Chris in the Weeds

This post is a bit chain of thought, so apologies if it doesn’t quite reach a firm and justified conclusion. It’s another indirectly conlanging related one… Ch’ubmin so far also has incorporated nouns sandwiched between the coverb and verb root, but I’m still not sure why that’s the “right” position for them.

The next topic to ponder in relation to Algonquian and Ojibwe is why the medials are where they are. Algonquian verb stems contain many elements, but the core is a sequence containing one or more of:

Initial – an open class of morphemes, many of which with adjectival or adverbial meanings, which modify the generic action type specified by the final and make it more specific. They may be bound (roots requiring a final affix to be used), or they may be free word-stems.

Medials – Noun-like elements which replace or classify arguments of the verb, especially but not exclusively the patient. Sometimes they are independent noun stems, but often they are special classifier morphemes which do not occur elsewhere in exactly the same form.

Finals – These are mostly affixes which require an initial to attach to, although a few can form complete verb stems with no additional material. They mark distinctions of valency, animacy of the patient/absolutive argument, and they may also encode more specific event type or dynamics. In my previous post, I argued that they were light verbs, although unlike most light verbs they have lost any kind of independent stem or word status. The finals are what make a verb stem verbal, and therefore they are semantically the head of the construction even if the morphological root is the initial.

Some examples with the abstract, mostly bleached final -e which appears to have transitive meanings like ‘have’ and ‘get’:

• naajmiijme naad-miijim-e fetch-food-e “fetch food”
• moon’hapnii moonah-apiny-e dig-potato-e “dig potatoes”

The thing which is bothering me a little bit is that in general, Algonquian morphology is mostly left-branching / right-headed. But if that’s the case, why is the noun-like element in the middle? I would have naively expected it to more naturally come first in such a left-branching language, since the initial+final together mostly define both the verbal transitivity, kind of action, and any selection effects on verbal arguments.

Of course you just say it’s templatic, you don’t have to explain it, but firstly Algonquian morphology does have a greater degree of recursiveness than I’d previously appreciated (you can stack finals in what’s called “secondary derivation”), and secondly, even if there’s no synchronic reason, there was presumably a historical one.

The first option you could consider is that the grouping is [Initial-[Medial-Final]], e.g. [dug-[potato-get]] “to have/get potatoes dug”. The final is mostly the component of the verb stem which has an argument structure, a valency and licenses arguments, so to a certain extent this grouping makes sense – especially since medials are associated with arguments, and if the medial groups primarily with the initial, then what argument structure is it associated with?

But according to Valentine, the initials seem to show stronger selection effects when it comes to medials than the finals do. He says:

Arguments can be made that within the word initials and classificatory medials show a closer connection to each other than either does to the final. For example, certain roots appear to select for the presence of medial classifiers, such as /ginw-/ 'long' and /dakw-/ 'short', whereas other roots such as /aabad-/ 'be useful, be used' and /-ishp-/ 'high' tend to occur in words lacking classifiers. Also, Rhodes 1981 points out that many initials that occur in verbs of breaking tend to be associated with specific classes of objects, such as /book-/, which is especially associated with breaking of sticklike objects, and /baashk-/, which is associated with three dimensional objects. But we rarely find finals that are associated with specific classes of objects, though in some cases, such as the element /-bii/ 'liquid, by means of liquid', as in giiwshkwebii, 'be drunk' (VAI) (with initial /giiwashkw-/ 'be dizzy, disoriented'), the boundary does blur somewhat.

These observations suggest that initials and medials are more closely linked to each other than either is with finals. Perhaps this is true of roots and classificatory medials. On the other hand, certain medials seem to select for a particular form of finals, suggesting a close link between medials and finals. For example, body part medials that end in /n/ often show up with a zero final, that is, with no apparent final at all, as in gzhiibaawgan 'have an itchy back' (M), with body part medial /-aawgan-/ 'back', and gaagiijgondaagan, 'have a sore throat' (M), with body part medial /-gondaagan/ 'throat'.

So it seems like we’re left with the initial+medial forming a secondary adjectival~adverbial predicate together, one which can be classified by a nominal element without losing its original adjectival sense.

In English, indefinite nominals and simple NPs without markers of specific reference can serve as depictive and resultative secondary predicates, although they are less common in this role than adjectives. Consider “he left the meeting a beaten man”… which you might even compound to say he “beaten-man-left” in some alternate dimension version of English.

Other Classificatory Processes in Ojibwe

There are some other aspects of Ojibwe which perform a classificatory function, but sadly the classifiers often don’t seem to have exactly the same form as verb medials, so most initial-medial combinations probably can’t be used as noun stems or as modifiers within NPs without forming part of a verb / relative clause / participle.

This poses a bit of a challenge to the model of medials as “nominalising” or adjectival classifiers, at least as a synchronic analysis of the language.

The main nouny classificatory structures reminiscent of medials that I’ve spotted so far in my attempt to digest 1000 pages of grammar are:

Noun Formation by Initial + Noun Final

Many noun stems are basic / unanalysable, but if I understand correctly:

1. Many noun finals are somewhat classificatory.
2. The initial that such classificatory finals can be attached to may have the same form as verb initials with adjectival senses in some cases, although it may also be an existing noun stem

A comparison of verb medials with classificatory and derivational noun finals shows some overlap. There are a few elements which derive nouns when used as suffixes but which occur as medials within verbs with no change in form like /-aakw/ ‘wood’, and many more where the verb medial and noun final form is similar but not identical: /-min/ vs /-minag/, /-aaby/ vs /-aabig/, /-aabik(w)/, /-iiginw/ vs /-iig/.

Numeral Classification

According to Valentine, when talking about quantifiers:

Traditional Nishnaabemwin has a set of classifiers which classify objects according to their physical properties. The most common classifiers are /-aatig/ 'wooden, pole-like', /-eg/ 'clothlike', /-aabik/ 'metal, glass, plastic, stone', and /-aabiig/ 'string-like', though others exist as well (see discussion in Chapter 9, beginning on page 331). These classifiers are used in counting in the traditional language, basically by adding the appropriate classifier to a numeral root to produce a classifier-sensitive number. Nowadays the most commonly used classifier is /-aabik/, used in counting money.     

Many of the medials used in verbs appear to have exactly the same form as the suffixes used on numerals and quantifiers within noun phrases, so with numerals at least there seems to be considerable overlap between classificatory elements incorporated into predicates and classifiers attached to attributive modifiers within NPs.

Bound Adjectival Classifier Languages

Anyway, this led me a bit to languages with productive adjectival noun classification via bound morphemes / suffixes / compounding, a good example of which is Tariana. Tariana has a large set of classifier suffixes which distinguish animacy, shape, and function, and the set of expandible via the use of suffixed noun ‘repeaters’ when no existing classifier is suitable. According to Aikhenvald’s grammar, adjectives within noun phrases can only be used when suffixed with an appropriate classifier or repeater:

All adjectives require classifier agreement with the head noun. 

...

Example 3.6 illustrates classifier agreement with an underived adjective, and 3.7 shows classifier agreement on an adjective derived from a verbal phrase (dithi sede 'his eye does not exist').

3.6 heku-na hanu-na
wood-CL:VERT big-CL:VERT
'a big tree'

3.7 tʃãri di-thi-sedite
man 3sgnf-eye-NEG.EX+NCL:ANIM
'an eyeless man'

...

5.3.1 The use of repeaters

Not all inanimate nouns in Tariana occur with one of the established classifiers listed in Table 5.1. These classifiers cover about 60 per cent of underived nouns with an inanimate referent. To classify the rest, a different morphological technique is used which consists in repeating the head noun on the modifier. The following semantic groups of underived nouns require a repeater, unless a particular shape property has to be focused:

(a) culturally important artifacts, e.g. ãda 'grater', mukutu, ka:me, kusiwa 'types of baskets', tuda 'fishing net';
(b) important abstract nouns, e.g. daikina 'afternoon', de:pi 'night', and sometimes also ke:ri 'sun, moon';
(c) some body parts, e.g. -daki 'body', -kapi 'hand', -heni 'ear'
(d) occasional loans, mostly from Portuguese, e.g. mesa 'table', pilya 'battery', garavadora 'tape recorder'

However, Tariana mostly doesn’t incorporate adjectives with classifiers into the verb, and nor does it have any kind of adjective-based secondary depictive or resultative predicate which agrees via a classifier. There are a small number of cases when classifiers do make it into the predicate:

• Topic advancing voice

Tariana has an affix which marks a topic fronting construction that resembles a cleft or a relative clause, and which involves affixation of a classifier which agrees with the topical referent on the verb. For example, the following clause has a verbal classifier agreeing with a topical instrument:

ha sipi itʃiri nu-inu-ni-pi-na
DEM:INAN gun animal 1sg-kill-ni-CL:LONG-REM.P.VIS
"This gun was for me to kill animals"

• On verbs in certain kinds of subordinate clause, e.g. verbs inflected for the purposive

Nor have I been able to find any other language with classifiers which uses them (optionally) on depictive or resultative secondary predicates as we’d usually understand them, although languages where adjectives show gender or case agreement often permit or require such agreement in depictives and resultatives too.

Googling this topic is hard though, and I only can only manually check so many language grammars.

Possible Conlanging Implications

One possible adjustment for Ch’ubmin would be to move in the direction suggested by the above vague speculation. Namely:

• Shift towards the coverbs / initials being the ‘root’ and making incorporated classifiers and the closed class verbs more suffix-like… this would be more economical in many ways, since currently most of my verb roots are CVC, but this feels a bit heavy as a firm requirement for a closed class with some high frequency semantically bleached members
• Make many of the coverbs describing stable or resultant state also function as adjectives within the NP (Ch’ubmin as written already has this)
• Have a set of bound classifier suffixes (including repeaters as an explanation for more generic noun incorporation) which occur both as incorporated absolutives within verbs, as nominal derivational affixes and as adjectival agreement / licensing markers

That is, unify the equivalent of Algonquian verb medials and noun finals and treat [coverb-classifier] ~ [initial-medial] combinations as the incorporation of a fully inflected depictive or resultative adjectival modifier. Since Ch’ubmin also already has noun derived coverbs with causal or instrumental nouns, e.g. rain-hit “get rained on”, this would also imply the kind of selectional restrictions on incorporation suggested by Valentine’s grammar for Ojibwe. Coverbs which are stative~adjectival in origin would be compatible with a wide array of classifiers, whereas a number of noun derived coverbs would be more or less semantically incompatible with incorporated classifiers. Compare:

• [split-CL:STRING.LIKE]-cut = cut it split-string = “cut it apart”
• ??? [rain-CL:PERSON]-hit = hit him/her rain-person = “get rained on (for a person)”