Confession: I’m bad at morphology when I try to invent it. Most of the conlangs I’ve designed in the past have been fairly isolating because I’m never quite happy with how words look when I try to invent lots of morphology. Everything I make feels either too regular to too arbitrary in terms of how morphology adapts to context. This is despite the fact that I speak languages which are not completely isolating. Here’s the ones I know best (or at least learned in the past), ordered from least to most inflectional morphology:

• English
• Spanish
• Basque

All of them also have a decent amount of derivational morphology. So I’m really not sure where this mental block comes from.

The Jigsaw Puzzle Approach

Along these lines, one thing that always interests me is the way that many languages have conspiracies so that morphemes mostly fit together without the need for complex rules of vowel or consonant merger/deletion/insertion or irregular allomorphs. To take Spanish as an example, almost all verb roots end in a consonant and inflectional endings begin with vowel. A few verbs have lost the final consonant of their root, like creer, but for the most part the generalisation holds.

If we take the verb andar, the root is and-, and then we have endings like -o, -amos, -aba, etc. Where these suffixes are segmentable, they often divide into a suffix that begins with a vowel and ends with a consonant, and then a final suffix that has a vowel and may or may not have a final consonant. An example here would be the imperfective -ab-, which is then followed by person agreement suffixes. The general pattern conspires to produce a legal Spanish word:

• Verb Root = CV(CV)(CV)…C
• Intermediate suffix = –(VC)(VC)..VC-
• Final suffix = -(VC)(VC)..V(C)

Examples with more than two morphemes would be words like and-ab-a and and-ar-emos.

A similar pattern occurs in some other languages. Mayan languages often seem to show this pattern of consonant final verb roots and many -VC- suffixes. Swahili and many other Bantu languages also prefer consonant final roots and then -VC- voice suffixes, with the verbal word ending with a final vowel that marks mood, aspect or tense distinctions. -pend-a “to love” → -pend-an-a “to love each other” is an example.

On the other hand, there are languages like Quechua and Yidiɲ where the verb stem must end in a vowel as a hard rule and (almost) all intermediate affixes also end in a vowel, while potentially contributing a coda to the preceding material. An example from Quechua would be miku-nki-ču, where the intermediate suffix -nki- relies on the preceding stem ending in a vowel to comply with Quechua’s phonotactics.

The Every Morpheme Looks Like a Word Approach

Then there are languages where most of the roots and affixes seem to form a whole number of syllables and could almost stand alone as independent words, so the morphemes can be simply concatenated together with only a bit of assimilation at the boundaries. Tariana is this kind of language, with examples like ma-thuka-i-ta-kade-kaka=tha=pidana=bosa=niki=ka. Because all the morphemes are composed of whole syllables, they require minimal adjustment to form a legal word. Kham would be another example mostly along these lines.

The Fix It Approach

Languages that take none of the above routes would regularly produce illegal words without doing something to remove illegal sequences of consonants and/or vowels. The obvious solutions are:

• Delete/merge some consonants and vowels from illegal sequences
• Regularly insert consonants and vowels to break up illegal sequences
• Have irregular allomorphs and select based on the shape of the preceding/following stem

Tariana is actually a partial example of the merger approach, since it does sometimes merge sequences of vowels (although this doesn’t seem to be an issue for the majority of affixes).

An example of the insertion approach would be Manambu, which has roots and morphemes ending in obstruents, but does not allow most word-internal obstruent clusters. When these occur, an epenthetic schwa is inserted, as in gwaj-ə-kə-tuá.

Allomorphs not generated by regular rules barely appear in some synthetic languages, but seem pretty common in others. A simple example from Nahuatl would be the choice of -huâ or –ê as the possessive suffix on noun, which is mostly determined by whether the noun stem ends in a consonant or a vowel (although there some messiness around body parts). These forms are more similar than the strange conventional spelling rules of Nahuatl might suggest, but I don’t believe the alternation can be explained by regular phonological rules.

Anyway, I know all this, but somehow, whenever I try to create very synthetic non-fusional morphology, it just doesn’t look or feel natural. And yet, every now and then I decide to try again…