Axes of inflection classes

[Kuchigakatai]

If a language has multiple classes for inflecting nouns, adjectives or verbs, what are some surface phonological characteristics that these classes can be identified by?

You don't need to mention irregular words, as it can be safely assumed that languages with inflections have irregular members. In the examples below, I only talk about nouns/adjectives/verbs that can be called regular.

- - - -

Latin nominal declension has classes loosely based around its five native vowel qualities /a o i u e/, using the vowels a/ā, u/ō, i/ē/ī, u/ū and e/ē as vocalic themes respectively.

Synchronically speaking, the main conjugation classes of Latin verbs are based on four particular vowel themes: ā, ē, e/i, ī. The third vowel theme, e/i, may or may not insert an extra short -i- in some of the forms depending on what verb it is, creating two subtype classes of itself (i.e. whether there is the extra -i-). And so the five classes come to be. (However, this only applies to infinitive-based and present-based TAMs, as Latin has no conjugation classes in perfect-based TAMs, where the verbs rely purely on unpredictable, lexically-determined stem changes instead.)

Spanish has three main verbal classes based on vowel themes in its endings: a, e, and i. However, there's also a further orthogonal axis involving roots whose vowel inside changes when stressed, /o/ > /ue/, /e/ > /ie/, or /e/ > /i/, which involves so many verbs that the two axes together, ending vowel theme + stressed root vowel change, should probably be considered when talking about the regular verbs of Spanish. (The traditional convention is to say that all those several hundred verbs with a simple vocalic stem change are "irregular".)

Japanese verb conjugation classes are based on the type of consonant at the end of the stem, if any: velar (-k/g), nasal (-m/b/n), sibilant (-s), stop-and-approximants (-t/r/w), and vocalic (-i/e). I should note that most derived verbs, which are naturally a lot, consistently use a special suffix -suru that has /s/ and is the same as the sibilant conjugation in some forms, but forcefully contains /si/ [ɕi] in others.

Standard Arabic's nominal declensions are a bit on the weird side, since some core suffixes are really their own declension, but the following can be identified:
- triptote: the default declension, includes most nouns and adjectives
- -aani: the only dual suffix
- -uuna: a plural suffix (used often with human masculine nouns)
- -aatun: a plural suffix (used often with human feminine nouns, also many deverbal nouns and borrowed nouns)
- diptote: words in the singular ʔaCCaC-u transfix, also words in the plural CaCaaCiC-u and CaCaaCiiC-u transfixes (e.g. falaafil-u 'peppers' i.e. the origin of English "falafel", balaayiin-u 'billions'), also the feminine-singular or masculine-plural suffix -aaʔ-u
- defective with /i/: words with /j/ as the last root consonant without a further suffix
- defective with /a/: words with /j/ as the last root consonant in a transfix that alters it with a short /a/
- a few (IIRC only seven) very short nouns (e.g. ʔab-un 'father')
- invariable: no inflection for case or definiteness
In Arabic linguistics, "defective" is not used the usual way. It doesn't refer inflection holes with missing forms, but seemingly-missing consonants in the spelling of words. Latin dēficiō, literally "off-do", can means 'to take away [sth]', so the root-final /j/ consonant is "taken away" from the spelling.

Meanwhile, Standard Arabic verb conjugations are mostly classified around the consonant root structure, with geminates or the glides /w j/:
- regular
- geminate: the 2nd and 3rd consonant roots are the same, e.g. r-d-d in radda 'to respond'
- glide-initial: e.g. w-j-d in wajada 'to find [sth]', present-active stem -jid- (/w/ dropped)
- glide-medial: e.g. q-w-l in qaala 'to say', present-active stem -quul-
- glide-final: e.g. b-n-y in banaa 'to build [sth]', present-active stem -bnii-
(Traditionally, roots with /ʔ/ are also talked about separately, but this is due to difficulties in spelling the forms, which are otherwise very regular phonemically.)

I'm aware that in the study of Biblical Hebrew there are similar categories, but I don't know whether they all really matter, i.e. whether they produce inflections that are unpredictable from the regular pattern and phonological rules. As I understand, the categories are:
- regular
- geminate (2nd and 3rd are the same: X-Y-Y)
- (/ħ ʕ/) pharyngeal-initial
- /ʔ/ glottal-stop-initial
- /n/ nun-initial
- (/w j/) glide-initial
- pharyngeal-medial
- glide-medial
- (/ħ ʕ h/) guttural-final
- glottal-stop-final
- glide-final
- verbs spelled with a final unpronounced he consonant.
(This last category is traditionally called something like he-final, but the /h/ is not pronounced, and if a verb has a final /h/ that is pronounced then it's classified under the pharyngeal-final verbs... hence why I renamed that category "guttural-final".)

[zompist]

Sanskrit divides verbs into 10 classes based on how they form the present indicative: same as root, various suffixes, reduplication, inserted nasal, etc.

Russian has two conjugations, in ать and ить. Plus a bunch of semi-regular variations generally involving infixes. Plus the chaos of the perfective.

The Spanish verbs with o > ue and e > ie are sound changes, which also affected French and Italian, but not Portuguese. I'm not sure how many would be regular if you knew the Latin root and had a good grasp of Romance historical linguistics.

[Richard W]

(However, this only applies to infinitive-based and present-based TAMs, as Latin has no conjugation classes in perfect-based TAMs, where the verbs rely purely on unpredictable, lexically-determined stem changes instead.)

Actually, the Latin ā, ē and ī conjugations mostly have predictable perfect forms; it's just that there is the opportunity to deviate in the perfect.

I'm aware that in the study of Biblical Hebrew there are similar categories, but I don't know whether they all really matter, i.e. whether they produce inflections that are unpredictable from the regular pattern and phonological rules.

The inflections are mostly predictable. Lamedh he and lamedh ayin have some special forms, and glide medial are a bit odd. Nun- and glide-initial have some oddities due to the initial consonant simply vanishing in some forms, notable imperatives and infinitives.

You forgot one other axis - the vowel between the second and third consonants, roughly correlating with active v. stative. It becomes quite significant when one includes the object suffixes. It's vaguely phonological, like the difference in the noun inflection between final syllable vowels that can take the accent and the ones that weaken instead.

Hebrew nouns have perhaps 3 basic structures depending on the vowels in the CVCVC frame. There are the segholates, stressed on the penultimate and typically having -e- as the final vowel (whence their name), those that keep the the stress on the original ultimate when a 3rd person pronoun is suffixed to the sinɡular, and those that move it to the penultimate. Those with fixed stress have the vowel pattern CV(C)CāC or CāCēC, or CV(C)CīC or CV(C)CūC. The vowel that loses stress is ē or ō.

Pali likewise works by stem vowels, with two axes in the nouns - lexical gender and phonological stem vowel. Gender is the more significant. In the verb, the contrast between the -a- and the -aya-/-e- is most striking in the 2s imperative active (-hi optional or not) and the present participle (the -a- and -aya- forms can have the consonant stem nominative singular). The -aya- alternative mostly works as the -a- form. There are also verbs in -o-, but I'm tempted to call the lot of them irregular. Adjectives in -mant-/-vant- have masculine nominative singular in -mā/-vā, while present participles in -ant- have masculine nominative singular in -aṃ (as well as the thematic forms in both cases).

In Sanskrit, athematic verbs can have an ablauting stem vowel or have a consonant before the 3s ending. Reduplication is one indicator of a 3rd person plural imperfect middle in -r - I forget the gory details. Consonant noun stems have subtleties depending on which consonant it is, but that doesn't affect the choice of endings. The vocalic classes do have different endings depending on the vowel.

[Estav]

The prosodic structure of a word can affect how it inflects. In German, the endings -e /ə/ and -∅ are in complementary distribution at the end of plural nouns: nouns that end in /ər/, /əl/, /ən/ or /əm/ in the nominative singular may take plurals in -∅, but cannot take plurals in -e, while nouns that end otherwise may take plurals in -e, but cannot take plurals in -∅. (German has other plural suffixes aside from these.)

In Latin, the small category of -io third-conjugation verbs originated historically as a short alternative to the fourth-conjugation ī-inflection; the long vowel was used after heavy roots (those ending in VVC or VCC), while the short vowel was used after light roots (roots ending in VC). I don't remember the exact details, or whether this conditioning factor could still be considered synchronically relevant by the time of Classical Latin. Light vs. heavy roots affecting the development of inflectional suffixes was also a thing in Old English, both in verb and noun inflection, although I remember even less of the specifics here.

[Richard W]

Sanskrit divides verbs into 10 classes based on how they form the present indicative: same as root, various suffixes, reduplication, inserted nasal, etc.

However, I'm not aware of any method of predicting the class from the root. Given the 3s singular present, the four thematic classes all have the same endings, except that the position of the unvarying stress may be different. Similarly, the two of the athematic classes, those in -nu- (5th) and -u- (8th), differ only by whether the -n- is part of the root, apart from the slightly irregular karoti. The reduplicating (3rd) and suffixless (2nd) class have nothing between root and affix, so might need subdividing by final consonant.

Russian has two conjugations, in ать and ить. Plus a bunch of semi-regular variations generally involving infixes. Plus the chaos of the perfective.

There are also verbs (with present endings basically as -ать) where the velar stem consonant palatalises. I don't know whether palatalisation or non-palatalisation is regular. Additionally, the 1s of labial stems inserts л.

[Richard W]

In Latin, the small category of -io third-conjugation verbs originated historically as a short alternative to the fourth-conjugation ī-inflection; the long vowel was used after heavy roots (those ending in VVC or VCC), while the short vowel was used after light roots (roots ending in VC). I don't remember the exact details, or whether this conditioning factor could still be considered synchronically relevant by the time of Classical Latin.

The 4th conjugation verb saliō 'to jump' doesn't follow the rule.

[zompist]

Sanskrit divides verbs into 10 classes based on how they form the present indicative: same as root, various suffixes, reduplication, inserted nasal, etc.

However, I'm not aware of any method of predicting the class from the root.

No, but Ser didn't ask for that. I don't know if there's a single form that can predict all the others, though the present root should come close.

[Estav]

In Latin, the small category of -io third-conjugation verbs originated historically as a short alternative to the fourth-conjugation ī-inflection; the long vowel was used after heavy roots (those ending in VVC or VCC), while the short vowel was used after light roots (roots ending in VC). I don't remember the exact details, or whether this conditioning factor could still be considered synchronically relevant by the time of Classical Latin.

The 4th conjugation verb saliō 'to jump' doesn't follow the rule.

That's interesting. It looks like there are some more exceptions (including the very common venio), and Baldi 2002 says the explanation of the capio type as resulting from Sievers' law is disputed.

[dhok]

Russian has two conjugations, in ать and ить. Plus a bunch of semi-regular variations generally involving infixes. Plus the chaos of the perfective.

Er...kind of but not really.

First, there are really two stems to the Russian verb: the non-past (future perfective in perfectives, present in imperfective) and the past. The infinitive is the always the basis of the past stem, which gives you your past tense in -л/-ла/-ло/-ли: жить 'to live' > жил '(he) lived.' The past tense is sometimes subject to stress alternations, as жила́ 'she lived', жи́ли 'they lived'. (There are also a few minor forms like participles that are built on this stem).

The non-past stem is sometimes predictable from the infinitive, in "regular" verbs in -ать and -ить, but it's best to treat it as entirely orthagonal to the past/infinitive stem: жить is first-conjugation живу́/живё́шь/живё́т, and спать is second-conjugation сплю/спишь/спит. (A somewhat analogous situation can be found in Latin, where most a-stem first-conjugation verbs have completely predictable perfects, but most thematic third-conjugation verbs don't). First-conjugation verbs conjugate like this:

	Singular	Plural
1	-ю/-у	-ем
2	-ешь	-ете
3	-ет	-ют/-ут

Second-conjugation, like this:

	Singular	Plural
1	-ю/-у	-им
2	-ишь	-ите
3	-ит	-ят/-ат

There's yet another catch, though. Orthagonal to the conjugations in the non-past is stress alternation. Each verb--of either conjugation--belongs to one of the following three classes in the non-past:

* stress class A: Stress is always on the root. Most "regular" verbs in -ать are like this: спра́шиваю, спра́шиваешь, спра́шивает 'I, you, s/he asks.'
* stress class B: Stress is always on the ending. Example жить: живу́, живё́шь, живё́т 'I, you, s/he lives'.
* stress class C: Stress is on the ending in the 1sg and on the stem everywhere else. Example сказа́ть 'to say (perfective)': скажу́, ска́жешь, ска́жет 'I, you, s/he will say.'

If I recall correctly, there's a rather tenuous connection to the infinitive stem: if a verb isn't stressed on its final syllable in the infinitive, its non-past is going to be class A. But don't quote me on that.

To complicate matters, in the first conjugation, the -e- of the 1pl, 2sg, 2pl and 3pl becomes -ё- /'(ʲ)o/ when stressed. Otherwise, it's just /ɪ/ and merges on the surface with the stem-stressed forms of the second conjugation, at least after consonants (second-conjugation verbs [almost?] always have consonant-final stems).

OK, we haven't touched the question of aspect. Aspect is sort of this weird chimera between a derivational and an inflectional process. Essentially:

* most basic verb stems, but by no means all, are prototypically imperfective. Example: просить 'to ask for'.

* adding a prefix turns a basic verb stem perfective. This may be a semantically null prefix added to the base stem, most frequently по-: попросить 'to ask for (perfective).' Or, it may be a derivational prefix that creates a new verb: спросить 'to ask [as a question] (perfective)'.

* because prefix-derived verb stems are usually perfective, their imperfectives are secondary derivations from the perfective root. This is usually done with the suffix -ывать/-ивать, which creates a 'regular' first-conjugation verb: спрашивать 'to ask [as a question] (imperfective).' (as we can see the stem has changed irregularly here, but the process is the same you see in dozens and dozens if not hundreds of verbs)

* finally, a few verbs just have irregular perfectives, like imperfective говорить and perfective сказать.

[vegfarandi]

Icelandic has 15 declensions, each with several minor categories, resulting in a total of 96 paradigms

The top-line distinctions are weak (genitive ends in vowel) or strong (genitive ends in consonant) + gender: masculine, feminine or neuter.

The minor variations sometimes depend on factors such as thematic vowel, the stem vowel (if it's a and the ending contains -u-, the a becomes ö due to u-umlaut), but most of the time they're lexical. For instance, strong masculine nouns can end in -ar or -ir in the nom. pl., and -a or -i, respectively in the acc. pl., which pair of endings must be learned by rote. For many masculine nouns, whether there is a dat. sg. ending -i or not is also unpredictable. Familial terms ending in -ir such as móðir 'mother' and bróðir 'brother' have a unique paradigm (although with different ablaut depending on stem vowel) that are the same despite being different genders.

Some things are fairly easy: the dat. pl. always ends in -um, the gen. pl. always ends in -a, although some nouns have a an -n- before the -a. Aaaaaand of course u-umlaut may apply in the dative due to the -u- (kettir 'cats' (nom. pl.) > köttum (dat. pl.)) and there may be vowel deletion before a preceding -r-: feður 'fathers' (nom. pl.) > feðrum (dat. pl).

Essentially, Icelandic declension is a nightmare for any learner.

And conjugation is not much better. There are weak and strong verbs, and minor categories such as preterite-presents and -ri verbs, but within each are multiple sub-categories. Weak verbs have different past tense endings (-di, -ti, -ði, -aði) which are somewhat phonologically conditioned, but also not predictable. Strong verbs are just a dumpster fire of unpredictability, and four principle parts must be memorized to derive all forms.

[Linguoboy]

Irish verbs are divided into two conjugational classes based on the presence or absence of long vowels in the non-past inflectional endings.

First conjugation: campálaim "I camp", cuirim "I put", creidim "I believe"
Second conjugation: ceanglaím "I bind", ceannaím "I buy", crúim "I milk"

Paedagogical works sometimes generalise that first conjugation verbs have monosyllabic stems whereas second conjugation verbal stems are disyllabic, but exceptions abound. All verbal stems ending in the productive derivational suffix -áil, for instance, are first whereas the second conjugation has syncopated verbs like crúim (2s.IMP crúigh), caím (2s.IMP caígh), etc.

Irish nouns have five declensional classes on the basis of how they form the genitive. I can't think of any internal unifying phonological characteristics. Members of the first class all end in broad consonants but final consonants can be broad or slender in the other classes, and members of both the fourth and fifth classes can end in vowels. (The fifth declension is rather like the Latin third, a grab-bag of historical consonantal stems.)

[Frislander]

Speaking of the prosodic factors, there are of course the Australian languages that have different forms of certain inflections depending on whether the root has an even or odd number of syllables, or whether is has two or more syllables, e.g. Yidiny gali > galiiny "go > went" vs. majinda > majindanyu "walk up > walked up".

In fact you could also say the same about Finnic's rhythmic gradation.

Neither of these are proto-typically "inflection classes" since they are generally lumped with agglutinative analyses, but they still count since they are factors that influence the forms in a paradigm.

[Salmoneus]

Icelandic has 15 declensions, each with several minor categories, resulting in a total of 96 paradigms
[...]
And conjugation is not much better.

Germanic is a nightmare in this respect in general.

In my Germanic conlang, Old Wenthish, I intentionally tried to simplify things - lots of analogy, and you never have to learn principle parts independently, they're all predictable from the conjugation...

...and there are still 26 verb conjugations (17 strong and 9 weak). Not including irregularities.

[Kuchigakatai]

The Spanish verbs with o > ue and e > ie are sound changes, which also affected French and Italian, but not Portuguese. I'm not sure how many would be regular if you knew the Latin root and had a good grasp of Romance historical linguistics.

The evolution of Latin > Spanish is quite irregular though. I wouldn't rely on the Latin etyma much.

For example, from Latin percontārī we'd expect *pregontar *pregüento, but Spanish has preguntar pregunto (although this might reflect a survival of the earlier Latin stem percunctārī). From Latin respondēre we'd expect responder *respuendo, but Spanish has respondo. From Latin contestārī we'd expect contestar *contiesto (cf. testa > Old Sp. tiesta), but Spanish has contesto.

In the opposite direction, from Latin findere we'd expect hender *hendo but Spanish has hiendo. From Latin pluere, infl. by pluvia, we'd expect llover *llove, but Spanish has llueve.

Most -ir verbs with historical Latin /ɛ/ in the stem end up with e > i, e.g. petere > pedir pido and vestīre > vestir visto, but there's the few exceptions of sentīre > sentir siento and venīre > venir viene (and, for other reasons, adquaerere > adquirir adquiero).

And then verbs borrowed from Latin or based off a borrowed Latin noun make the picture worse, as they don't normally take diphthongization, e.g. intentāre > intentar intento (cf. tentāre > tentar tiento), interrogāre > interrogar interrogo (cf. rogāre > rogar ruego).