Many of the world's natural languages, and some conlangs, have a free word order, so that the words can be put in any order with it still making sense.
If they can't use word order to indicate the structure, what strategies do they use instead? Have any conlangs used strategies not found in natlangs?
It should be noted that in addition to relying heavily on one strategy, it's also possible to mix strategies is various different ways, relying on different strategies to back each other up. For example in Fore(Kainantu-Goroka(TNG), PNG), there is a hierarchy like this, where if each strategy fails, the one lower in the list can be relied on. I'm putting "case" in quotation marks, because the "ergative" case marker in Fore is not really a case-marker as much as it is an often optional and occasionally prohibited derivational marker:
Point 2, "animacy" means that referents higher on the animacy hierarchy are interpreted as being more likely to be agents, as such the NPs in a sentence like aebá nanita: yaga: amiye he pig food 3sg:gave_to:3sg "he gave the pig food" can be freely reordered without a change in truth value, despite the fact that the verbal agreement is wholly inadequate to disambiguate the roles of the participants. Only in the few instances where the other strategies fail is word order a primary disambiguative device.
There are some languages where a primary disambiguation strategy is marking on the verb whether such a hierarchy a broken or not, called direct/inverse marking (this can disambiguate situations with two 3rd persons of equal animacy as well by considering e.g. topicality or obviation).
The way these languages do this is with inflections. Nouns, for example, can be declined to show cases, which tell the speaker things about what they are doing. Verbs can be used to show who the subject is, and the tense and mood and stuff. For example, in English we would say:
The boy loves the girl
However, in Latin (I don't know any other languages with cases well enough for this answer) we would say:
Puer puellam amat.
There's a lot of info here: the first word (puer), meaning "boy" is in the nominative case, meaning it is the subject of the sentence. Puellam, girl, is in the accusative, meaning it is the direct object. In addition, the verb, amat, has the 3rd person singular present indicative ending -t. In this case this isn't enough to tell us who the subject is, as both the nouns are 3rd person singular, but coupled with the nominative, we can easily figure out who is loving whom. Therefore, the same thing can be expressed as:
Puellam amat puer
or
Amat puer puellam
or any other combination you'd like, and the listeners or readers would understand because of the cases of the nouns. Latin has six or seven cases (depending), but some languages have more. Finnish has 15, Hungarian has 18; Wikipedia has a whole bunch of cases that appear, all to free up the word order.
Because of these, many prepositions can be omitted entirely, verbs can use inflections to show the subject and stuff.
Adjectives that modify nouns must agree in case with the nouns--so a reader can easily infer their meaning from what case they are in. Because of these, word order having any grammatical function is unnecessary and redundant.
As another example, my own conlang Simean has four cases. Here is a sentence with all of them:
Nuler loyile shuir edeï sefmou.
Translated, this is: "The soldier (Nominative) is giving the king (Dative) of the land (Genitive) a book (Accusative)." The verb, loyile, is the stem loy- plus the imperfect affix -i- plus the third person singular -le. If instead, we wanted to say "I am giving the king of the land a book," we could take out sefmou, soldier, and we wouldn't have to add in the word for I because of the verb inflection. Instead, the subject could be expressed through the verb. Instead of the -le we would use -ë, and the listeners would know what we were talking about--who gave whom what and all that.
Technically, any one of these words could be moved somewhere else--the only thing in Simean that dictates word order is convention.
There are three main strategies for indicating grammatical relations in languages with free word orders. It is common for languages to have more than one of these, and to my knowledge all free word order natural languages have either case or verbal agreement.
Case refers to grammatical markers attached to nouns which indicate the noun's role in the sentence. Some languages have a very large number of cases, but in general you can expect such a language to have at least two cases: nominative and accusative, or ergative and absolutive. Here is an example from Walmajarri:
parri-ngu pa manga-Ø nyanya
boy-subj AUX girl-obj saw
'The boy saw the girl.'
parri-Ø pa nyanya manga-ngu
boy-obj AUX saw girl-subj
'The girl saw the boy.'
In Walmajarri ngu is the ergative case marker. Walmajarri has a null absolutive case marker, shown here as Ø.
The second strategy is agreement, where markers are attached to the verb to give information about the subject and object. Depending on the language, these markers might indicate the person, number, or gender of each noun. Here's an example, again from Walmajarri:
nayili nya-ka-kyaka-ji-pila
north see-IRR-REDUP-1sgO-3dlS
'You two watch out for me in the north.'
Here the verb has two suffixes (actually clitics), -ji and -pila, which mean a 1st person singular object, and a 3rd person dual subject. While it is most common for these verbal agreement markers to be attached to the verbs, in some languages they are attached to other elements in the sentence. In Walmajarri the most common sentence form actually has these markers attaching to a modal auxiliary:
ngayirta ma-rnalu majurra kang-ka-rla
NEG AUX-1plincS matches carry-IRR-PAST
'We didn't take matches.'
Here ma is the unmarked modal auxiliary, to which -rnalu is attached, the marker for a 1st person plural inclusive subject. Matches is a third person singular object and so has a null marker.
The third strategy is incorporation, where a language productively forms compound words where the verb includes the object. English doesn't do this very productively - our compounds are more like individual set combinations. So in the verb babysit the object baby is incorporated, but the meaning of the compound is not obviously derived from its component parts. In incorporating languages this process happens productively with the components often keeping their normal sense. Here is an example from Nuu-chah-nulth:
maḥt'a-ʔa-mit-ʔiš čakup
house-buy-past-3ind man
'A man bought a house.'
ʔu-ʔaa-mit-ʔiš maḥt'ii čakup
Ø-buy-past-3.ind house man
'A man bought a house.'
Nuu-chah-nulth is called an obligatory incorporating language because if the object is not incorporated a dummy morpheme ʔu must be used.
Now while incorporation is used in many languages, we have to be careful about making generalisations as each language incorporates nouns according to its own rules. Some incorporate adjectives, and some don't. Or for example, in Nahuatl an independent noun indicates a specific event, whereas an incorporated noun indicates a more general or habitual meaning:
ni-c-qua in nacatl
I-it-eat the flesh
'I eat the flesh.' [particular act]
ni-naca-qua
I-flesh-eat
'I eat flesh' or 'I am a flesh-eater.'
A free word order conlang could require incorporation of all objects and not use either case or agreement, a typological combination that does not exist in natlangs to my knowledge.
Apart from the morphological / grammatical angle, we can also understand meaning by context.
A mournful song sang the choir.
Except in the Land of Strange Tales, we know that choirs sing songs. Songs don't sing choirs. The inversion is startling when interjected into speech of ordinary pattern, but it's quite understandable. Even though we don't have morphological clues to tell us, we intuitively know that this sentence is not to be "read straight through".
Man bites dog.
Again, context informs our understanding. Here, the (seeming!) inversion is nòt used for poetic effect, but is instead the epitome of the ironically prosaic. In this case, the context is journalism and headline writing style and is a commentary on what is newsworthy.
The key principle to understanding what is being uttered or has been written is to know how to put the words into a structural context. This is often explained in simple terms using the W questions: who is doing something, what are they doing, how are they doing it, to whom are they doing it, when and where are they doing it, etc. In grammatical terms, these can be labelled as subject, verb, adverbs, objects, adverbials and more.
In some languages, quite a few categories are distinguishable a priori just by their ending. For example in Esperanto, singular nouns are marked with terminal -o, adjectives end in -a and conjugated verbs usually in -[vowel]s. Likewise, a general knowledge of the conjugation tables of Finnish often allows a quick and dirty rough estimate whether a word happens to a noun or adjective or a verb form. English is generally terrible at this distinction with a number of words being written and pronounced the same regardless of the category (see: Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo).
However, what is mentioned in the paragraph above only accounts for the word type; if a sentence has two nouns we may not immediately be able to distinguish between the subject noun and the object noun without additional information. Thus, languages generally use one of two methods to clarify what exactly is meant:
The information is encoded in word order. This is often the rule for languages with rather low levels of inflection such as English or Chinese. In English, the following two sentences are different because of the different subject and object functions even though all the words are the same:
The man bites the dog.
The dog bites the man.
The information is encoded in inflection, prepositions, participles, suffixes and the like. While very likely all languages make use of this scheme partially, some do to a much larger extent than English; for example German or Finnish. In German and Finnish, the following two sentences with the same words but different order mean the same:
Der Mann beißt den Hund.
Mies puree koiraa.Den Hund beißt der Mann.
Koiraa puree mies.
(I’m not sure if the second Finnish example is strictly allowed in that way; however, my Finnish course includes a similar precedence of an OVS sentence).
By marking the nouns according to case — and thus, according to their grammatic function — we have one clear subject (the agent, the man; nominative case in both examples) and one clear object (the patient, the dog; accusative or partitive case). Once the relation between the two is established like this, one is typically free to move the components around.
Other than using declension by case, modifiers can be placed around a word. In Japanese, that would be the particles が, は, を, に, へ and others. The word or phrase that precedes the particle が is understood to be the subject of a sentence, whatever precedes を is an object. In theory, this would allow the parts suffixed in that way to be moved around as seen fit. (Whether Japanese actually does that I do not know as my knowledge of it is still too basic.) Other options of the same general flavour include prepositions: in the house shows that the house is used to denote location and is not an agent or a patient of the sentence. Even in the otherwise relatively strict word order of English, adverbials such as that can be moved around to a certain degree.
tl;dr:
Languages that allow a more or less free word order will not rely on word order to assign syntactical roles. Instead, markers or modifiers of certain types will likely be used; cases and declension being a rather common choice.
The real answer is that languages don't actually have free word order. Rather, languages with 'free' word order use the rearrangements to signify additional meaning. Here, look at English:
Hey, what did you eat?
This is a plain question, indicated by shifting the 'what' from its underlying position to the front.
Wait, you ate what?!
From the perspective of a logician, this sentence should mean the same thing, right? But it doesn't. When you don't shift a wh-word, it doesn't remain a question, it's a statement of surprise and maybe a request for clarification. It's also stylistically informal, and wouldn't be used in a sentence. In other words, it's what linguists call a 'pragmatic shift'.
Every language with 'free' word order does something like this. Traditional grammar and logic tends to downplay or be unaware of word order and pragmatics, and so people who speak such languages tend to believe that words can be arranged in any which way you want. But underlyingly, every arrangement of word order carries a stylistic or pragmatic significance, and using the 'wrong' arrangement will make it sound off to a fluent speaker, even if they couldn't explain why.
Esperanto is an interesting example -- officially speaking, Esperanto has totally free word order. But in practice, Esperanto word orderings follow unspoken stylistic and pragmatic rules (typically lifted from Germanic and Slavic). So this would presumably occur with any fluently spoken language, conlang or not.
Many languages, like Japanese and Korean, have particles added to words to show which part of speech they are. While word order is still necessary for these languages to make sense, a conlang might have grammar where each word's part of speech is not defined by their place in the sentence, but what particles are attached to the words. A similar effect could be achieved by using different pitches or different word forms to define a word's part of speech.
