We have seen the data on the spread of o-vocatives in Modern Greek. I will post how I make sense of the data. But first, some preliminaries about analogy.

How analogy works

Analogy in language change takes a linguistic rule that applies to one word or paradigm or category, and starts applying it to another word or paradigm or category. To take an example that’s rather removed from o-vocatives: analogy in non-rhotic dialects of English takes the rule “keep the word-final [ɹ] before a vowel, as in more or less [mɔː-ɹ ɔː lɛs]”, and starts applying it in novel contexts, like law and order [lɔ-ɹ ənd ɔːdə].

There was no historical reason for there to be an r after law; that’s creating a new rule where there is no historical justification for it (there is no r to keep). That’s what makes it a linguistic innovation, after all. But there is a linguistic justification for it: it is that more looks like law. When they’re not followed by a vowel, the two words end in the same vowel, and they have the same syllable structure: [mɔː, lɔː].

And that’s how analogy works in general. If a linguistic rule is generalised from A to B, it doesn’t happen at random. It’s because A and B have a feature in common. And linguistics is not powerless to make sense of analogical change: linguistics can identify what features A and B have in common, and use that to explain the change.

This is a diagnostic, not a predictive explanation: it’s the kind of explanation at home in historical linguistics, not synchronic linguistics, because we’re not dealing with a rule here, but a tendency, that could have happened, and could have not happened. But analogy is a bunch of contingent tendencies, and not rules; that’s why language undergoing analogical change is not particularly rule-bound. Or, to put it more informally, a mess.

As analogies spread, if an analogy spreads from A to B and from B to C, that spread can be explained because A, B, and C all have a feature in common. But again, analogy is not that rule-bound or predictable, and the analogy between A and B can be quite independent from the analogy between B and C. The feature B and C have in common may not be a feature A and B have in common; it may well not be present in A at all.

To give a sneak preview of how the o-vocative spread: I will argue that it spread from “old man” to “Nick” to “Venizelos”. The words “old man” and “Nick” [ˈɣeros, ˈnikos] have phonological features in common: they are bisyllabic, and penult-accented. The words “Nick” and “Venizelos” have one out of the two phonological features in common: they are penult-accented, but there is no longer a requirement for the word to be bisyllabic. Yet that’s not enough to explain the analogy: there are lots of penult-accented words that don’t have an o-vocative. The feature that “Nick” and “Venizelos” have in common is that they are names. And that’s a feature that “old man” does not have: as the o-vocative spread from class to class of words, the NAME feature is something it picked up when it started applying to given names like “Nick”. It was not a feature present at the start.

So there are features involved in the spread of a phenomenon by analogy. Those features are how the distribution of the analogy can be made sense of, while it is in progress. And when the analogical spread settles down, they will be the basis of the new rules that govern the phenomenon.

And if an analogical change involves more than one feature, not all features are necessarily equal. The more common or “basic” a particular feature is in a language, the more likely it is that new instances of analogy will be based on it. If an analogy is spreading among, say, names of ethnic groups, there are limits to how far the change is going to spread. If the analogy hops across from ethnonyms to, say, proper names in general, the analogical change has the potential to take over much more of the language.

Similarly (though this is not as obvious), a very common or psychologically salient word, like “old” or “mother”, is likelier to form the model for further analogies than a less common or salient word, like “vague” or “ogre”. I haven’t seen this particular notion articulated explicitly anywhere, though I suspect something to it will have been stated by Kuryłowicz or Dressler, who have theorised extensively on analogy.

What analogy does

Analogy is the mechanism that spreads language change, and makes it make sense. Analogy takes a local change in one instance of language, and spreads it further afield. Analogy spreads such changes in a way that, eventually, can be made sense of by a simple rule, that ends up replacing whatever other simple rule was there before—as the features that enabled the analogy are generalised.

Eventually. Because until the language change goes all the way through, analogy in action is a mess. It’s two rules, the old and the new, clashing, and the context in which one rule or the other applies is completely idiosyncratic: it varies from social group to social group, from semantic category to semantic category, from individual to individual. You’ll have seen a little of that in the disagreements among commenters I related, as to whether “uncle” or “Marinos” had an o-vocative, and what the connotations were if it did.

And that’s not an idiosyncrasy of Greek vocatives: that happens whenever there’s a language change underway. Moreover, while analogy usually runs to completion, and you have a clean new status quo, sometimes it remains messy. Hence the unpredictability of the pronunciation of <ea> in English, for example.

Analogy is the culprit for language learning being exasperating for adults, from a book: it’s where the attempts to formulate simple, learnable rules for language founder. That’s where the infuriating laundry lists of exceptions, and long lists of ifs and unlesses in textbooks come from. Language in preliterate societies, and child acquisition of language, deal with these exceptions and conditions. In fact, they are valuable in tidying them up for the next generation, through logical abduction: confronted with a whole lot of messy data that doesn’t follow nice rules, language learners make up their own rules, which tend to be simpler, and more learnable, and wrong: they overgeneralise from the data they get. Which ends up being a good thing for the language.

Which is fine if you’re two years old, or if you’ve married into a tribe, and have a couple of years on your hands. If you’re in a bit more of a hurry, grammars and textbooks give you shortcuts: they’ve worked out the rules for you in advance. Except where the rules are a mess, because that aspect of the language is in flux, and being pummelled back into non-mess over generations through drift of analogy.

So to any language learner who has groaned about why they have to learn laundry lists of genders, and why there’s so many exceptions, and why there are so many irregularities, I present to a cartoon villain for your dartboard. Commenter Kostas in the discussion of o-vocatives at Sarantakos’ blog:

I rejoice that the unruly people finds a way to escape the norms and shackles set by learned linguists, and in their own unique way they transform, develop, and enliven their language.

And I don’t think that’s by accident. As they learn their mother tongue, they create their own internal mechanisms and rules and they then apply them in its development. I call that language instinct, and I follow it in my life, in my spoken language. In my written language I conform more with the rules posed by linguists and other specialists.

Kostas is of course correct in the second paragraph, in how “language instinct” works: that’s logical abduction in the face of messy linguistic input. Linguists will be shocked to see themselves caricatured in the first paragraph: all descriptive linguists are trying to do is make sense of the rules that people do carry with them in their “internal mechanisms”, which is what the logical abduction is producing: they aren’t setting any shackles for anyone. But of course prescription does have a role to play in the social aspect of language, and linguistic descriptions are recruited for prescriptions. That’s not an evil; that’s just what happens.

Language learners can keep Kostas on their dartboard though. His convenience and unruliness is your inconvenience and rote-learning. And those language learners do the same in their own native language. (“English is a mongrel language, English has no Academy”, etc. etc.)

And that’s how speakers like Kostas, and everyone else who speaks Greek, comes up with rules for second declension vocatives that look like this:

• Most of the time use -e
• The word for “old man” uses -o
• “Nick” uses -o
• Words like “old man” and “Nick” use -o

Words like “old man” and “Nick”?! What kind of rule is that?

A very porous, ambiguous rule. That’s why speakers disagree so much. And that’s what happens when one aspect of a language is in flux.