Eventually I'll make a blog or something.
Four functions of phonotactics in constructed languages
A language's phonotactics is its restrictions on the patterning of sounds. English phonotactics, for example, prohibits words from starting with the th and n sounds back-to-back, so Dr. Seuss was breaking the rules when he coined the word thneed. Phonotactics often noticeably contributes to the character of a spoken language. Tolkien relied on phonotactics over phonemes to give the Black Speech its harsh and foreign feel. Take the famous couplet that appears inscribed on the One Ring in The Lord of the Rings:
Ash nazg durbatulûk, ash nazg gimbatul,
ash nazg thrakatulûk agh burzum-ishi krimpatul.
Leaving aside that spooky diacritical mark, only the gh-sound is foreign to an English speaker; it’s the preponderance of closed syllables and the final zg cluster that really achieve the effect. Such expressiveness alone makes phonotactics worthy of serious attention from any conlanger. But it can serve three quite important functions besides aesthetics.
Some sequences of sounds are easier to pronounce than others. A transition between a uvular consonant and a close front vowel is going to require more effort than a transition between a palatal consonant and the same. Most conlangs will try to avoid sequences of sounds that are hard to do right. Certain consonant clusters are tricky to produce for most people, although which ones cause the most trouble varies by linguistic background. Because of this, in addition to aesthetic reasons, nearly all constructed languages put limits on either the composition or the length of clusters. Phonotactics, needless to say, can make speaking and learning a language easier. It can even aid the perception of contrasts; certain sounds are more distinct when they occur in certain positions in a word. When phonotactics serves these purposes, it can be called ergonomic. Perhaps, where necessary, phonotactic rules of an ergonomic nature could collectively be called E-phonotactics (the E could also stand for “easiness” or "ease-enabling").
Most conlangers never exhaustively list the principles guiding the design of their language, even if it is an “engineered language.” One goal that that type of language almost always has, but which seldom gets mentioned, is stability. A language of that type (think of Lojban or Esperanto) should be built so that it will not wear down too quickly with frequent use and lose its more important features. Sound changes are to be guarded against wherever possible, as they generally lead to one or more of three outcomes, all of which are bad. These are the loss of the orthography’s predictability and its one-to-one relationship with pronunciation; the creation of new contrasts difficult for some learners to pick up; or — what is most likely — the loss of contrastiveness between segments, entailing homophony between words that were once distinct. These things are bad enough for any schematic conlang; they are fatal to a logical language.
It appears that sound changes can be limited, if never totally prevented, through good design. Naturally this involves phonotactics.
To illustrate: certain types of assimilation are very common across the world of languages and thus easy to foresee. For instance, nasal consonants, especially /n/, tend to assimilate to the place of articulation of following segments. A language that contrasted /an.ba/ with /am.ba/ would be likely to collapse the two words into the latter form. To prevent this, a language can do a bunch of things. It can have one phonemic nasal in coda position, unspecified for place. Or it can increase the functional load of the /nb/-/mb/ contrast — that is, have lots of words differentiated by said contrast — until assimilation would be too costly in information lost. Or it can take the most common approach, which uses phonotactics preventatively: make /nb/ illegal as a consonant cluster. No native morphemes shall have it; and if it arises as a result of compounding, it must be broken up with an epenthetic vowel. This is exactly what Xorban does. Lojban, as well as Xorban, also takes care to prevent voicing assimilation: the two languages do not allow a voiced obstruent and an unvoiced one to be adjacent. Both require the “buffer vowel” when this would happen in a compound word.
A third example, from Lojban, is the rule prohibiting the clusters /nts/, /ndz/, /ndʒ/ and /ntʃ/. This rule exists because it is very common for speakers to insert a stop between a nasal and a fricative at the same place of articulation. It is a notable rule because it bans the forms that would result from a sound change, as opposed to the forms likely to undergo one. The effect is the same as if /ns/, /nz/, etc., had been prohibited — though the orthography loses a bit of its accuracy as a result.
Rules like those above could be characterized as preventative, precautionary or prophylactic. It is useful to distinguish them as belonging to a specific module of the phonological grammar, which might be called P-phonotactics for short.
Certain rules are not classically phonotactical but may be grouped into this module for convenience. Consider rules like “Two words may not differ solely in the voicing of a single segment.” (This rule is found in Lojban, too, where its scope is root content words.) Aside from adding redundancy that is useful for communication in noisy environments, they also serve to prevent accidental homophony.
A fourth kind of phonotactics features only in the least naturalistic of conlangs. It is most often called “morphology” in the logical-language community, which is not exactly wrong, but a nonstandard use of the term. What is meant by “morphology” here is restrictions on the phonological structure of morphemes that are necessary to ensure morphological self-segregation (that is, unambiguity about where morphemes begin and end in a stream of speech.)
Such rules are crucial to a logical language, since you can’t have any kind of unambiguity without morphological unambiguity. They typically restrict word-initial phonemes, or syllables, to some set smaller than what the phonotactics would otherwise permit — or do the same thing word-finally. Then words can be identified as always starting, or ending, with a member of the set. Something similar is usually done for morphemes within larger words. However, there are a great number of possibilities for how self-segregation can be implemented, and more than one style of rule can be used at once.
This area of design might more properly be called morpho-phonology, but unfortunately that term is more closely associated with rules of phonological alternation in the surface forms of morphemes, such as vowel harmony, and the study of related phenomena. Most engineered languages do not have very much morpho-phonology by that definition, but they do have some; Lojban's affixes, or rafsi, may have several allomorphs that are used in different phonological contexts. In principle, nothing is stopping an engineered language from being rich in allomorphy. For this reason, I believe it is better to come up with a different name for the phonological constraints that function to bring about self-segregation; I propose morpheme-level phonotactics, or morpho-phonotactics, or M-phonotactics for short. In less technical discussion, using the traditional term “morphology” should be fine, too.
Are there kinds of phonotactics other than the four I have described (aesthetic or artistic, ergonomic or ease-enabling, prophylactic or precautionary, and morphological or morphemic)? Phonotactics can be used to serve most any of the goals of a language. It is key to designing for brevity, naturalism and its opposite, and auditory distinctness, among other things not given their own heading here. However, it seems to me that phonotactics has most often been consciously put to use for the four general reasons outlined above. Hopefully this analysis will help to clarify the motivations behind phonological choices, which can be hard to tease out, and may be obscure even to the designer.
Self-segregating morphology for logical languages: An overview
Unambiguous syntax is the sine qua non of a logical language. One way to understand this is that it is always clear where one part of a sentence ends and the next begins. Logical languages since Loglan have tried to implement this criterion at all structural levels, from the sentence down to the morpheme. (For our purposes, a morpheme is a string of phonemes that encodes meaning and cannot be further broken down into meaningful parts. Examples of morphemes in English include dog and the plural marker s in dogs.) It seems that this is somewhat easier to do in written language than in spoken language. Self-segregation in speech typically requires some rather unnatural rules governing the phonological patterns, or shapes, of morphemes.
Various conlangers have written about the ways in which self-segregation can be built into a language. Jim Henry, notably, put together the well-known "List of self-segregating morphology methods" found at Frathwiki and the Conlang Wikia. I have found this list to be invaluable as a starting point. However, it lacks in generality.
At the most basic level, self-segregation can be accomplished in only a few ways. It involves some basic unit and some kind of higher-level unit made up of basic units. I will call the basic units elements. In the schemes most often discussed, elements are either phonological segments (phonemes) or syllables; but they can also be morae, strings of segments of determinate length, or any other linguistic unit. Higher-level units may be morphemes or words (among other things); but for simplicity's sake, I will not distinguish between morphemes and words, and so refer only to words.
The easiest way to have words self-segregate is to fix their length at some number of elements, for all words. This inflexibility is not very attractive: Who wants a language where, for instance, every single word is two syllables long? Most existing languages use what might be termed an "A-B system." This involves classifying elements into two sets, which are loosely speaking word-initial and word-final: Set A and Set B. Words are then defined in terms of these classes, in one of three ways.
1. The left-breaking method: A word is defined phonologically as a string consisting of any element from A followed optionally by any number of elements from B. Words thus come in the patterns A, AB, ABB, ABBB, etc.
2. The right-breaking method: A word is defined as a string consisting of any element from B, preceded optionally by any number of elements from A. Words come in the patterns B, AB, AAB, AAAB, etc. (The prototypical method is found in Xorban, where a word consists of one or more consonants followed by a peripheral vowel.)
3. The bidirectional method: A word is defined as a string consisting of one or more A-elements followed obligatorily by one or more B-elements. Words end at the juncture where a final B-element meets an initial A-element, or at the end of an utterance. They come in the patterns AB, AAB, ABB, AABB, AABBB, etc.
It is worth noting the resemblance of these three possibilities to prefix, postfix and interfix notation, as well as to the head directionality parameter that features prominently in typology. (In fact, The World Atlas of Language Structures' classification of fixed-stress systems into left- and right-headed types inspired my analysis here.) The formal structure underlying such linear arrangements must have received attention from theoretical linguists and mathematicians; I have regrettably not dug into the literature yet.
This is, of course, a simplified picture. The nuances that can be worked in to a self-segregation system are multifarious. One common trick is a class of special joining elements. For illustration's sake, let us use the left-breaking method. We can define a class C whose members link the element to their left to that to their right. Morphemes can now look like ACA, ACAB, ABCAB, etc., in addition to the patterns shown earlier. Something similar is possible with right-breaking and bidirectional types.
Another twist is to have deferred, or long-distance, effects. For instance, we can define a class of joiners, D, that connect the following two elements to the previous element; and another class, E, that connect the following three. This is similar to what Jeff Prothero's "Plan B" language does by having morpheme-initial segments encode the length of the morphemes they belong to.
The resemblance is superficial, however. Plan B has a qualitatively different self-segregation system. There is no A and B class; every segment carries information about how many segments to the right are bound to it when it occurs word-initially. We can call this a length-encoding system. There are many other ways one can be implemented. It does not have to be the first element in a word that encodes the length. Although that is most practical, it is also possible to have a right-headed or retrogressive system in which the last element determines the number of preceding elements. Alternatively, elements might determine the length of a word to the right and to the left. With more complex variations, it is necessary to either determine levels of precedence for the different classes of element, or have all but one of the elements in a multi-element word belong to a null class, carrying no morphological information. Let me sketch an improbable but theoretically possible system like this:
Figure 1: A length-encoding morphology
* Class A: carries no morphological information. * Class B: extends a word zero elements to the left and zero elements to the right. * Class C: extends a word one element to the left and zero to the right. * Class D: extends word zero elements to the left and one to the right. * Class E: extends a word two zero elements to the left and two to the right. * Class F: extends a word one element to the left and one to the right. Example sentence: ACBDAAFABDA --> AC B DA AFA B DA.
Note that, although here, elements determine the length in elements of the same kind, that is not necessary. A leading consonant, vowel or tone can determine the length in syllables or in morae, and so on. Also worth noting is that the now-familiar threefold classification of directionality seems to apply to this type of system as well.
As demonstrated in Figure 1, it is entirely possible to have rules of differing directionality operating simultaneously. This may be done to permit a greater variety of word-forms, to implement self-segregation at multiple levels of the grammar, or for the sake of redundancy. Lojban exemplifies such a hybrid system. It has a deferred right-breaking rule in the form of fixed penultimate stress. On top of that, it has various joining elements, such as the vowel y and heterosyllabic consonant clusters like rp; and something like a left-breaking rule in its requirement that content words have a consonant cluster in the first five segments. More analysis is needed. Questions of interest include: Is Lojban's cluster requirement best understood as a rule concerning segments, segment sequences or syllables? Which rules, if any, are redundant? Which rules belong to each level of self-segregation?
Most existing logical languages draw a distinction between words and morphemes: they have words containing more than one morpheme. This necessitates two forms of self-segregation operating at once (at least — intermediate-level groupings of morphemes within a word are possible, and occur in Lojban). One elegant system, which allows discrimination between three types of morphemes within a word, is presented by Rick Morneau in his essay on morphology. It provides a good demonstration of how right-breaking rules can be stacked. Morneau's approach is to "[ensure] that each type of morpheme can always be identified by its shape, and . . . that each type can occupy only one position in a word." Three morpheme types are possible in his simplified model: prefix, root and suffix. Only suffixes are obligatory; prefixes and roots may appear zero or more times. Prefixes are of the form consonant-semivowel-vowel; roots, consonant-vowel-nasal; and suffixes, consonant-vowel. Altogether, this constitutes a right-breaking system at the word level, with CV syllables the B-elements; and at the morpheme level, with syllable rimes the B-elements. Morneau uses a similar but more elaborate scheme in his magnum opus.
I believe I have covered the most common and useful ways of implementing morphological self-segregation. Needless to say, there is still much more to investigate in this area. Many combinations of methods remain untested. Some may be impossible or practically useless; others may be quite fruitful. Greater complexity seems often to result in more agreeably naturalistic languages.