier predict the third from the second than the second from the first. At the third level, I have put in the schemas at the level of morphology. If there are e.g. quirky subjects in Russian, there could be more than one construction inherited by each box at the second level. I put in the lexical items on the bottom level.

I've made lots of simplifcations though, including:

• 'Mum' and 'frame' should be linked to the root, not the whole word. I was too lazy to create even smaller boxes within the box (and it would have been hard to see anyway.)
• Agent, patient, predicator should still be inherited at the bottom level.

In addition, as I focused only on a single construction, there are cross-constructional relations not represented:

• 'Mum', 'wash' and 'frame' should in turn be linked to a network of semantically associated words, e.g. 'window' for 'frame' - i.e. lexical relationships in Diessel's framework.
• I did not put in the 'lateral' associations between constructions at the same level, or constructional relations in Diessel's framework
• Some words are more likely to be put in a slot than others. I did not represent filler-slot relationships in this diagram either, which is necessary to capture facts like e.g. 'The frame washed mum' is much less likely than 'mum washed the frame'.

Croft, William. 2001. Radical construction grammar: syntactic theory in typological perspective. Oxford: Oxford Univ. Press.

Diessel, Holger. 2019. The grammar network. Cambridge University Press.

Answer 2

You say you assume that TGG assumes underlying SVO order for all languages. Why do you assume this? I don't recall any TGG linguist ever proposing such a thing (and I've been around awhile). I just did a quick google search, and the closest I could come was a definition by Haj Ross of "the universal base hypothesis" that makes base rules universal except for the order of sister constituents. And I don't think Ross was actually proposing this -- it was a definition.

You must think American linguists are really dumb.

Answer 3

Good points. Exactly this criticism led to the creation of other frameworks in the 1970s and 1980s. In LFG, say, your example sentence would be analysed by the (exocetric) rule S -> X, where X is a metasymbol standing for both NP and V(P). Grammatical relations are captured in a separate structure (an AVM in LFG). The AVM for “mama” would be the value of the attribute ‘subject’ in the predicate’s AVM etc. In this kind of framework, there’s a phrase-based structure for word order and a dependency structure for grammatical functions. This approach is called ‘constraint-based’ because the AVMs are created in the course of parsing by applying constraints on morphological and syntactic features of the constituents.

Answer 4

Generative grammar is supposed to be universal, so it must be that any language can be described using the same structure. The relations between words that are intuitive in Russian are less important in languages that don't mark cases, like English.

Generative grammar is definitely more elegant in a language like English, with strict word order, than in Russian, with free word order. This is the downside of a universal theory. Whether a universal theory is better, or truer, than a particular model suited to each language is an open question.

Answer 5

I just want to clarify that there is no preferred order.
Spoken language uses a serial channel, but it represents a tree (actually a net), (which is necessarily 2D and approaching it from different directions exposes the nodes in more than 2 ways).
Synthetic languages use morphemes to decline the nodes, and analytic languages use order, but there is nothing that prevents a language from using both methods (other than Zipfian waste).
Note that there is a third method: a single added declension for each analytic order, but no NatLang seems to recognize that.

(response to comment)
The major division of a sentence is into subject and predicate and there is no preferred order there. The predicate may be divided into verbal and adverbial phrases (I include the direct object as a type of adverbial) and, again, there is no preferred order.
Within these phrases (subject, verb, object/adverbial) there are lexical items (things, qualities, and actions - the open classes), and functional items (relations [between things, which may be lexical or deictic] - the closed classes, often reduced to morphemes), and again there is no preferred order, but there is a tendency (within a given language) to a consistent order; ie, a functional item will typically precede, or typically follow, the lexical item. Here it may be conjectured that the two classes (open vs. closed) may be stored differently in the brain and so there is some efficiency in a consistent way of constructing a 1D string for speaking, resulting in head first vs. head last languages. This ordering within phrases can look like a preference between phrases (such as where a copula has no lexical verb and an object has no adverbial head), but this order is illusory, and there can be head last languages with SVO order.

Answer 6

It would be difficult to analyze the Russian language from the viewpoint of GG since it is synthetic, not analytical language. However, it is, of course, possible, yet you must not forget to mark the inflections. In more complicated sentences, it would be very difficult I'm afraid.

Some kind of dependency grammar would be better for the analysis of synthetic languages with rather free word-order.