Almost as well-known as Jespersen’s Cycle is “Givón’s Cycle”, the change from independent personal pronoun to grammatical person marker (= person index) to zero, with subsequent renewal:
Givon’s Cycle (so dubbed by Seržant 2021): independent personal pronoun > subject index > zero > [renewal…]
Givón’s (1976) article is one of the most famous early papers of the modern grammaticalization literature, and the change also figures prominently in van Gelderen’s (2011) book on cyclic changes. The development of new subject indexes from independent pronouns is well-known from the Romance languages (e.g. Poletto & Tortora 2016) and the Germanic languages (see Fuß 2005, who also discusses other languages). And the loss of subject indexes is well-known from English, Swedish, French, as well as many creole languages. So it may appear that Givón’s Cycle is a common type of change.
But in recent cross-linguistic work by Ilja Seržant and associates (Seržant 2021; Seržant & Moroz 2022), we see a rather different picture, and it appears that the widespread impression that Givón’s Cycle is a common type of change is a misconception perhaps based on a Eurocentric perspective. Seržant (2021) examines the person(-number) indexes (for present-tense verbs, intransitive subject function) of 310 ten languages from 16 families whose proto-languages have been reconstructed. For each language, he counts the segmental length of the indexes and measures the length difference between the proto-language and the modern language, as illustrated in Table 7.
He measures both shape deviations (where indexes have different shapes in the modern language) and loss deviations (“deviations leading to demise”, i.e. changes leading to shortening, zero, or syncretism). While there are substantial shape deviations in the dataset, the amount of loss is not particularly high. In 11 out of his 16 families, the “degree of demise” index is below 0.1, and only 8% of the languages show a degree higher than 0.3. The amount of loss (or “demise”) found in English or Swedish in comparison with Proto-Indo-European is very high in comparative terms, and not at all typical of the world’s languages. Overall, Seržant concludes that “Cyclic changes in verbal person-number indexes are unlikely” (which is the title of the paper).
With his global perspective, he identifies three or four areas where there does seem to be a trend toward loss of subject indexes, as seen in his Figure 5.
Clearly, one area is western Europe, whose languages have had an outsize influence on linguists’ thinking. Seržant points out that these unusual changes may well be due to special historical circumstances, as argued by authors such as McWhorter (2016) (in general) and Donohue & Denham (2020) (specifically for Austronesian languages). Seržant’s database is not fully representative of the world’s languages and does not include many smaller families (and he cannot say anything about isolates because of his method that relies on reconstructed proto-languages). So there are some limitations that must be kept in mind, but I think there is no doubt that this paper represents a big step forward in our understanding of the dynamics of subject person indexes.
Another important observation of the 2021 paper is that subject person indexes may have sources different from independent personal pronouns: They may derive from auxiliaries as in Polish (this case is well-known also from Hopper & Traugott 2003: 145-148), from possessor indexes as in Cariban languages, from number markers (and other nominal markers, e.g. the participial suffix giving rise to 3rd person -b in Estonian), from directional deictic markers (also in some varieties of English, e.g. give it here ‘give it to me’), and by accidental reanalysis from various other sources. Thus, the properties of subject indexes cannot be entirely explained from the properties of their sources (as a source-oriented approach would expected), but we rather observe multi-convergence from different sources toward similar outcomes. (That subject person indexes are extremely widespread in the world’s languages had already been shown by Siewierska’s (2005) WALS map.)
In a 2022 follow-up paper, Seržant and his coauthor George Moroz identify a universal attractor for subject person-number indexes: On the basis of a study of (intransitive) subject indexes in 383 languages from 53 families, they find that the plural indexes show a strong tendency to be longer than the singular indexes, and the 3rd person indexes tend to be shorter than 1st and 2nd person indexes (in both numbers):
Now by comparing the length of the indexes to thee (reconstructed) proto-language forms, they find that changes are not necessarily reductions, as one might expect – they can be shortenings or lengthenings, but in any event, they tend to be in the direction of the attractor. (For example, Polish lengthened its 1st person plural suffix -m to -my, as in ma-my ‘we have’, perhaps on the analogy of the independent personal pronoun my ‘we’.) Thus, a mutational explanation (Haspelmath 2019) whereby synchronic states can be explained via tendencies of change does not seem to be appropriate – on the contrary, a diverse range of changes give rise to a uniform outcome, strongly suggesting the existence of a pull force (or attractor) driving the changes. Seržant attribute this to Zipfian efficiency of coding: “articulatory efficiency plays an important role here: the more expected the sign is, the shorter it is.” This study is thus an important confirmation of the predictions made by the efficiency theory of asymmetric coding in grammar.
Donohue, Mark & Denham, Tim. 2020. Mechanisms of language dispersal across southern Island Southeast Asia and the collapse of Austronesian morphosyntax. In Schapper, Antoinette & Gil, David (eds.), Austronesian undressed: How and why languages become isolating, 447–482. Amsterdam: Benjamins.
Fuß, Eric. 2005. The rise of agreement: A formal approach to the syntax and grammaticalization of verbal inflection (Linguistik Aktuell). Amsterdam: J. Benjamins.
Givón, Talmy. 1976. Topic, pronoun and grammatical agreement. In Li, Charles N. (ed.), Subject and topic. New York: Academic Press.
Haspelmath, Martin. 2019. Can cross-linguistic regularities be explained by constraints on change? In Schmidtke-Bode, Karsten & Levshina, Natalia & Michaelis, Susanne Maria & Seržant, Ilja A. (eds.), Competing explanations in linguistic typology, 1–23. Berlin: Language Science Press. (http://langsci-press.org/catalog/book/220)
Hopper, Paul J. & Traugott, Elizabeth Closs. 2003. Grammaticalization (2nd edition). Cambridge: Cambridge University Press.
McWhorter, John H. 2016. Is radical analyticity normal? Implications of Niger-Congo and Southeast Asia for typology and dichronic theory. In van Gelderen, Elly (ed.), Cyclical change continued, 49–92. Amsterdam: Benjamins.
Poletto, Cecilia, and Christina Tortora. 2016. Subject clitics: syntax, in A. Ledgeway and M. Maiden (eds.), The Oxford guide to the Romance languages. Oxford: Oxford University Press.
Seržant, Ilja A. 2021. Cyclic changes in verbal person-number indexes are unlikely. Folia Linguistica 55(s42–s1). 49–86. (doi:10.1515/flin-2021-2014)
Seržant, Ilja A. & Moroz, George. 2022. Universal attractors in language evolution provide evidence for the kinds of efficiency pressures involved. Humanities and Social Sciences Communications. Palgrave 9(1). 1–9. (doi:10.1057/s41599-022-01072-0)
Siewierska, Anna. 2005. Verbal person marking. In Haspelmath, Martin & Dryer, Matthew S. & Gil, David & Comrie, Bernard (eds.), The world atlas of language structures. Oxford: Oxford University Press. ((http://wals.info/feature/102A))
van Gelderen, Elly. 2011. The linguistic cycle: Language change and the language faculty. New York: Oxford University Press.
To Eugen Hill (I was not able to reply directly to your latest comment, possibly because of too much nesting?)
By phonological extensions, I mean any case where the number of segments in words is (regularly) increased from their original number. For example, epenthesis (commonly used to break up clusters, or added after word final consonants to remove closed syllables). Another example of phonological extension is vowel breaking (when a single vowel changes into two vowels.) During the Great Vowel Shift in English, /u/ –> /au/, /i/ –> /aj/.
As I am not aware of any study which attempts to quantify how common different sorts of phonological changes are, and in particular whether phonological reductions are more, less, or equally common as phonological expansions, I believe it is reasonable to assume that the two are equally common as the null hypothesis. (If such a study is in fact available, and shows that the two types of changes are not equally common, that would be really interesting to know!)
As far as the following:
“The standard theory says, reductions in length are mostly due to sound change, whereas gain in length is most usually due to univerbation. ”
I have not heard of this standard theory, and would be curious to hear about any sources that discuss/support this theory.
Are we talking about the same? Isn’t the whole discussion about the length of personal indexes on verbs (not about sound change in general)?
The study being discussed is about length of personal indexes. I am stating that it is reasonable to assume that sound change in general is not directional (extensions and reductions on average cancel out).
Therefore, if there are any patterns in the relative lengths of different affixes (such as singular tends to be shorter than plural, or third person tends to be shorter than other persons), it is not unreasonable to look for “attractors”, as the study does. We do not expect purely phonological factors to cause such patterns to show up. Thus, we do not need to “throw away” data points like Latvian, whose third person singular is shorter than the original PIE form and support the idea of attractors, merely because the “direct cause” of the change is a phonological law.
Re-inventing historical phonology from scratch (‘extensions and reductions on average cancel out’) is not promising. It’s etsablished that word-edges, especially the word-ends, are very often subject to special sound laws. It’s precisielly where personal index markers usually sit. The numerical relation between phonological reductions and phon. extensions in word-final position is roughly the following: in Indo-European family I could easily name severel dozen reductions (such as consonant cluster reduction in Sanskrit, etc.) but I know of just one case of phon. extension (lengthening of word-final vowels in Old Avestan). That at word-ends reductions in length are mostly due to phonology and extensions mostly due to univerbation is a well-known descriptive fact (not some theory-driven assumption which can be ingnored).
I have ceveral issues with the methodology applied by Ilja and his colleagues. Here are the two I think to be most important:
(a) First, they use reconstructed proto-languages. This means they don’t assume that the history of the languages they work with is irrelevant for establishing cross-linguistic diachronic trends. This is very remarkable but also raises the following question: why no attention is paid to the time span between the proto-language (such as Proto-Indo-European) and the individual languages used in the data-base. For instance, Latvian indeed shows 1sg.prs. -u ~ 3prs. -Ø. However, we know that this is because of a very recent sound change which eliminated all short vowels but u in the last syllable of a polysyllabic word everywhere in the language (i.e. also in nouns, etc.). The original system is preserved in the most closely related Lithuanian, where we find 1sg.prs. -u ~ 3prs. -a or -i. This means that we KNOW why the 1sg.prs in Latvian is longer than the 3prs. and this is not because of any cross-linguistically valid attractor state but rather due no a recent sound change specific to Latvian.
I would expect cases like this to be removed from the data-base. But they aren’t. Why? Doesn’t this necessarily mean that the calculations might be seriously distorted?
(b) The team uses reconstructions of inflected word-forms. The standard procedure of reconstucting inflection works roughly in the following way: one compares the form in question (such as the 1sg.prs of a particular inflectional class) in the many daughter-languages, observing the relevant sound laws (about which one knows from correspondence sets etsablished in the lexicon). If the different languages match, we simply substitute the relevant hypothetical proto-language sounds for those of the daughter-languages and enjoy our reconstruction. However, very often the languages won’t match. In such a case we suspect reshaping and/or restructuring in one or several of the daughter-languages or, finally, in all of them. Our hypothesizing about such restructurings is most usually deeply influenced by our theoretical expectations, i.e. on what kinds of change in inflection we believe to be typologically natural (or frequent, unremarkable, etc.). This means that the reconstructed inflectional forms one finds in the relevant literature are very often shaped by our typological beliefs, such as ‘the 3sg. should be originally zero’ and so on. Doesn’t that affect the conclusions reached in the papers?
Eugen, I don’t understand how your points would affect our study.
(1) Our idea was to see the general picture about what happens in languages between two stages: the proto-stage and the modern stage. Why would we have to exclude some changes (e.g. recent changes) and take other onboard? And why would the “recency” of a change matter? (Btw, the change in Baltic you are talking about is at least 1T y.o. and thus not as recent, as you know).
(2) Of course, there might still be some fine ammendments necessary in the reconstructed patterns. But you seem to make a stronger point that, e.g., the reconstruction of the thematic paradigm (the best attested one across all IE lgs) of PIE as we know it from the text books is not a good approximation of the (Late-)PIE thematic paradigm and something totally different has to be reconstructed because the current reconstruction is entirely subjective and driven by “theoretical expectations” of particular scholar? Precisely because a reconstruction is typically not just tracing back the sound changes but also reconstructing reanalyses and analogies, etc., it is more precise to rely on manual reconstructions that are there than on some automated reconstructions which technically cannot take into account reanalyses, analogies, etc., and would model only sound changes. And so we relied on reconstructions. But you don’t seem to believe in reconstructions…
My point is basically the follwoing:
we know that sound change (a) is morphologically blind, affecting all morphological structures and lexicon the same way, (b) may affect the length of personal indexes in verbs. It seems to follow that languages affected by established sound changes also in personal indexes shouldn’t be used in the study. It’s about eliminating factors potentiually interfering with the phenomenon under investigation. Imagine, a physicist interested in weak magnetic fields wouldn’t care whether his experiments were done in front of a refrigerator or not.
As for reconstructions: peronally I strongly believe that your PIE reconstruction of thematic present endings (3sg. *-eti, etc.) is correct. But this is not the point. The point is: reconstructions are not facts. Accordingly, in IE linguistics we have not one reconstruction of the relevant set of indexes but two competing reconstrcutions. As you know, people in Leiden and elsewhere would reconstruct not *-e-ti but *-e with zero-index and cite not Meier-Brügger’s handbook but that by Beekes. This position presupposes that 3sg.prs. of thematic verbs became longer at least in Celtic, Italic, Germanic, Anatolian and Indo-Iranian.
Finally, using reconstructions presupposes establishing which structure in the proto-language yields what in every particular daughter-language. As you perfectly know, this is often far from trivial. For instance, 3sg.prs. -e of Ukrainian thematic verbs is often assumed to reflect not PIE 3sg.prs. indicative *-e-ti (which would be difficult for phonological reasons) but 3sg.prs. injunctive *-e-t (which would be regular). Personally, I think this is wrong, but the point is: is it ok to simply count PIE *-e-ti > Ukr -e?
If the purpose of the study is to determine whether there are any overall trends in the length of various suffixes over long periods of time, it makes sense to include languages even when we know a particular change was due to purely phonological patterns.
Unless we have specific reason to think otherwise, it is reasonable to assume that phonological reductions are just as common as phonological lengthening.
Thus, if we study a lot of languages over time, on average phonological errors should “cancel each other out” and thus any trends in suffix lengths that remain may be reasonably assumed to reflect a true pull factor.
If we ignore length changes that are due to known phonological processes within a given language, we won’t necessarily know that the remaining data set will be truly representative of typical trends (for example, maybe it’s easier to find examples of reductions than increases, or vice versa). “Cleaning” the data in this way might then introduce a bias not present in the original data.
What is your reason to assume that ‘phonological reductions’ are as common as what you call ‘phonological lengthening’? Why do you think they would cancel each other out? ‘Reductions’ – which are very common – are often loss of segments. ‘Phonological legthenings’ – which are quite rare at word-ends – are changes in vowel length. The standard theory says, reductions in length are mostly due to sound change, whereas gain in length is most usually due to univerbation. That this is indeed the case has been repeatedly demonstrated for individual languages and whole groups of related languages. Why do you think we should ignore this fact and rather look for some different pull factor?
Fascinating post, as always!
It seems that Ilja Seržant has shown that loss of subject indexes isn’t as common worldwide as it is in Europe.
I’m curious though: of languages that have lost subject indexes, or never had them in the first place, are there any studies that look into how common various sources of subject indexes are?
In this post you mention that some possible source of subject indexes include independent pronouns, auxiliaries, possessor indexes, number markers (and other nominal markers), directional deictic markers, and various other sources. But which of these are most common?
Good question, but I’m not sure that we will be able to quantify this rigorously, as we cannot really count cases of change. But maybe there will be still more comprehensive follow-up studies.