The One Grammar... TO RULE THEM ALL!!!
...or OGTRTA, for short, is a meta-grammar for constructed languages. It's kind of like a grammar template that you can customize for your conlang.
It is designed to minimize syntactic ambiguity (think Lojban) without requiring agreement-marking, while remaining at least somewhat naturalistic. You can dial the syntactic precision up or down to your liking.
This document describes the OGTRTA framework through the lens of a specific language that uses it: English. I cover not only how OGTRTA and English do things, but why. Along the way, I'll also discuss the decisions that I made when designing English, and explore some of the roads not taken. Example sentences will be in bold, and counterfactual examples (e.g. using different grammar rules from those I chose for English) will be marked with an asterisk *like this.
The most interesting thing about OGTRTA is probably its treatment of verbs, prepositions, adjectives, and adverbs. OGTRTA collapses all of these parts of speech into one, predicates. This design decision was inspired by a series of observations about English:
- Many verbs can be either transitive or intransitive. E.g. "The bird ate the rice" → "The bird ate".
- English prepositions can lose their object and function like adverbs. We can say "Is Josh in the house?" or "Is Josh in?"
- These two derivational processes (if indeed they are derivational) look awfully similar! A word that takes a complement can lose it.
- Participles function syntactically much like prepositions. Some pairs of prepositions and participles are nearly synonymous: "I broke the rock with a hammer" (preposition) vs. "I broke the rock using a hammer" (participle).
- Adjectives can take complements, much like verbs and prepositions. "The elf is eager" → "The elf is eager to please."
- Adjectives, prepositions, and verbs can all modify nouns (the latter via relative clauses) and can be the predicate of a sentence (though English requires a copula for non-verbs).
It certainly seems like there are similarities between these parts of speech! So why not unify them?
The Celtic languages also provided a great deal of inspiration. The reference implementation of OGTRTA uses initial consonant mutations, as well as tense-marking particles similar to those in Cornish.
This section presents a "reference implementation" of OGTRTA—a language designed according to OGTRTA conventions. To avoid confusion, the reference language is called English.
OGTRTA is a "reversible" grammar: all the production rules can be reversed to get a language with the opposite word order. The basic supported word orders are head-initial VOS and head-final SOV. OGTRTA languages can implement transformations can move the subject to the opposite end of the sentence, so SVO and OVS word orders are also possible.
The other two possible word orders, VSO and OSV, are not directly supported, but can be achieved using clitic subject pronouns or noun incorporation.
A noun phrase describes a person, place, thing, or idea. The production rule for noun phrases in a head-initial OGTRTA language is:
NP = DET? N MP*
Where DET is a determiner, N is a noun, and MP* is zero or more modifier phrases (I'll get to those in a bit).
Examples in English:
- edhen "bird"
- am edhen "the bird"
- edhen velen "yellow bird"
- am edhen velen "the yellow bird"
- am edhen velen pig "the little yellow bird"
The DET? part of the rule is optional; OGTRTA languages need not have determiners at all. I've chosen to include determiners in English to demonstrate more of the range of OGTRTA's capability, but I could just as easily have chosen to leave them out.
Even if you do have determiners in your language, that doesn't mean you have to use them for the same set of things that English does. Possessive pronouns, for example, can be either determiners or modifiers:
- *nin edhen "my bird" (determiner) (the asterisk indicates a counterfactual, since English doesn't actually work this way)
- am edhen nin "my bird" (modifier)
Demonstratives, too, can be determiners or modifiers
- *hen edhen "this bird" (determiner)
- am edhen hen "this bird" (modifier)
What about head-final languages? If your language is head-final, the production rule is reversed:
NP = MP* N DET?
This rule would generate:
- *edhen am "the bird"
- *velen edhen "yellow bird"
- *velen edhen am "the yellow bird"
- *pig velen edhen am "the little yellow bird"
Variants of the basic production rule are possible. Since the determiner is essentially a clitic on the noun, we can move it to the other side without ambiguity:
NP = N DET? MP*
We can also choose to allow "simple" MPs (e.g. adjectives with no complements or modifiers of their own, denoted below as SMP) between the noun and its determiner:
NP = (DET SMP*)? N MP*
Note that we cannot allow SMPs before the noun in the absence of a determiner without creating syntactic ambiguity, since in an actual sentence the NP might be preceded by another NP with modifiers of its own.
A second production rule for NP describes the use of conjunctions like "and" and "or":
NP = NP CONJ NP
There's not much to say about this. Conjunctions: they exist!
While English has conjunctions, other OGTRTA languages might not. It's possible to use modifier phrases instead:
- *am edhen do am pesh "the bird with the fish" (i.e. "the bird and the fish")
We have already seen a few examples of modifier phrases above: the adjective-like words pig and melen (sic—the spelling velen arises from a consonant mutation). Technically, though, these are not adjectives, but predicates.
Predicates in OGTRTA play the roles that verbs, adjectives, and prepositions do in English. The following words are all predicates:
- melen "yellow"
- math "eating"
- ga "of"
- gavo "giving"
Some predicates take complements. Complements are noun phrases that follow the predicate (in head-first languages) and complete its meaning. Predicates must have all their complement "slots" filled. The number of complement slots a predicate has is called its valence.
Adjective-like predicates like melen generally don't take any complements, so they have a valence of 0. Math is a transitive-verb-like predicate and has a valence of 1. Preposition-like predicates like ga also have a valence of 1. Gavo is ditransitive, and has a valence of 2.
The production rule for an MP is:
MP = P_n MP* NP{n}
That is, an MP consists of a predicate (which has some valence n), zero or more modifier phrases, and n noun phrases to fill the complement slots.
Examples of MPs with complements:
- math pelt "eating rice"
- ga am edhen "of the bird"
- gavo pelt am edhen "giving rice to the bird"
Since the complements are noun phrases, they can contain MPs of their own. With just two production rules, we have a recursive grammar, capable of producing phrases of any length or complexity.
- gavo pelt am edhen vowr math edhen big "giving rice to the big bird eating a small bird"
The predicate of an MP may have modifiers of its own. These modifiers are essentially adverbs.
A common adverb is mowr (mutated form vowr) meaning "big, very, fully"
- melen vowr "very yellow"
- math vowr "devouring, eating up"
- gavo vowr pelt (nebban) "giving all the rice (to someone)"
In complex phrases, it can sometimes be hard to tell which word an MP is modifying. Consider the phrase am edhen velen vowr. Does this mean "the big yellow bird" or "the very yellow bird?" If vowr is modifying edhen, it means "the big yellow bird". If vowr is modifying velen, it means "the very yellow bird".
Actually, there is no ambiguity here, because English uses consonant mutation to resolve it. That is, the initial consonant sound of a predicate changes when that predicate immediately follows the word it is modifying. If any other word comes between the modifyee and the predicate, there is no mutation. Thus, we have:
- am edhen velen vowr "the very yellow bird"
- am edhen velen mowr "the big yellow bird"
I personally like this mechanism; I think it's elegant and fairly naturalistic. However, it doesn't completely resolve all ambiguity about modifier attachment. If there are more than two levels of nested phrases, there is no way to know if an unmutated modifier attaches to the word one level up or two (or more) levels up.
For example, in the sentence am edhen vath edhen velen mowr, is the big bird the one eating, or the one being eaten?
It is possible to resolve this ambiguity if we distinguish three kinds of MPs:
- Those that are first among their modifyee's modifiers.
- Those that are last among their modifyee's modifiers.
- Those that are in the middle, i.e. neither first nor last.
The consonant mutation rule in English distinguishes first modifiers from the others, but there is no distinction between last and "middle" modifiers. Let's imagine we used a prefix ac- on last modifiers. Then we could construct phrases like:
- *am edhen vath edhen ac-velen ac-mowr "The big bird eating the yellow bird"
- *am edhen ac-vath edhen ac-velen ac-vowr "The bird eating the very yellow bird"
- *am edhen ac-vath edhen velen ac-mowr "The bird eating the big yellow bird"
While this is nice and logical, it has several practical weaknesses.
First, human speakers do not always know when they are about to utter the last modifier of a word. In general, it's a losing battle to try to force humans to parse or generate unambiguous nested structures of arbitrary depth. We just don't have the working memory for it.
Second, the resulting structure is difficult to parse in real time. It's a bit of a logic puzzle to deduce which modifiers attach to which modifyees.
Finally, some ambiguity about modifier attachment is normal in natural languages, and in many cases it is unclear even to the speaker which word a modifier should attach to. In the sentence the suspect murdered the man in his bedroom, does the prepositional phrase in his bedroom attach to the man, or is it an adverbial modifier on murdered? Practically, it doesn't matter. The meaning is nearly the same in either case.
One way to reduce ambiguity about modifier attachment is to require modifiers to agree with the modified noun. Agreement assumes that each noun is assigned a "class", and both the speaker and listener know what class any given noun is in. In European languages, the classes are generally masculine, feminine, and (sometimes) neuter, and are assigned to nouns arbitrarily. But this is by no means the only possibility. For example, many languages distinguish between animate and inanimate nouns. Tamil divides the world into rational and irrational beings (and children are irrational!) And Swahili has 15 different noun classes.
English has very minimal agreement (an animacy distinction in third-person pronouns only). But for the sake of making the idea concrete, let's imagine what English would be like with modifier agreement.
Let's suppose that "bird" and "fish" are in different noun classes - e.g. "bird" is in the noun class for warm-blooded vertebrates ("class W"), and "fish" is in the noun class for cold-blooded vertebrates ("class C"). Predicates modifying class-W nouns infix -ey- before their last vowel. Predicates modifying class-C nouns infix -imm- before their last vowel. Here are the results:
- *edhen meleyen "yellow bird"
- *pesh melimmen "yellow fish"
- *am edhen mayeth pesh melimmen mimmowr "the bird eating the big yellow fish"
The production rule for MPs described above requires us to put all modifiers of a predicate before its complements. This is awkward, because those modifiers might be lengthy prepositional phrases.
- *am edhen vath vowr na goweth pelt "the bird devouring rice in the tree" (lit. "the bird eating up in the tree rice")
We can relax the rule, at the cost of some ambiguity, to:
MP = P_n MP* NP{n} MP*
That is, we can allow post-complement modifiers of the predicate.
The ambiguity results from the fact that the NP complements of the predicate might have modifiers of their own, and it might not be possible to tell where the sequence of modifiers for the last NP ends, and the modifiers for the MP begin.
However, if we assume that the post-complement modifiers are prepositional phrases, this is "good ambiguity". As discussed above, the attachment of prepositional phrases is usually not crucially important, so allowing the attachment to be ambiguous reduces the burden on the speaker without increasing the burden on the listener.
Now we have all the pieces in place to create complete sentences. Sentences in English have the following production rule:
S = MP* SN
SN = NP TAM MP
SN = TAM MP MP
That is, a sentence consists of zero or more modifier phrases followed by a sentence nucleus (SN).
An SN contains an MP containing the main predicate of the sentence, preceded by a tense-aspect-mood particle (TAM). The subject of the sentence is an MP, which can either come first or last.
Here are some examples of sentences without leading MPs:
- Am edhen a canu. "The bird sings."
- Am edhen a math pelt. "The bird eats rice."
The modifier phrases at the start of a sentence play the role of sentence-level conjunctions and conjunction-like adverbs ("however", "additionally") and attitudinal adverbs ("hopefully", "fortunately"). They can also be prepositional phrases ("on Tuesday").