In singular-plural-dual triples, personal pronouns show the patterns in (1)-(4), but not the ABA pattern in (5), as shown by Smith et al. (2019), on the basis of data from many dozens of languages from around the world.
(1) AAA, e.g. Mapudungun 2nd person eymi – eym-ün – eym-u
(2) ABB, e.g. Kayardild 3rd person niya – bi-lda – bi-rra
(3) ABC, e.g. Savosavo 2nd person no – me – pe
(4) AAB, e.g. Yagua 2nd person jiy – jiryéy – sááda
Smith et al. (2019) also find that in nominative-acc/erg-oblique triples, personal pronouns show the patterns in (6)-(9), but not the ABA pattern in (10).
(6) AAA, e.g. Lezgian 1SG zun – za – zaz
(7) ABB, e.g. Latin 1SG ego – me – mihi
(8) ABC, e.g. Khinalug 1SG zɨ – jä – asɨr
(9) AAB, e.g. Wardaman 3SG narnaj – narnaji – gunga
This looks like a significant new finding which should be explained. Suppletion is strongly related to absolute high frequency (as I noted in an earlier post discussing Moskal 2015), so one might take usage frequencies as a starting point for an explanation, but I can only think of a possible way of explaining the relative rarity of the ABC and AAB patterns. (By far the most frequent patterns are AAA and ABB.)
The dual is much less frequent than the plural (which is in turn less frequent than the singular), and the oblique case forms are much less frequent than the accusative/ergative (which is in turn less frequent than the nominative). Suppletion needs high frequency, which disfavours the ABC pattern. And when there is analogical reorganization, it is usually a rarer form that is eliminated and takes on the stem of a more frequent form (thus, English kneel/knelt is leveled to kneel/kneeled, not to knell/knelled, because the present-tense form is more frequent). Thus, ABC will be regularized to ABB in preference to BBC (=AAB).
However, I cannot think of a good explanation of why *ABA should be even rarer than AAB and ABC. Is the dual somehow semantically closer to the plural than to the singular? Is an oblique form semantically more similar to an accusative or ergative than the nominative? These don’t seem plausible suggestions.
Thus, it is very interesting to see that Smith, Moskal, Xu, Kang, and Bobaljik (= Smith et al.) give an explanation for the absence of *ABA that is based on “containment” plus priority insertion of more specific forms (as per the Elsewhere Principle), much as in Bobaljik’s (2012) explanation of the absence of *ABA in comparatives and superlatives (*good – bett-er – good-est). This is an intriguing suggestion, and in the absence of a better explanation, it seems that it should be adopted. (Unfortunately, I cannot see how this explanation can possibly work, as I will explain below).
Smith et al. (roughly) propose the following underlying featural structures:
[[[pronoun] PL] DUAL]
[[[pronoun] ACC/ERG] OBL]
The idea is that in patterns like Latin ego ‘I.NOM’, m-e ‘I-ACC‘, and m-ihi ‘I-OBL’, the exponent m- is specified as occurring in the context “__] ACC]”, while the exponent ego occurs in the elsewhere case. Since the accusative case is contained in the representation of the oblique, the oblique must also have the exponent m-. It could have a different exponent only if the language had a more highly specified vocabulary item (like Khinalug asɨr, which is specified as occurring in the context “__] ERG]OBL]).” Thus, the hypothesis about containment plus the Elsewhere logic would explain the absence of ABA patterns.
It seems to me that unfortunately, this explanation cannot work (not for case suppletion, not for number suppletion, and not for comparatuve suppletion either), because it makes four presuppositions that can hardly be right:
(1) Homonymy is not possible. If homonymy were possible, then an ABA pattern could be easily generated by having two homonymous exponents, one of which is the default exponent and one of which is highly specified (to occur only in oblique or dual contexts). Homonymy is generally possible in languages, so a theory that relies on its non-existence can hardly be correct.
(2) There are no doublets. The Elsewhere logic only works if the existence of a suitable form blocks the generation of an alternative form, e.g. if the existence of a form better blocks the generation of *good-er. Now this is generally the case with comparatives and superlatives, but there are also languages where a regular and a suppletive comparative exist side by side (e.g. French meilleur and plus bon ‘better’). The Elsewhere principle would predict that plus bon should be blocked by the more specific exponent meilleur.
(3) Synthetic and analytic formations are completely different. One might want to say (like Bobaljik 2012 does) that plus bon is not blocked because it is analytic. But synthetic and analytic patterns obey basically the same generalizations elsewhere in grammar, as is (ironically) often pointed out by linguists working in the framework of Distributed Morphology. In Turkish, for example, the “postposition” ile and the instrumental “case suffix” -le/-la work in much the same way, with only a slight phonological difference. There is no good reason for saying that plus in plus bon is not a prefix in French.
(4) The effects of UG are not probabilistic. If one thinks probabilistically (as Greenbergian typology has done since the 1960s), then there is something odd about the idea that UG effects must be completely categorical. It is a perfectly coherent view that UG is not an absolute limitation on what is possible in languages, but merely guides grammars in certain directions, as envisaged by Jackendoff (2002: Chapter 4; “a set of attractors toward which languages tend statistically”), and a priori this is of course much more likely. It so happens that current generative grammar is committed to a discrete view of UG, but this leads to all kinds of artifacts, e.g. the distracting emphasis on explaining away a few problematic cases. It also makes it impossible to distinguish between rare patterns and frequent patterns (as discussed above), and thus to see further patterns that may be relevant to understanding what is going on.
In the last paragraph of the paper, the authors address an issue that I had raised earlier with them: How do we know that speakers actually analyze these forms, rather than simply storing them all, so that they are all suppletive (i.e. that all patterns are basically ABC)? The honest answer is: We don’t know it, except that if speakers did not do what Smith et al. claim they do, we would have no explanation of the cross-linguistic patterns. Thus, what we have here is a strong UG claim that is exclusively based on typological patterns. (Moreover, for the claim that case features are arranged hierarchically so that the containment explanation can work, again the motivation seems to come exclusively from typological patterns.)
But of course, there are different kinds of explanations for typological patterns, as was emphasized repeatedly and prominently by Stephen R. Anderson (discussed on this blog here and here). And I think that it should be clear that non-UG-based explanations (of the functional or diachronic sort) should take precedence over UG-based explanations, which are inherently less likely.
In the case of the missing ABA patterns, I do not know of an alternative explanation, as I noted above. Thus, this could somehow be hard-wired into the language faculty. But how this could happen is a real puzzle, because logically possible patterns like English-prime “good – better – goodest” or Latin-prime “ego – me – egohi” are readily learnable. Learnability cannot be the issue, so what is the problem?
I am still looking for an explanation.
Bobaljik, Jonathan David. 2012. Universals in comparative morphology: Suppletion, superlatives, and the structure of words. Cambridge, MA: MIT Press.
Jackendoff, Ray. 2002. Foundations of language: Brain, meaning, grammar, evolution. Oxford: Oxford University Press.
Moskal, Beata. 2015. Limits on allomorphy: A case study in nominal suppletion. Linguistic Inquiry 46(2). 363–376. doi:10.1162/LING_a_00185.
Smith, Peter W., Beata Moskal, Ting Xu, Jungmin Kang & Jonathan David Bobaljik. 2019. Case and number suppletion in pronouns. Natural Language & Linguistic Theory. doi:10.1007/s11049-018-9425-0. http://lingbuzz.auf.net/lingbuzz/003110
Something I find particularly interesting given that ABA patterns appear absent (or nearly absent given the few potential counterexamples given in the other comments) in pronouns with regards to number is that number marking on nouns can readily be found following an ABA pattern.
This occurs both with specific subsets of nouns where there arguably is a semantic basis for such a distinction, like the Kiowa class III nouns describing things that naturally come in pairs and which recieve a suffix when they do not regardless of whether there is one or more than two, and likely more relevantly for testing a number of theories also with nouns in general, for example in various places in Siberian Yupik, e.g. 3rd person nonsingular possessor “relative” (ergative-genitive) case which has SG: -(ng)ita DL: -gkena PL: -(ng)ita (with quite different morphophonological behaviour between the two as well) (there are many arguably less “awkward” cases of SG=PL≠DL in the paradigm, however this is as far as I can tell the only one that is suppletive rather than with a mostly transparent dual, and regardless of transparency the collapse seemingly only occurs in possessed forms).
I wonder whether it would be possible for a single hypothesis to account adequately for both the (near?-)absence of number-related ABA patterns in pronouns and their (in my admittedly limited and nonrepresentative sample of languages I have some familiarity with) seemingly greater presence among regular nouns.
So looking back at this comment which was made late at night in not exactly the most clearminded state it looks like I fell into the same trap as Y down below, reading ABA as sg-dl-pl rather than sg-pl-dl, which more or less voids my entire comment.
A slight delayed addendum here, but I have recently found a case of the “right” kind of ABA suppletion within regular nouns, with the singular and dual sharing a stem but the plural being different. In Yimas, accoring to W. A. Foley’s (1991) The Yimas Language of New Guinea, there is a significant number of nouns, primarily concentrated in certain word-classes, which have quite irregular number marking, where the dual is generally formed from the singular (though there are different suffixes for the different noun classes and some morphophonological rules in play as well as some occasional irregularities as well), with the plural being significantly more irregular, in a couple of cases outright suppletive (given here in the order SG-PL-DU):
narman – ŋaykum – narmprum “woman”
panmal – payum – panmalɲcrm “man”
tnum – tpwi – tnumul “(canonical) sago palm”
He mentions that this sort of irregular plurals are also attested in other languages in the area, but there’s no mention of whether they follow the same system of the dual being derived from the singular in such cases.
Two people commented on Facebook (https://www.facebook.com/martin.haspelmath/posts/10216359067141673), as follows:
“However, I cannot think of a good explanation of why *ABA should be even rarer than AAB and ABC. Is the dual somehow semantically closer to the plural than to the singular? Is an oblique form semantically more similar to an accusative or ergative than the nominative? These don’t seem plausible suggestions.”
But I find these suggestions to be very plausible indeed. Restricting attention to number, I can think of several respects in which dual patterns with plural to the exclusion of singular, suggesting that they are indeed closer. A random sample …
1. Many languages have a quantifier that means one, and another quantifier that means two or more; no languages have a quantifier that means one or two, and a quantifier that means three or more. (By quantifier here I mean a monomorphemic form that can occur as a free form, e.g. in response to a question such as ‘How many houses does he own?’
2. Dual and plural can function as distributive keys (aka antecedents to distriutive share expressions), for example, in an “English plus” with a dual”
*He / TheyDU / TheyPL carried three suitcases each
3. Similarly, dual and plural can function as antecedents for reciprocals, while singulars can’t:
*He / TheyDU / TheyPL hate each other
4. In languages with dual nouns, but whose targets of agreement lack a dual, the dual controls plural agreement, not singular, for example, marginally, in Hebrew
The weekSG passedSG quickly
The weekDU passedPL quickly
The weekPL passedPL quickly
Why not the historical explanation, “ex origine”? 🙂 It seems to account very well for these facts. Superlative is just more often derived from the comparative than from the positive (if it exists at all); dual may be derived from plurals because the word for “two” would require a plural form on the noun; then, dual forms often develop into plurals (if they don’t disappear at all) and oblique cases often derive from case stacking where the internal case is a structural case (genitive or accusative).
Some Uto-Aztecan languages have singular, dual and plural suppletive stems for certain verbs. Haugen and Everdell (Language Dynamics & cChange 5, 227, 2015) document many of these. ABA is rare but attested, e.g. Goshiute Shoshone ‘arrive’ (p. 257) S pitɨ́, D pippitɨ́, P pitɨ́.
Clearly I misread ABA as referring to Sg-Du-Pl rather than Sg-Pl-Du. But that is attested in the same source, too, e.g Hopi ‘kill’, Sg. and Du. niina, Pl. qöya; Tümpisa Shoshone ‘kill’, Sg. and Du. pakkah, Pl. wasü”, and about six others.
Another paper addressing exactly that came out just recently: John E. McLaughlin, Expanding to the edges; Central Numic dual number. IJAL 84(3), 359, 2018. McLaughlin clearly presents the different patterns of suppletion (all possible ones are present), and discusses various paths toward the grammaticalized dual in these languages.
Martin, just a brief comment, this time ☺. The explanation of the impossibility of ABA proposed in terms of ‘containment’ (which seems to me the best one available, especially in the area that I know best, which is case syncretism exhaustively analyzed by Caha 2009) is not, as you suggest, a universal property that is “hard-wired into the language faculty”. Both suppletion and syncretism are phenomena produced by linguistic changes and, therefore, UG does not pronounce itself in this regard. What these approaches propose is that the linguistic changes that give rise to syncretisms are sensitive to a universal hierarchy of traits (which, in this case yes, are supposedly ‘hard-wired’ in human cognition). These models are based on a specific and independent hypothesis of ‘containment’ of features/categories in a syntactic hierarchy that produces a contiguity (nom-acc-dat, etc.) that would explain the impossibility (except phonological accident) of the ABA pattern.
In addition, you know, in science the best available explanation is ‘the explanation’.
Yes, I always adopt the best available explanation – if necessary, a UG explanation. But it must be an explanation that could work, and the containment-based explanation simply does not work (I think). There might be a simpler UG explanation though, which posits a direct innate dispreference against *ABA, without all the other highly questionable presuppositions (feature hierarchy, impossibility of homonymy, etc.). This one also seems very unlikely, but it’s the best we have – I agree.
The generalization is not quite true, at least with respect to number:
Sumi Naga (Tibeto-Burman, Nagaland, India) nò / ò-kùʒó / no-ŋù
Idu (Tibeto-Burman Arunachal Pradesh, India) ȵu / e-ne / ȵu-ka-ȵu
In both cases the plural is formed from the independent singular, while the dual morpheme is inflected with the possessive pronominal clitic.
Thomas Graf has an upcoming paper on this topic, making an effective theory argument from monotonicity:
Thanks, I hadn’t been aware of that paper!