I would not call myself a good conlanger, since I have neither the education nor experience. But I'd like to offer what I think is some good advice anyway.
Many people, when they first start creating languages, simply create vocabulary by writing new words corresponding to each English word. This is often known as a relex, since it is equivalent to re-creating the lexicon. It is, of course, enough for the beginning conlanger, who finds all the novelty they need in simply speaking alien words. Leže žakbidak, I would say, wishing those around me a happy birthday—for žak meant "birth", and bidak "day". Such was Qoqqkyuk1, my very first conlang.
This is the first instinct: to simply do everything the same way as you are used to, with trivial differences, and for its purposes it serves.
A variant of the first instinct arises when some other language is learned about and emulated. Thus, when I learned elementary Latin, I found myself desiring to recreate the language; partially though a perception of its superiority—after all, the presence of inflectional endings allowing free word order is obviously better and more logical!—and partly through a desire to create that which I knew to be interesting. Thus Kyndae was born, and it had nouns that inflected like this:
The terminations of the nominative are especially damning. The vocative is also to be noted: the spelling ë is simply a Tolkienism. I was beginning to realize that I should use some of my own ideas: the pattern of acute and circumflex accentuation is indeed exceedingly Greek, but I had never set eyes on the Greek language when I created Kyndae, and so this must have been of my own invention. Yet in the end this is still primarily created from the first instinct, for here too I was simply doing everything in a way that I had seen things done.
But not everything. I did not quite copy the cases wholesale; you will see here the split of the genitive into "possessive" and "relative", and the merger of the dative and ablative. This shows a pattern of thought that is also all too common, not only among beginners but also among those who have been conlanging for many years. They see imperfections in real languages: redundancies, ambiguities, arbitrariness of words and inflections, and they wish to eliminate these, to make their language better. So they take words whose meaning covers a wide semantic field, and split it up, so that "man" becomes one word when meaning "person", another when meaning "male", and another when meaning "adult". And they collapse synonyms together, so that "kid" and "child" both translate to a single word. And often as well they will create inflectional endings, or groups of related words, that differ only by a single sound, so that "spring", "summer", "autumn", "winter" become keta, keti, keto, kete.2 With Kyndae I did this; you can see the collapse of the dative and ablative into a single case (since, at the time at least, I did not understand their differing usages in Latin), and the splitting of the genitive into its "possessive" (that is, describing ownership, like "the father's book") and "relative" (that is, describing relation to something, like "the lord of the rings"). The phenomenon of minimal distinction didn't show up in the nouns, but it does show up in the verbs, where the pattern of inflection was þôtu, þôntu, þôtë, þôta, þônta, þôti, with a different suffixed consonant for each TAM.
This is the second instinct, for it is that which rejects the first instinct. Such a conlanger decides not to create a language like the language they know, but rather to do it differently, to do it better. Unfortunately, this is in many ways a bad way to create a language. Redundancy is important in languages, since most communication is noisy (that is, the signal is distorted by some form of noise, whether it be real auditory noise or a failure to correctly say what you mean). Apparent synonyms often differ in slight connotation, or, if nothing else, in register. Conversely, words with a wide semantic field are useful when multiple aspects of that field are meant: "man" brings to mind an image of someone who is adult and male, but perhaps in a way that implies they are, for the purposes of the conversation, a person. And words (or morphemes) of a certain class who differ only slightly are in fact a terrible idea, for any aforementioned distortion can cause one to hear the wrong word—it is so much easier to hear /keta/ for /keto/ than to hear /spɹɪŋ/ for /ɔtəm/—and in such a case context will not help you; whereas even such a small phrase as "in the spring" precludes "bring" from making any sense, the same sentence in Arka would easily lead a listener to false conclusions.
Sadly, far too many auxlangers fall victim to this design flaw. Esperanto, by far the most well-known and successful auxlang, distinguishes the present, past, future, and conditional by a single vowel: -as, -is, -os, -us. Those who design artlangs seem less likely to do so, being not necessarily motivated by the ideal of making their language better, but it has nevertheless happened. There is of course nothing wrong with a little of this—especially when it comes to inflectional endings, though also lexically (consider English "where", "when")—but it should not be considered a design ideal.
One who has read H13's advice on conlanging knows the way to make a good conlang: learn typology. Conlanging is like any art. To understand how to write a song, to know everything that makes a song good, well—you can be told some of these things, to be sure, but most important is listening to lots of music. Trying to create a language when all you know about is English is like writing a song when all you've ever heard is a single Beatles record, or a single symphony by Brahms. Even if you know that a song must have sound, melody, harmony, orchestration—it doesn't help you if you've never heard different kinds of harmony or orchestration applied. Even if you know all of the rules and guidelines surrounding first species counterpoint, this won't help you compose a beautiful chorale—unless you've spent a long time listening to Bach, and understanding how those rules are applied. Take it from experience, I've tried. And similarly, if you try to put an antipassive into your language, even if you follow every universal that WALS lists, it won't be quite right, and it certainly won't be interesting.
The way to create a good conlang, then, is to have an existing knowledge of many different languages and language features, to have spent years reading linguistics papers, and then to pick out features from each and cobble together from them a language. This will get you a phonology, and a grammar, and many important details of how your language behaves on a fundamental level. If you spend a lot of time reading dictionaries, or learning other languages in a practical sense, it will get you a vocabulary as well.
Thus all of the conlanging you ever need to do can be done by following your first instinct. Your language will be realistic; it will not resemble too greatly any one world language, being rather composed from bits of many; it will be interesting to be sure, for there are many languages out there with absolutely bizarre features.
But this is work.
When all you know is English, and maybe also Latin, and you want to make something new, and interesting, and different, you resort to the second instinct. The goal doesn't have to be to make the language better, it should just be different. The second instinct is the second instinct because it rejects the first; whereas the first instinct is to do things the same, the second instinct is to do things differently. But the second instinct is a problem, because it's too predictable. It divides words and grammatical concepts up by different boundaries than a natural language does, but it divides them up by obvious boundaries, and those, with a little variation, are objective and universal.
Every time I look at a word and think how to translate it, what semantic field to give it, I think first of how it is done in the few languages I know. Then I consider what the obvious semantic boundaries are. Then I consider both of these options, but ask myself: what other possible ways could I define the range of this word? And without ever consulting a dictionary, I decide. The word avé means 'hill', but not just any hill. Say it means a hill that is covered in green grass or wood, but not dry weeds or bare dirt, or a rocky outcropping. And, having given such a definition, I ask myself: what similar words exist? I reject the second instinct that tells me to create a corresponding word for a bare hill—perhaps such a concept can only be expressed with an adjective—but I will create a word for "tor", hueton. A hill contrasts with a mountain, just as in my native language: the word for mountain is edris, and edris is used of any type of mountain, whether alone or in a range, no matter how tall, no matter if properly snow-capped or not. Perhaps the size is a little different, though, so that a thing I would call a hill is still an edris. And then I will add a word to describe a range of hills, eresé. In this too I reject the first instinct—for such a word does not exist in any language I know—but I reject the second also, for that would tell me just to refer to such a thing using the plural of avé. In fact, I decide to reject the second instinct so completely that I say the plural of avé is incorrect.
This is the third instinct, the instinct to reject the second instinct—sometimes in favor of the first, as with my word for "mountain", but often enough in favor of something different, a third way out, simply considering all of the ways one can use "hill" or something similar to "hill", and drawing arbitrary lines in the sand. By so doing I have created a semantic map of four words, that do not correspond to the semantic division any language that I know, and quite possibly do not correspond to any language at all. Yet I do not think they are unrealistic. And this is partly the boundaries of realism are wide—as I said, there are some truly astounding natlang features out there—but also partly because realism doesn't need to be defined by what exists. In fact, it absolutely shouldn't be.
The third instinct helps create interesting vocabulary; in fact, in my opinion it's very good at it. It takes more time, of course, than simply borrowing from the first instinct the way a learned one would do. Yet the third instinct does not preclude the use of the first instinct, and in fact doesn't always need to be used over the second either. H13 says that a language only needs a few interesting quirks to be really notable. I don't think it's a bad idea to write dictionary entries for 90% of the Swadesh list with definitions as simple as "woman", "two", "long", "fish"—but then take the other 10% and put a lot of detail into them. You don't even need to roll dice—just write down simple definitions until an idea pops into your head while looking at a word. And this makes sense too, typologically, because semantic ranges for the most part don't tend to vary that much.
And the third instinct can be used outside of the lexicon. Suppose I have a near demonstrative and a far demonstrative. Logically they should behave the same way grammatically, since they're the same kind of word; and the first and second instincts agree here. But the third instinct rejects this. One of them is used with the article, and the other isn't. In fact, I could then go on to expand this to create a general pattern of demonstratives that are used instead of an article, and draw arbitrary lines for what words fall into this class. Then I decide to add another word that is a normal demonstrative, so now we have one far demonstrative that's used with the article and one that isn't. But I follow my third instinct again, and add an extra twist: this new word isn't put before the noun, but after.
And to be sure, knowing just what you could possibly do to a word is harder if you don't study typology. Even having the idea to put a word after its noun instead of before is harder if you've never studied a language that did that.
Harder, but not impossible. All you need to do—all the third instinct is—is to write out a description of how your grammar works, even if it's just a copy of your high school English textbook, and then to point at some word at random and ask yourself, "what if it didn't?"
And even this doesn't have to preclude realism. It's hard to do something truly novel in this way, because the phenomenon of ANADEW is ubiquitous, but maybe you did. Maybe you broke a universal by having a language with rising tones but no falling tones. I don't think it matters. I think, when it comes down to it, the real kind of universals you need aren't typological. They're the kind of rules that typology helps you follow, but you can still do statistics with a sample size of one, especially when the population standard deviation is so wide.
Therefore I prescribe the following universals:
And the more you follow the third instinct, the easier it becomes to use, even outside of the field of conlanging. In my fantasy conworld, how should the culture treat homosexuality? The first instinct says it should be shunned and disgraced, as it is and was in many cultures. The second instinct rejects this, for there is no real reason to consider homosexuality bad, and it should be considered just the same as heterosexuality. The third instinct rejects the second: homosexuality will indeed be considered immoral. But just going with the first instinct isn't interesting enough, so I add a twist: homosexuality is considered acceptable under some condition; it's considered acceptable if the relationship is between a male human and a male elf, because elves are neotenous. Here the analog of typology helps, as I am largely borrowing the concept of pederasty, but I am using my own ideas, and it was the third instinct that led me to this.
I find the third instinct consistently helps me create things that feel realistic but also interesting. I hope it will help you similarly.
Many people, when they first start creating languages, simply create vocabulary by writing new words corresponding to each English word. This is often known as a relex, since it is equivalent to re-creating the lexicon. It is, of course, enough for the beginning conlanger, who finds all the novelty they need in simply speaking alien words. Leže žakbidak, I would say, wishing those around me a happy birthday—for žak meant "birth", and bidak "day". Such was Qoqqkyuk1, my very first conlang.
This is the first instinct: to simply do everything the same way as you are used to, with trivial differences, and for its purposes it serves.
A variant of the first instinct arises when some other language is learned about and emulated. Thus, when I learned elementary Latin, I found myself desiring to recreate the language; partially though a perception of its superiority—after all, the presence of inflectional endings allowing free word order is obviously better and more logical!—and partly through a desire to create that which I knew to be interesting. Thus Kyndae was born, and it had nouns that inflected like this:
| First declension | Singular | Plural |
|---|---|---|
| Nominative | kínda | kíndae |
| Possessive | kíndek | kíndaek |
| Relative | kînd | kîndon |
| Dative | kíndes | kíndos |
| Vocative | kíndë | kindêa |
| Dative/Ablative | kîndin | kíndept |
| Temporal | kîndiʃ | kíndeʃ |
| Locative | kîndig | kíndeg |
The terminations of the nominative are especially damning. The vocative is also to be noted: the spelling ë is simply a Tolkienism. I was beginning to realize that I should use some of my own ideas: the pattern of acute and circumflex accentuation is indeed exceedingly Greek, but I had never set eyes on the Greek language when I created Kyndae, and so this must have been of my own invention. Yet in the end this is still primarily created from the first instinct, for here too I was simply doing everything in a way that I had seen things done.
But not everything. I did not quite copy the cases wholesale; you will see here the split of the genitive into "possessive" and "relative", and the merger of the dative and ablative. This shows a pattern of thought that is also all too common, not only among beginners but also among those who have been conlanging for many years. They see imperfections in real languages: redundancies, ambiguities, arbitrariness of words and inflections, and they wish to eliminate these, to make their language better. So they take words whose meaning covers a wide semantic field, and split it up, so that "man" becomes one word when meaning "person", another when meaning "male", and another when meaning "adult". And they collapse synonyms together, so that "kid" and "child" both translate to a single word. And often as well they will create inflectional endings, or groups of related words, that differ only by a single sound, so that "spring", "summer", "autumn", "winter" become keta, keti, keto, kete.2 With Kyndae I did this; you can see the collapse of the dative and ablative into a single case (since, at the time at least, I did not understand their differing usages in Latin), and the splitting of the genitive into its "possessive" (that is, describing ownership, like "the father's book") and "relative" (that is, describing relation to something, like "the lord of the rings"). The phenomenon of minimal distinction didn't show up in the nouns, but it does show up in the verbs, where the pattern of inflection was þôtu, þôntu, þôtë, þôta, þônta, þôti, with a different suffixed consonant for each TAM.
This is the second instinct, for it is that which rejects the first instinct. Such a conlanger decides not to create a language like the language they know, but rather to do it differently, to do it better. Unfortunately, this is in many ways a bad way to create a language. Redundancy is important in languages, since most communication is noisy (that is, the signal is distorted by some form of noise, whether it be real auditory noise or a failure to correctly say what you mean). Apparent synonyms often differ in slight connotation, or, if nothing else, in register. Conversely, words with a wide semantic field are useful when multiple aspects of that field are meant: "man" brings to mind an image of someone who is adult and male, but perhaps in a way that implies they are, for the purposes of the conversation, a person. And words (or morphemes) of a certain class who differ only slightly are in fact a terrible idea, for any aforementioned distortion can cause one to hear the wrong word—it is so much easier to hear /keta/ for /keto/ than to hear /spɹɪŋ/ for /ɔtəm/—and in such a case context will not help you; whereas even such a small phrase as "in the spring" precludes "bring" from making any sense, the same sentence in Arka would easily lead a listener to false conclusions.
Sadly, far too many auxlangers fall victim to this design flaw. Esperanto, by far the most well-known and successful auxlang, distinguishes the present, past, future, and conditional by a single vowel: -as, -is, -os, -us. Those who design artlangs seem less likely to do so, being not necessarily motivated by the ideal of making their language better, but it has nevertheless happened. There is of course nothing wrong with a little of this—especially when it comes to inflectional endings, though also lexically (consider English "where", "when")—but it should not be considered a design ideal.
One who has read H13's advice on conlanging knows the way to make a good conlang: learn typology. Conlanging is like any art. To understand how to write a song, to know everything that makes a song good, well—you can be told some of these things, to be sure, but most important is listening to lots of music. Trying to create a language when all you know about is English is like writing a song when all you've ever heard is a single Beatles record, or a single symphony by Brahms. Even if you know that a song must have sound, melody, harmony, orchestration—it doesn't help you if you've never heard different kinds of harmony or orchestration applied. Even if you know all of the rules and guidelines surrounding first species counterpoint, this won't help you compose a beautiful chorale—unless you've spent a long time listening to Bach, and understanding how those rules are applied. Take it from experience, I've tried. And similarly, if you try to put an antipassive into your language, even if you follow every universal that WALS lists, it won't be quite right, and it certainly won't be interesting.
The way to create a good conlang, then, is to have an existing knowledge of many different languages and language features, to have spent years reading linguistics papers, and then to pick out features from each and cobble together from them a language. This will get you a phonology, and a grammar, and many important details of how your language behaves on a fundamental level. If you spend a lot of time reading dictionaries, or learning other languages in a practical sense, it will get you a vocabulary as well.
Thus all of the conlanging you ever need to do can be done by following your first instinct. Your language will be realistic; it will not resemble too greatly any one world language, being rather composed from bits of many; it will be interesting to be sure, for there are many languages out there with absolutely bizarre features.
But this is work.
When all you know is English, and maybe also Latin, and you want to make something new, and interesting, and different, you resort to the second instinct. The goal doesn't have to be to make the language better, it should just be different. The second instinct is the second instinct because it rejects the first; whereas the first instinct is to do things the same, the second instinct is to do things differently. But the second instinct is a problem, because it's too predictable. It divides words and grammatical concepts up by different boundaries than a natural language does, but it divides them up by obvious boundaries, and those, with a little variation, are objective and universal.
Every time I look at a word and think how to translate it, what semantic field to give it, I think first of how it is done in the few languages I know. Then I consider what the obvious semantic boundaries are. Then I consider both of these options, but ask myself: what other possible ways could I define the range of this word? And without ever consulting a dictionary, I decide. The word avé means 'hill', but not just any hill. Say it means a hill that is covered in green grass or wood, but not dry weeds or bare dirt, or a rocky outcropping. And, having given such a definition, I ask myself: what similar words exist? I reject the second instinct that tells me to create a corresponding word for a bare hill—perhaps such a concept can only be expressed with an adjective—but I will create a word for "tor", hueton. A hill contrasts with a mountain, just as in my native language: the word for mountain is edris, and edris is used of any type of mountain, whether alone or in a range, no matter how tall, no matter if properly snow-capped or not. Perhaps the size is a little different, though, so that a thing I would call a hill is still an edris. And then I will add a word to describe a range of hills, eresé. In this too I reject the first instinct—for such a word does not exist in any language I know—but I reject the second also, for that would tell me just to refer to such a thing using the plural of avé. In fact, I decide to reject the second instinct so completely that I say the plural of avé is incorrect.
This is the third instinct, the instinct to reject the second instinct—sometimes in favor of the first, as with my word for "mountain", but often enough in favor of something different, a third way out, simply considering all of the ways one can use "hill" or something similar to "hill", and drawing arbitrary lines in the sand. By so doing I have created a semantic map of four words, that do not correspond to the semantic division any language that I know, and quite possibly do not correspond to any language at all. Yet I do not think they are unrealistic. And this is partly the boundaries of realism are wide—as I said, there are some truly astounding natlang features out there—but also partly because realism doesn't need to be defined by what exists. In fact, it absolutely shouldn't be.
The third instinct helps create interesting vocabulary; in fact, in my opinion it's very good at it. It takes more time, of course, than simply borrowing from the first instinct the way a learned one would do. Yet the third instinct does not preclude the use of the first instinct, and in fact doesn't always need to be used over the second either. H13 says that a language only needs a few interesting quirks to be really notable. I don't think it's a bad idea to write dictionary entries for 90% of the Swadesh list with definitions as simple as "woman", "two", "long", "fish"—but then take the other 10% and put a lot of detail into them. You don't even need to roll dice—just write down simple definitions until an idea pops into your head while looking at a word. And this makes sense too, typologically, because semantic ranges for the most part don't tend to vary that much.
And the third instinct can be used outside of the lexicon. Suppose I have a near demonstrative and a far demonstrative. Logically they should behave the same way grammatically, since they're the same kind of word; and the first and second instincts agree here. But the third instinct rejects this. One of them is used with the article, and the other isn't. In fact, I could then go on to expand this to create a general pattern of demonstratives that are used instead of an article, and draw arbitrary lines for what words fall into this class. Then I decide to add another word that is a normal demonstrative, so now we have one far demonstrative that's used with the article and one that isn't. But I follow my third instinct again, and add an extra twist: this new word isn't put before the noun, but after.
And to be sure, knowing just what you could possibly do to a word is harder if you don't study typology. Even having the idea to put a word after its noun instead of before is harder if you've never studied a language that did that.
Harder, but not impossible. All you need to do—all the third instinct is—is to write out a description of how your grammar works, even if it's just a copy of your high school English textbook, and then to point at some word at random and ask yourself, "what if it didn't?"
And even this doesn't have to preclude realism. It's hard to do something truly novel in this way, because the phenomenon of ANADEW is ubiquitous, but maybe you did. Maybe you broke a universal by having a language with rising tones but no falling tones. I don't think it matters. I think, when it comes down to it, the real kind of universals you need aren't typological. They're the kind of rules that typology helps you follow, but you can still do statistics with a sample size of one, especially when the population standard deviation is so wide.
Therefore I prescribe the following universals:
- Don't have too much complexity, or too little. "Too many" is hard to define. You can compare it to your native language, but it can also be hard to realize what complexity there is. That said, following this rule isn't that hard. We live in a world where people have written books—thick books—on the uses of the Ancient Greek article. Chances are, you'll get bored of thinking up rules long before your rules become too complex. And as for the other side, well, a language entirely without rules is just some sort of caveman speak, and that would just be boring. Half of the fun is inventing the rules, or should be. And you can get close. In fact, getting close helps you do interesting things. It's a good way to reinvent serial verbs, for example.
- Don't make everything surjective. Sometimes the members a given category, be it lexical or grammatical or phonological, won't span the whole range of the category itself. The words for "hill" that Maotic has vary depending on the covering that hill has, but you can't describe every kind of hill with just one word. Sometimes you need an adjective. The cases that Finnish has to describe the relative position of objects are extensive, but not exhaustive. And when it comes to grammar, make some words defective. English has no simple non-finite form of "can". You can do all of these things because of periphrasis. Come up with some other way to say things in your language, something that's a little bit more difficult.
- Don't make everything injective. In fact, make lots of things not injective. That is to say, introduce redundancy. English has several words for "breast" depending on register, gender, and size, but in most cases more than one of these apply. A certain verb in Ancient Greek can be in the active or middle with no difference in meaning; it can take the accusative or dative.
- Avoid homogeneous redundancy. At the same time, there are almost always differences, however slight, in each pair of synonyms. To be sure, you can have similar morphemes that are synonyms with literally no difference, like -ness, -hood, and -ship, but in general try to limit it to a couple. If nothing else, the words "ask" and "inquire" have different lengths, and different languages of origin. There are many ways to express intent in English ("for", "to", "in order to", "in order that", "so as to", "for the purpose of") and they are not all conjunctions of the same size or the same construction.
And the more you follow the third instinct, the easier it becomes to use, even outside of the field of conlanging. In my fantasy conworld, how should the culture treat homosexuality? The first instinct says it should be shunned and disgraced, as it is and was in many cultures. The second instinct rejects this, for there is no real reason to consider homosexuality bad, and it should be considered just the same as heterosexuality. The third instinct rejects the second: homosexuality will indeed be considered immoral. But just going with the first instinct isn't interesting enough, so I add a twist: homosexuality is considered acceptable under some condition; it's considered acceptable if the relationship is between a male human and a male elf, because elves are neotenous. Here the analog of typology helps, as I am largely borrowing the concept of pederasty, but I am using my own ideas, and it was the third instinct that led me to this.
I find the third instinct consistently helps me create things that feel realistic but also interesting. I hope it will help you similarly.