How (not) to build a generic verb system
The problem
With this post I’m returning to the previous language in this thread, ‘Proto-Savanna’ (as recently mentioned, the ancestor of Eŋes). As you may recall (but probably don’t), perhaps the most distinctive attribute of this language is that its class of true verbs is closed (i.e., new verbs cannot be created by borrowing or derivation) and relatively small.
Earlier I wrote that there are ~150 true verbs; I feel that number should be revised upwards somewhat, but it’s in the right ballpark.
Of course, the language still needs to express verbal concepts beyond this set of basic roots. ‘Proto-Savanna’ has two means of expanding its verbal inventory: serial verb constructions (SVCs) and verb adjunct constructions (VACs). This example shows both:
bradrn wrote: ↑Thu Jun 23, 2022 11:21 am
- Be
- 1s
- fasi
- stand.PFV
- sasay
- home
- thaŋ
- DEF.SG
- waalhi tsagif
- go.PFV running
- qaathan
- fall.PFV
- fawetli
- speak.PFV
- feqe.
- cry
I ran down away from home and cried.
(Reminder: ⟨q g e lh th⟩ are /ʔ ɣ ə ɬ tʰ/; the rest should be obvious.)
This post focuses on verb adjunct constructions, like the above
*walha tsagif, lit. ‘go running’ = ‘run’. In these constructions the main verb belongs to a smaller set of ~20 verbs with highly generic meanings, while its complement provides the specific verbal meaning. There is a class of ‘coverbs’, which occur exclusively as verb adjuncts, but they can also be nouns, adjectives or ideophones. The verb adjunct is not a direct object of the verb, as VACs can take another NP as the direct object. It is not an NP itself, although to some extent the verb adjunct can be modified by adjectives or quantifiers.
Since the verb class is closed, VACs (and SVCs) are used to express a wide range of meanings where there is no suitable true verb. And since the verb class is small, this extends even to seemingly basic meanings — thus we have
*lhiise meŋul ‘see’,
*siwe tsisa ‘want’,
*walha naŋa ‘walk’.
However, there is a problem. As should be obvious by now, ‘Proto-Savanna’ tends very strongly towards disyllabic roots. The consequence is that most verb adjunct constructions are four syllables long. And I’m increasingly coming to believe that this state of affairs is simply impractical for a human language. No-one wants to take the time saying /ɬiː.sə.mə.ŋul/ when they just want to say they saw something. (And
zompist agrees.)
(There is actually a single verb which is monosyllabic:
waq~mah ‘do, make’. This is also the most common verb for VACs, which helps somewhat. But on its own this isn’t nearly enough to fix the problem.)
At this point it’s worth looking at some comparable natlangs. VACs are probably most widespread in north-western Australia, where many languages have small, closed verb classes, alongside an open class of coverbs. The most famous case is probably Jingulu, with only three full (inflecting) verbs:
-ardu ‘go’,
-jiyimi ‘come’ and
-ju ‘do’. (Most VACs occur with ‘do’ there.) However, this is an extreme case. Within the region, VACs are probably best studied in Jaminjung, with 35 full verbs (
Schultz-Berndt 2000). These full verbs mostly have highly generic meanings: for instance
-irdba ‘come to be in a locative relation’,
-ijga ‘move/extend along a path / to a state / be involved in sth. for a long time’,
-mili / -angu ‘be in contact or in the same place affecting’. These combine with adjuncts to encode, for instance,
jag -irdba ‘fall down’,
digirrij -ijga ‘die’,
ngalyag -angu ‘lick’.
As with ‘Proto-Savanna’, we can see that many of these verbs and adjuncts are disyllabic or even longer. (This is not unusual for roots in Australian languages, which impressionistically are often rather verbose.) However, many are monosyllables also, as in Jaminjung
birl -ma ‘blow off’. Also note that verbs mark various categories as well — subject, object and TAM in Jaminjung — and often with stem suppletion, so these verb stems can carry more information than those in ‘Proto-Savanna’. (The extreme case is probably Murrinhpatha, where 39 verbs occur in 1638 separate verb forms with few regularities.)
Furthermore, it is common to use verbs without an accompanying adjunct. In Jaminjung, Schultz-Berndt states that this occurs in 40% of clauses. In such cases the specific meaning of the verb must be inferred from context: ‘the verb
-ijja ‘POKE’ can be read as ‘spear’ in a kangaroo hunting context, as ‘dig with digging stick’ in a yam digging context and as ‘stab’ in a knife fight context’ (p118). By contrast, in ‘Proto-Savanna’, when a specific meaning can be expressed using a VAC, I’ve tended to stick with that construction: for instance I’ve always expressed ’see’ as
*lhiise meŋul, never just
*lhiise. Being more flexible on this would already be enough to make the system more usable.
(For much more on verbs in north-western Australia, see McGregor’s 2002 book,
Verb Classification in Australian Languages.)
Outside Australia, VACs are widespread in Papuan languages. This is especially the case in Trans-New Guinea, where even open-verb-class languages often use them for borrowed verbs. My impression is that most TNG languages have far more verbs than Australian languages do: for instance, Kalam has a closed class of ~300 verbs, but still makes very extensive use of VACs to express meanings like
wdn nŋ- ‘eye perceive = see’,
suk ag- ‘shouting say = shout’,
dad amn- ‘carrying go = carry’. (Kalam in particular also uses SVCs, like ‘Proto-Savanna’.) As in Australian languages, the verbs which occur in these constructions have highly generic meanings.
In Kalam, a great many of both verbs and adjuncts are monosyllabic, and it seems rare for both to be disyllabic. There is no explicit information on the use of generic verbs by themselves, but at least in some cases it appears very common: for instance, in
Kalam Hunting Traditions (Majnep & Bulmer 1990), I can find no instances of
wdn nŋ- for ‘see’, which appears to be expressed with
nŋ- alone. (In the
Kalam Dictionary, sample sentences for
wdn nŋ- seem to have a rather emphatic meaning — ‘see with my own eyes’, and similar.)
Outside Australasia, VACs are widespread in Indo-Iranian and Dravidian languages. However, the only closed-verb-class language I’m aware of in the region is Kurmanji Kurdish, where the construction occurs with only three main verbs. In other languages such as Hindi-Urdu, VACs occur with more verbs than this, but they don’t take on anything like the importance they have in Australian or Papuan languages (or even Kurdish) with their closed verb classes.
I find it interesting that all the above languages are verb-final, unlike ‘Proto-Savanna’. Are there any verb-medial languages with true VACs? It appears not. However, Ewe (which has a closed verb class) has something similar in the form of ‘inherent complement verbs’ (
Essegbey 1999): highly generic verbs which obligatorily occur with a specific nominal complement. But unlike a true VAC, the objects here are syntactically just ordinary direct objects, making them similar to English light verb constructions like ‘have sex’ (my go-to example because English has no single-verb equivalent, similar to Ewe).
So, the problem I must solve is: how do I make the ‘Proto-Savanna’ verb adjunct system less impractical, with as little change as possible? And, secondarily, how do I square this with the fact that it’s a verb-medial language?
Whence did it come?
I am of the opinion that even very unusual or unstable linguistic features are fine in a conlang (especially a con-protolang), as long as they can be justified diachronically. After all, such things
do turn up on occasion — just look at the PIE stop system, or the marked-nominative case system in Old French and Old Norse. None of those systems lasted for long, but they did all exist at some time, and there’s no reason why ‘Proto-Savanna’ couldn’t have been in a similar situation. (If anything, I
like having an unstable protolang: it gives me more opportunities for descendants to go different ways.)
So, what are the possible origins of VACs? It’s tempting to look at ordinary V+O idiomatic constructions, present in many languages including English (as mentioned). But very few languges seem to have turned those constructions into VACs where the adjunct is not a true object.
The key seems to be the prior presence of some other VAC-like verbal construction. For instance, Haig (
2002) traces the origin of Indo-Iranian VACs back to IE particle verbs. He suggests that various structural factors conspired to make nominal and adjectival objects behave similarly to particles, such that it was easy for the particle verb construction to be generalised to VACs — a tendency which was encouraged by the paucity of methods for deriving new verbs. Unfortunately, this particular instance doesn’t really help with ‘Proto-Savanna’, which has no analogue to IE particle verbs.
In North Australia, VACs are extremely old and widespread. However, McGregor has noticed that coverbs in these languages possess some interesting properties:
- They are much more likely to be monosyllabic than other parts of speech: across Warrwa, Gooniyandi and Gunin, ~30% of adjuncts are monosyllabic, compared to only 1–3% monosyllables in other parts of speech.
- They are much more likely to end in consonants or contain consonant clusters than other parts of speech. In some languages these consonants appear to be phonaesthetically motivated: e.g. in Gooniyandi, stops are used with ‘abrupt activities’, and continuants with ‘continuous or iterative events’ (McGregor 2002, mentioned above).
- Some languages allow coverbs to be used independently with an expressive meaning, outside their normal syntactic position.
- Of the limited modification allowed on coverbs, reduplication is the most common process.
- In some languages there are elements which can refer either to quotes or to coverbs (e.g. Gooniyandi yiniga ‘how, in what manner’).
For these reasons, McGregor posits that VACs in Australian languages derive ultimately from
ideophones — words denoting ideas in a sound-symbolic fashion. He suggests a historical process in which ideophones, originally restricted to minor clauses and used for expressive purposes, become increasingly common in speech. As they occur more often next to verbal clauses, ideophones become associated with main verbs, creating a proto-VAC which becomes increasingly common in the language, at the expense of simple verbs.
(It’s worth noting that Ewe makes extensive use of ideophones too. Why have they not triggered the evolution of VACs there? I think it’s because ideophones don’t form a single word class in Ewe, and thus have no distinctive syntactic behaviour; instead they are distributed across the classes of adjectives, verbs, nouns, etc. In this light, the issue of OV vs VO word order may not be as big an issue as it seemed.)
Very conveniently, ‘Proto-Savanna’ has ideophones! And not only that — I already specified that those ideophones are most often used as verb adjuncts. Some Eŋes verbs are even derived from verb+ideophone combinations already. Thus, ‘Proto-Savanna’ already has the perfect conditions for developing widespread VACs. (Or near-perfect: Haig noted that the development of Kurdish VACs was aided by a lack of articles, but I feel comfortable handwaving this away.)
In fact, not only does this give me a justification for the system, it even gives me a way to bypass the bisyllabicity issue with minimal changes. As mentioned, ideophones are phonologically unusual in many languages, and ‘Proto-Savanna’ is no exception. So far I’ve made most ideophones reduplicated, but there is already at least one monosyllabic ideophone. It’s easy enough to change the situation to make such ideophones the usual type, then use those ideophones as verb adjuncts. Indeed, the protolanguage might not even need a separate class of coverbs — they could all just be ideophones. (Though I’m not sure I love the idea of going that far.)
Whither can it go?
Even though this system is now justified, and can be made less impractical than it was, I still think it’s somewhat unstable. How could it evolve in the descendant languages?
Looking at Australian and Papuan languages, it seems that there are basically three possible outcomes:
- The system continues more or less as-is, with a variety of simple verbs, only some of which are generic and occur in VACs.
- The set of true verbs is much reduced, to perhaps only a few dozen, nearly all with highly generic meanings; new coverbs appear to replace the old verbs.
- Verb and adjunct both become obligatory, forming a bipartite stem, or being reduced even further to opaque and/or unproductive conjugational classes.
Most Papuan languages have a system like (1). The languages of northwestern Australia mostly have some form of (2): in some cases with the verb and adjunct being distinct elements (even with some freedom of word order), or in other cases having them integrated into a single word.
By contrast, Eŋes has undergone something like (3). But it’s an unusual example of the type, because the verb and adjunct fused at a fairly early stage: massive syncope obscured the previous system of verb roots, leading to verb+aspect+adjunct combinations being lexicalised as a new (and open) verb class. The forms of the original verbs can still be distinguished, just barely — e.g. the widespread /w-/ from previous
*waq ‘do, make’, or /siw-/ and /rw-/ from
*siwe ‘have’. But their original semantics have largely been overridden by the lexicalisation of the aspect system, which have come to encode similar notions of
Aktionsart and/or manner.
Intriguingly, McGregor suggests that the verbal conjugation system of Pama-Nyungan languages may also have come from (3). Most Pama-Nyungan languages have a set of 2–6 conjugation classes, distinguished partly by the first consonant of verb suffixes. McGregor’s suggestion is that these conjugation classes may ultimately be the remnant of a small set of inflecting verbs, with the modern verb class deriving from the verb adjuncts. This seems plausible to me, but the original situation is probably unreconstructable (unless by some miracle Pama-Nyungan is ever linked to one of the other Australian families).
Another unusual (3)-like outcome is found in Gooniyandi. This language still has distinct adjuncts and verbs, but the relation between them has been flipped: the former adjuncts are now the obligatory verbal element, and the former verbs are now optional classificatory morphemes. McGregor suggests that this could be a precursor stage to the Pama-Nyungan–type conjugational classes.
In any case, there’s enough possibilities here to inspire many years of happy conlanging!