Revalency

This page is Copyright John N. Shutt 2007, 2012, 2015–2017.
Last modified:  9 March 2017.

Introduction

In David Peterson's essay Ergativity (which I recommend), he describes a peculiar case system he calls duative-unitive (DU), and which contributors to the CONLANG mailing list call Monster Raving Loony (MRL).  There seems to be general consensus that DU/MRL is a pathological language development that doesn't serve any useful function, so I wanted to work out a scenario in which it would evolve non-pathologically to serve a useful function.  Ideally, the scenario would be within the bounds of credulity.  One product of my efforts is a case system I'm calling valent-revalent (until a better name comes along), a noun-marking-based form of a more general technique I'm calling revalency.  Revalency serves roughly the same purpose as the passive in accusative languages and the antipassive in ergative languages.  The valent-revalent system can exhibit a duative-unitive pattern, and comes equipped with a backstory of how it evolved from the passive form in an earlier accusative system.

This essay explains the valent-revalent system, including its backstory; and also (briefly) discusses other possible forms of revalency.  Since I'm not presenting a whole language here, I'll illustrate language features using Ergato, the toy language that Peterson used to illustrate the many variations on case systems in his essay.  I'm also not looking for an airtight historical progression, just a sketch of a process that might produce certain features, to show that a more carefully and thoroughly designed history for a conlang would have at least one strategy available to explain where those features came from.

Contents

Ergato

Table 1 lists about half the vocabulary, and all the inflections, of Ergato.  (Peterson's essay provided another two verbs, three nouns, two pronouns, and a preposition.) 

Table 1: Ergato words and suffixes.
Words Words Words Suffixes
 English   Ergato 
to sleep
to pet
to give
sapu
lamu
kanu
 English   Ergato 
panda
woman
book
man
fish
palino
kelina
kitapo
hopoko
tanaki
 English   Ergato 
she li
 English   Ergato 
Valency Reduction
Past Tense
Plural
Default Case
Special Case
 Recipient/Dative Case 
Oblique Case
Extra Case
-to
-ri
-ne

-r
-s
-k
-m
 English   Ergato 
and i

In a nominative-accusative (NA) system, a transitive verb has a nominative "default case" subject, and an accusative "special case" object; the sole argument of an intransitive verb gets the default case, nominative.  In an absolutive-ergative (AE) system, a transitive verb has an ergative "special case" subject, and an absolutive "default case" object; the sole argument of an intransitive verb again gets the default case, absolutive.  In a duative-unitive (DU) system, a transitive verb has duative "default case" subject and object, while the sole argument of an intransitive verb has the unitive "special case".  Table 2 illustrates all three systems, using the more orthodox of two DU systems Peterson describes; for comparison I've shown them all with VSO order (Verb, Subject, Object), although I tend to think of the nominative-accusative system as SVO, possibly because I'm used to English (but, then again, possibly because I'm sensing something structurally fundamental), and Peterson says he finds SOV natural for an absolutive-ergative system.  I think I agree with Peterson about SOV for absolutive-ergative, but for valent-revalent (where we're headed) I believe VSO is more natural, for reasons I'll try to clarify later.

Table 2: Basic sentences in NA, AE, DU.
English NA Ergato AE Ergato DU Ergato
The woman is sleeping.
The woman is petting the panda.
The woman is giving the book to the panda.
Sapu kelina.
Lamu kelina palinor.
Kanu kelina kitapor palinos.
Sapu kelina.
Lamu kelinar palino.
Kanu kelinar kitapo palinos.
Sapu kelinar.
Lamu kelina palino.
Kanu kelina kitapo palinos.

For valency reduction in the NA and AE systems, the valency-reduction suffix is attached to a transitive verb, the argument with default case is dropped, the other argument changes to default case, and the dropped argument may be restored at the end of the sentence as an oblique.  In NA, the valency-reduced form is called passive; in AE, antipassive.  Table 3 illustrates valency reduction in those two systems.

Table 3: Valency reduction in NA, AE.
English English
The woman's petting the panda.
The panda's being petted
   (by the woman).
The woman's petting the panda.
The woman's petting
   (and what she's petting is the panda).
NA Ergato AE Ergato
Lamu kelina palinor.
Lamuto palino (kelinak).
Lamu kelinar palino.
Lamuto kelina (palinok).
English English
The woman's giving the book to the panda.
The book's being given to the panda
   (by the woman).
The woman's giving the book to the panda.
The woman's giving to the panda
   (and what she's giving is the book).
NA Ergato AE Ergato
Kanu kelina kitapor palinos.
Kanuto kitapo palinos (kelinak).
Kanu kelinar kitapo palinos.
Kanuto kelina palinos (kitapok).

Peterson conjectures a valency-reduced form for the duative-unitive that he calls the ambipassive.  Embracing the weirdness of the duative-unitive system, which fails to provide a way to distinguish between the two arguments of a transitive verb (a major function of the other two systems), he envisions that one of the duative arguments is eliminated, but the audience can't tell if the remaining argument is subject or object, so that starting with  "The woman is petting the panda",  you drop either the woman or the panda, and end up with either a sentence that means  "The woman is petting or being petted,"  or a sentence that means  "The panda is petting or being petted."  Adding back in the dropped argument as an oblique, you get either  "The woman is petting, or being petted by, the panda,"  or  "The panda is petting, or being petted by, the woman."  The audience has to decide from context what is meant.

Valent-revalent Ergato

Here's the backstory.  Some time ago, an ancestral form of our language Ergato had completely eradicated all its noun case markings (even the recipient/dative and oblique), depending strictly on word order to determine the roles played by the various arguments of a verb, but it did have an NA valency-reduction mechanism, i.e., a passive.  Thus,  Lamu kelina palino (The woman is petting the panda),  Lamuto palino (The panda is being petted).  As the language evolved, the suffix  -to  kept its place in the sentence but changed its binding, so that it became a prefix on the noun:  Lamu to-palino.  The speakers of the language came to think of the  to-  prefix as meaning  "Skip one argument of the verb, and maybe we'll add it back onto the end of the sentence as an oblique."  They also took to the habit (or were already in the habit) of simply ending a sentence if they didn't have any more arguments they wanted to fill in.  Table 4 illustrates the treatment of transitive verbs at that point.

Table 4: Diachronic noun prefix, for a transitive verb.
English Ergato
The woman is petting the panda.
The woman is petting.
The panda is being petted.
The panda is being petted by the woman.
Lamu kelina palino.
Lamu kelina.
Lamu to-palino.
Lamu to-palino kelina.

To bring this noun affix more in line with the rest of the morphology, let's suppose that the noun prefix  to-  then became a noun suffix and lost the trailing vowel, becoming  -t;  then we've got Table 5.

Table 5: VR, for a transitive verb.
English Ergato
The woman is petting the panda.
The woman is petting.
The panda is being petted.
The panda is being petted by the woman.
Lamu kelina palino.
Lamu kelina.
Lamu palinot.
Lamu palinot kelina.

The  -t  is acting as a noun case marker.  We call the default case valent because it respects the valency of the verb, and the marked case revalent, because it rearranges the order of the valency bonds of the verb.  The case system as a whole is valent-revalent (VR).

At this point, the Ergane (the speakers of the language) might start using either a valent or revalent argument with an intransitive verb depending on whether or not the verb entails volition by the argument (an active verb) or not (a stative verb).  Setting a sentence with a stative verb alongside a sentence with a transitive verb (with valent arguments), one would then have Peterson's duative-unitive pattern of cases:  Sapu kelinatLamu kelina palino.

We can now form a number of permuted sentences with a ditransitive verb, keeping in mind that if the second argument of a ditransitive verb is marked revalent, that means that whatever argument was going to occur there has been skipped over.  Table 6 illustrates these permuted sentences.

Table 6: Minimal VR, for a ditransitive verb.
English Ergato
The woman is giving the book to the panda.
The woman is giving to the panda the book.
The book is being given to the panda by the woman.
The book is being given by the woman to the panda.
Kanu kelina kitapo palino.
Kanu kelina palinot kitapo.
Kanu kitapot palino kelina.
Kanu kitapot kelinat palino.
The woman is giving the book.
The woman is giving to the panda.
The book is being given to the panda.
The book is being given by the woman.
Kanu kelina kitapo.
Kanu kelina palinot.
Kanu kitapot palino.
Kanu kitapot kelinat.
The woman is giving.
The book is being given.
Kanu kelina.
Kanu kitapot.

This table also provides a second instance of the duative-unitive pattern, with the cases reversed:  Sapu kelinaKanu kitapot kelinat.  Moreover, this instance (unlike the earlier one) occurs in the pristine VR system, without supposing a split active-stative treatment of intransitive verbs.

What about skipping the first two out of three arguments, so that the third argument comes first?  In the prefix-marking form, we might just have used the prefix twice, indicating two skips over arguments:  Kanu to-to-palino (The panda is being given to).  At the next stage, that might become  Kanu palinototo.  What would happen when noun suffix  -to  becomes  -t,  though, isn't clear to me.  -toto  might mutate first into  -tot  and from there into  -to,  only stopping then because it has to remain distinct from the single-skip suffix  -t.  Or perhaps at some point in the process it would move sideways, and end up as  -k,  or  -s,  or something.  I'll arbitrarily use  -s  (a helpful choice for my English-trained ears).  Call it the double-revalent case.  If the first two arguments are skipped, and then one or both are tacked onto the end obliquely, presumably they'll be tacked back on with the same relative order they would have had in the first place (as if the first was skipped and tacked onto the end, and then the second was skipped and tacked onto the end).  That is,  Kanu palinos kelina kitapo  should mean  "The panda is being given by the woman the book,"  rather than  "The panda is being given the woman by the book."  Allowing the double-revalent case gives us Table 7.

Table 7: Full VR, for a ditransitive verb.
English Ergato
The woman is giving the book to the panda.
The woman is giving to the panda the book.
The book is being given to the panda by the woman.
The book is being given by the woman to the panda.
The panda is being given by the woman the book.
The panda is being given the book by the woman.
Kanu kelina kitapo palino.
Kanu kelina palinot kitapo.
Kanu kitapot palino kelina.
Kanu kitapot kelinat palino.
Kanu palinos kelina kitapo.
Kanu palinos kitapot kelina.
The woman is giving the book.
The woman is giving to the panda.
The book is being given to the panda.
The book is being given by the woman.
The panda is being given to by the woman.
The panda is being given the book.
Kanu kelina kitapo.
Kanu kelina palinot.
Kanu kitapot palino.
Kanu kitapot kelinat.
Kanu palinos kelina.
Kanu palinos kitapot.
The woman is giving.
The book is being given.
The panda is being given to.
Kanu kelina.
Kanu kitapot.
Kanu palinos.

It isn't necessary to suppose that native speakers of VR Ergato would take all these sentences in stride, only that they would understand them all, and grudgingly admit that all are technically correct.  Presumably they would find the more vigorously permuted argument orders rather stilted.  They might be no happier with  Kanu palinos kelina kitapo  than English speakers with  The panda is being given by the woman the book;  and  Kanu palinos kitapot kelina  should be worse.

Noun case is a marked property of a noun that determines its role in the surrounding phrase or sentence.  Under NA and AE, case is mildly context-sensitive, in that knowing the case marking of an argument underdetermines its relation to the verb unless one also knows whether the verb is valency-reduced.  Under VR, case is aggressively context-sensitive, in that one has to know the case marking of the argument, the natural valency of the verb, and the case markings of all preceding arguments.  Missing any of those other data, one might have no idea what role is played by an argument of known valent/revalent marking; the above table has valent nouns in every role (giving, given, and given to), and revalent nouns in every role.  Note, though, that the valent/revalent marking of the argument in question is just as essential:  missing the marking of the one argument, even given its complete context, one might again have no idea what role the argument plays.  For example, sentence  Kanu palino(-?)  could mean, depending on the marking  (-?),  that the panda is giving, given, or given to.  So valent/revalent markings are case markings.

IMHO, verb-initial sentence order plays an essential role in the feasibility of VR.  Revalency is essentially a "rotate left" operation on the remaining arguments of the verb, which could be somewhat taxing for the listener to track; presumably those fluent in Ergato would be used to it, but still it seems rather unfair (as in, not as naturalistic as I'm trying to be) to perform this rotation, sending a skipped argument off to the far end of the sentence, without first telling the listener how far off that is, so they know when to expect it to come 'round again.

Relative clauses

Peterson suggests that valency-reduction systems exist because they allow a noun to be related to a verb without enumerating other arguments that aren't currently relevant.  In English we have, e.g.,  I saw the car that was hit,  where  the car  is the topic of the passive  was hit  — telling us that the car we're talking about is the object of the verb  hit,  without forcing us to specify in this sentence the subject of  hit.  Without that use of the passive we'd have to say something like  I saw the car such that another car hit it,  which is very wordy and has caused us to say more than we'd probably wanted to about the other argument of  hit.  (The hitting incident is probably already known about from context, or we wouldn't have wanted to say  the car that was hit  in the first place; there's no need to reiterate in this sentence that the car was hit by another car rather than by a truck or a paint gun.)

This English example uses a relative clause, which is something we can't yet do in Ergato:  a clause (verb together with its arguments) that modifies a noun by indicating that that noun is also some particular one of the arguments in the clause.  To support them in Ergato, suppose there's a dedicated relative pronoun (i.e., a pronoun for this use and no other),  le.  Syntactically, a relative clause in Ergato comes immediately after the noun it modifies; and the clause uses the usual verb-initial order, with the  le  pronoun positioned within the clause at the appropriate place for its role, appropriately marked for case.  Table 8 gives a few examples.

Table 8: Relative clauses in NA, AE.
English English
The woman is petting the panda
   that the man gave the fish to.
The woman is petting the panda
   that the man gave the fish to.
NA Ergato AE Ergato
Lamu kelina palinor kanuri hopoko tanakir les. Lamu kelinar palino kanuri hopokor tanaki les.
English English
The woman is petting the panda
   that the man gave to her.
The woman is petting the panda
   that the man gave to her.
NA Ergato AE Ergato
Lamu kelina palinor kanuri hopoko ler lis. Lamu kelinar palino kanuri hopokor le lis.
English English
The woman is petting the panda
   that was given to her.
The woman is petting the panda
   that gave to her.
NA Ergato AE Ergato
Lamu kelina palinor kanurito le lis. Lamu kelinar palino kanurito le lis.

It should be clear that this relative clause device works just as well for VR as for NA or AE.  In a sense it works better for VR than for the other two systems, because, thanks to VR's support for arbitrary argument orders, we can always put the relative pronoun  le  at the end of the clause, and by requiring that order (perhaps it started as preferred usage and gained momentum from there), we can always tell unambiguously where each relative clause ends with no need for punctuation within the sentence.  Table 9 illustrates.

Table 9: Relative clauses in VR.
English Ergato
The woman who gave the fish to the man is petting the panda.
The woman who gave the fish is petting the panda.
The woman who gave to the man is petting the panda.
The woman who gave is petting the panda.
The woman who was given is petting the panda.
The woman who was given to is petting the panda.
Lamu kelina kanuri tanakit hopoko le palino.
Lamu kelina kanuri tanakit let palino.
Lamu kelina kanuri hopokos le palino.
Lamu kelina kanuri le palino.
Lamu kelina kanuri let palino.
Lamu kelina kanuri les palino.

Note the particular utility of the verb-initial order here, which allows us to know immediately when a relative clause starts without having to mark it by some special introductory word.  In, say, SVO order, we would probably put the relative pronoun  le  first — leaving us without a good way to indicate the end of the clause, since VR likes to end sentences by just leaving off some of the trailing arguments.

Topic

In NA and AE languages, number marking on the verb agrees with the argument with default case.  Table 10 illustrates that; but since the VR system ordinarily gives default marking to both arguments of a transitive verb, case marking on the arguments can't be used to decide which argument the verb should agree with.  This suggests a weakening of verb-noun linkage in the VR system, and I therefore suppose that the verb won't inflect for number of any argument.

Table 10: Number agreement in NA, AE, VR.
English NA Ergato AE Ergato VR Ergato
The woman is petting the panda.
The woman is petting the pandas.
The women are petting the panda.
Lamu kelina palinor.
Lamu kelina palinorne.
Lamune kelinane palinor.
Lamu kelinar palino.
Lamune kelinar palinone.
Lamu kelinarne palino.
Lamu kelina palino.
Lamu kelina palinone.
Lamu kelinane palino.

The verb-noun agreement under NA and AE is an expression of sentence topic, that is, which noun the sentence is primarily about.  The identity of the topic can become critical when a sentence says more than one thing about the topic, as in Table 11.

Table 11: Topic sharing in NA, AE.
English English
The woman is petting the panda
   and [the woman] is sleeping.
The woman is petting the panda
   and [the panda] is sleeping.
NA Ergato AE Ergato
Lamu kelina palinor i sapu. Lamu kelinar palino i sapu.

Table 12 shows how this topic-sharing interacts with valency reduction.

Table 12: Topic sharing and valency-reduction in NA, AE.
English English
The woman is petting the panda
   and [the woman] is giving the fish to the man.
The woman is petting the panda
   and the fish is giving [the panda] to the man.
NA Ergato AE Ergato
Lamu kelina palinor i kanu tanakir hopokos. Lamu kelinar palino i kanu tanakir hopokos.
English English
The woman is petting the panda
   and [the woman] is being given to the man by the fish.
The woman is petting the panda
   and [the panda] is giving to the man the fish.
NA Ergato AE Ergato
Lamu kelina palinor i kanuto hopokos tanakik. Lamu kelinar palino i kanuto hopokos tanakik.
English English
The panda is being petted by the woman
   and [the panda] is giving the fish to the man.
The woman is petting the panda
   and the fish is giving [the woman] to the man.
NA Ergato AE Ergato
Lamuto palino kelinak i kanu tanakir hopokos. Lamuto kelina palinok i kanu tanakir hopokos.
English English
The panda is being petted by the woman
   and [the panda] is being given to the man by the fish.
The woman is petting the panda
   and [the woman] is giving to the man the fish.
NA Ergato AE Ergato
Lamuto palino kelinak i kanuto hopokos tanakik. Lamuto kelina palinok i kanuto hopokos tanakik.

There are two reasons why VR can't do topic sharing.

The second reason [sic] is that, since revalency is marked on the arguments rather than on the verb, we can't tell the role of an argument without seeing its case marking.  Consider the attempted topic-sharing sentence  Sapu tanaki i kanu.  If we had used a relative clause, this would have been  Sapu tanaki kanu le(-?),  and we would know from the case ending  (-?)  whether the fish is giving, being given, or being given to; but since the whole point of topic sharing is to omit the topic of the second verb, we can't determine the shared topic's role in the second verb (except in some degenerate situations, such as when the second verb is intransitive).  That's a show-stopper:  there is no way we can parse the second half of the sentence without something, such as relative pronoun  le,  to exhibit the case marking of its topic.

So the show-stopping second reason is that we can't determine the shared topic's role in the second verb.  The more elusive first reason is that we can't tell which argument of the first verb is the topic.  Consider an attempted topic-sharing sentence where  kanu  is the first verb, which has all explicit arguments.  Kanu kelina palino hopoko i sapu.  Who is sleeping — the woman, the panda, or the man?  We could, of course, solve this problem by simply adopting some convention, such as that the first argument is always the topic, or, for that matter, that the last argument is always the topic (either of which would allow us to vary the topic by permuting the arguments).  That sort of convention, though, would surely have arisen only by accident (perhaps a holdover from before the language became revalent); it doesn't arise naturally from the VR system, and we don't have to steel ourselves to put up with it in our otherwise pristine VR language, because we already know that the topic-sharing mechanism will lose miserably on the second verb anyhow.  Remember that when we were looking at verb-topic number agreement (back in Table 10), the verb of the VR example didn't agree with either of its arguments.  Granted, that was my choice, made because at that point we didn't have any strong reason to prefer one or the other argument as topic — but here we are, still with no strong reason to prefer one or the other.  I suspect that neither argument is the topic:  that for classic transitive sentence  Lamu kelina palino,  the topic in an AE language is  palino,  the topic in an NA language is  kelina,  and the topic in a VR language is  lamu.  That would certainly cast an interesting light on the verb-initial word order.

Other forms of revalency

Thus far, I've been trying to illustrate the VR system with as few other distracting language features as possible, the better to focus on the essence of VR.  Also, part of my original goal was to build a case system that could exhibit a duative-unitive pattern, so my revalency indicators had to be case markings on nouns.  Moreover, putting these two factors together, I made valent, revalent, and double-revalent the only case markings on nouns.  There are however variations of the language feature that still seem to qualify as 'revalency'.  In this section, I'll suggest a few of these.

Here are three features that seem to me to be essential to the notion of revalency:  (1) revalency indicators indicate all deviations from, and only deviations from, an expected argument order;  (2) all possible argument orders are supported; and  (3) the revalency indicators occur in argument positions.  If a language supports arbitrary argument orders merely by marking all the arguments for semantic role, it fails (1).  If a language explicitly indicates alternative argument orders merely by marking the verb (akin to NA passive and AE antipassive), it fails (3).  If a language supports alternative argument orders merely by marking an argument to indicate passive (or antipassive), it fails (2).

A particularly simple variation of the VR system already described is to freeze the evolution of the ancestral language just after suffix  -to  detaches from the verb, and just before it attaches to the following noun.  That would give us a revalency clitic,  to,  and thus a more isolating language. Table 13 illustrates. 

Table 13: Revalency clitic.
English Ergato
The woman is giving.
The woman is being given.
The woman is being given to.
The woman is giving to the panda.
The book is being given to the panda by the woman.
Kanu kelina.
Kanu to kelina.
Kanu to to kelina.
Kanu kelina to palino.
Kanu to kitapo palino kelina.

Note that clitic revalency is clearly not a case system.  Moreover, since it doesn't mark nouns, it can readily coexist with a case system (there being no requirement that revalency be the sole determiner of argument order).  Supposing, for example, that the ancestral language had endings  -m  for nominative,  -r  for accusative,  -s  for dative, one would have Table 14. 

Table 14: Revalency clitic, with redundant case markings.
English Ergato
The woman is giving.
The woman is being given.
The woman is being given to.
The woman is giving to the panda.
The book is being given to the panda by the woman.
Kanu kelinam.
Kanu to kelinar.
Kanu to to kelinas.
Kanu kelinam to palinos.
Kanu to kitapor palinos kelinam.

With all that redundant information around, this system seems ripe for evolution of irregularities.

Another straightforward alternative, with a different backstory, would be to start with a complete set of explicit case markings (i.e., none of the case markings is "—"), and drop the case markings from an argument just when it is the next one in the expected order.  This has about as simple a backstory as can be:  it evolves as a form of abbreviation.  The sentences formed look exactly like those for VR Ergato from Table 7 except that the marked arguments are marked by semantic role instead of valent/revalent/double-revalent.  Table 15 illustrates this form of revalency, assuming the same explicit NA case markings as before. 

Table 15: Revalency by semantic role marking.
English Ergato
The woman is giving.
The woman is being given.
The woman is being given to.
The woman is giving to the panda.
The book is being given to the panda by the woman.
Kanu kelina.
Kanu kelinar.
Kanu kelinas.
Kanu kelina palinos.
Kanu kitapor palino kelina.

Here, revalency is a feature of the case marking system.  Note that this system, in its pristine form, cannot exhibit a duative-unitive pattern, since a verb can only have two arguments with the same marking if that marking is "—", which the pristine system always uses for the argument of an intransitive verb.  (Cf. the remark immediately following Table 6.)

An interesting possible source of variations on revalency is the devising (and diachronic justification, to taste) of other kinds of argument permutations for revalency, instead of (or, more bizarrely, in addition to) the left-rotate operation of our passive-derived  to  clitic.

Conclusion

I've tried to present the valent-revalent type system, and the more general principle of revalency, as clearly as I could — and, at the same time, present them in their best possible light, bringing out the most plausible and/or useful aspects of the scenario.  Some elements of the material that seem, to me, to have worked out particularly well are (1) the smooth way the revalent interpretation of  to-,  or  to,  replaces its passivizing interpretation; (2) the capacity of the revalent system to support arbitrary permutations of arbitrary subsets of the arguments without changing their semantic roles, and the neat way that meshes with the relative-clause syntax; and (3) the pure symmetry of identifying a transitive verb as the topic of its sentence.

Anyway, that's my idea.  I've learned a lot in working it out (I hope that was learning rather than befuddlement), and I hope you've gotten something useful/thought-provoking/fun out of reading it.  I also welcome your feedback on it (as of this writing, I'm here) — practical or conceptual, corrections, objections, questions, suggestions.  If anyone more facile than I with conlang projects has used, or does use, something at all like revalency —or if anyone can point out a revalency-like device in a natlang— I'd be fascinated to hear.








endnote: Ergane

Since it no longer makes sense for the name of the language to be a play on  "ergative",  here's a half-baked alternative.  Somewhere in the backstory of the name  Ergato  is a productive rule for deriving nouns for the different roles in a verb by putting case endings on a stem as if the derived noun were to appear as the sole argument of the verb.  (Vaguely reminiscent of English derivation  nominate >> nominator / nominee ).  If the verb meant something like  'to speak',  the stem of the derived nouns was  erga,  and it all happened at an intermediate stage of the language where the revalent suffix hadn't yet lost its final vowel,  ergane  would be the valent plural (speakers),  ergato  the revalent singular (that which is spoken).  Admittedly, this doesn't explain why we have no prior attestation of a syllable of the form vowel-consonant, nor of any voiced plosive, nor of a noun with only two syllables (or perhaps all three are related to the way the verb is transformed to make the noun stem . . .).