Recently I attended a nice conference in Paris on grammaticalization, held in a 17th century mansion with a view over the Seine, and while there were pleasant distractions (such as a memorable visit to the famous café Les Deux Magots on the last day), the conference also helped me reflect a bit about some developments over the last few decades.
As the conference was organized by the “Société d’histoire et d’épistémologie des sciences du langage (SHESL)”, there was a certain focus on historical questions, and so in my talk, I made some reference to my earlier papers on agglutination (2009), synthetic-to-analytic changes (2018a), and polysynthesis (2018b), where I questioned some traditional stereotypes which ultimately go back to the Romantic period of the early 19th century. Although they have not been as prominent in linguists’ discussions in recent decades as they used to be in earlier times, these old terms are still very much with us and many readers of our works must think that they are well-established typological concepts. This is not true, unfortunately, and the very notion of the “word” that they are all based on is now known to be of doubtful cross-lingustic validity.
So what about grammaticalization? This term goes back to Meillet (1912) and became increasingly more popular since Lehmann’s (1982/2015) book “Thoughts on grammaticalization”. Do we now have a better understanding of the developments than linguists in the 19th century had? For quite some time, I was very optimistic. I remember hearing about the first conference on grammaticalization organized by T. Givón in Eugene (Oregon) in May 1988 (which resulted in two edited volumes, Traugott & Heine (eds) 1991ab), and at that time I was very excited about this new trend. I wrote a book review of the two volumes, and after my more specific papers about grammaticalization of passives and infinitives, I started writing general papers: on grammaticalization in morphology (1992), on its relation to generative functional categories (1994), and on the unidirectionality or irreversibility of grammaticalization (Haspelmath 1999). That was timely (and helpful for my career), because in the decade after 1999 grammaticalization reached the peak of its popularity, with four conferences (“New Reflections on Grammaticalization”, NRG) organized in western Europe during this time: NRG 1 in Potsdam (1999), NRG 2 in Amsterdam (2002), NRG 3 in Santiago de Compostela (2005), NRG 4 in Leuven (2008), and NRG 5 in Edinburgh (2012). After that, there were no follow-up conferences as far as I know, perhaps because the topic was felt to have been exhausted, and/or it was not so fashionable anymore. Or was this also because the new concept of constructionalization had come up (Traugott & Trousdale 2013) and linguists were unsure about the role of grammaticalization in this new context?
Some of these issues were discussed at the Paris conference, but no clear answer was proposed. So here I will focus on the reasons why I (largely) stopped working on grammaticalization after publishing half a dozen papers on it in the period between 1989 and 2004. One factor was clearly that my position at Leipzig’s MPI for Evolutionary Anthropology (after 1998) allowed me to consider global grammatical diversity in a depth that was not easy to do while working at a German university. And globally, it is hard to do historical linguistics, whereas global typology is perfectly possible and actually more meaningful than the more restricted (often Europe-focused) typology that I had done earlier. The World atlas of language structures (2005/2013) gives us a lot of food for thought, but it does not clearly relate to grammaticalization.
Around 2007, I started thinking more and more about the relation between cross-linguistic concepts and language-particular concepts (some of the key ideas of my 2010 paper on comparative concepts were presented in two 2007 talks at MPI-EVA, May and November), and I became more skeptical about widespread practice of assuming that cross-linguistic categories are present in individual languages, which also extended to core concepts such as “word” and “affix”, and “syntax” and “morphology” (Haspelmath 2011b). If these are not generally applicable concepts, then how can grammaticalization be a generally applicable concept? Of course, there are many striking similarities in the kinds of diachronic sources of grammatical markers across languages (Bybee et al. 1994 had shown this conclusively), and also in the kinds of change processes, but how exactly do we talk about these changes in such a way that our claims can be tested?
In my 2004 paper, I defined grammaticalization as follows:
“A grammaticalization is a diachronic change by which the parts of a constructional schema come to have stronger internal dependencies.” (Haspelmath 2004: 26)
Now this was extremely vague, and I did not specify further what I meant by “internal dependencies”. And without further assumptions, it does not really capture the core intuition that grammaticalization involves a change from a free form or lexeme to a grammatical marker. When I was asked a few years later to contribute to a grammaticalization handbook (edited by Narrog & Heine 2011), I said that I would contribute a chapter on coalescence (Jespersen’s 1922 term to replace the older “agglutination”), in which I emphasized that wordhood does not really have a clear cross-linguistic definition (Haspelmath 2011a). But I still had intuitive confidence in the claim that grammaticalization is a gradual process, and that there are synchronic gradients as a result of these gradual changes (Lehmann 1985 presented this idea in a very accessible and (to me) convincing way).
But how do we test the claim that there are synchronic gradients, or that diachronic changes are categorially gradual (that they are socially gradual goes without saying)? As I became more and more doubtful about the universality of synchronic categories, I also started to ask harder questions about diachronic changes, and I became less satisfied with the earlier answers. (The critical discussion of categorial gradience by Croft (2007) probably also contributed to this.) In general, we can distinguish between sharp boundaries and gradient transitions only when we have fine-grained measurements: If a light detector distinguishes only between “dark” and “light”, then there can be no intermediate period of dawn between night and day. We need more fine-grained light detectors to measure the gradual transition from the night’s darkness to daylight. So how do we determine that grammaticalization is indeed categorially gradual? To my knowledge, there are just two teams of linguists that have tried to measure degrees of grammaticalization: Bybee et al. (1991; 1994) and Bisang & Malchukov (2020) (Walter Bisang reported on some of these results in his invited talk at the Paris conference). As there are multiple parameters of grammaticalization, such measurements are difficult, and both these studies focused on cross-linguistic evidence for covariation of different parameters (especially form-related and meaning-related parameters). There was no focus on determining whether the changes are gradual within each language, and in fact it is not clear what units of measurement one would use to answer such questions.
So even though I still have the intuition that something like categorial gradualness may be what is going on, I do not want to presuppose it any longer. If one looks at the full range of phenomena, languages have a lot more morphosyntactic classes than any existing description tells us about (this was observed by Croft 2001: 36), and it seems difficult to fit all this variation into a single scale of “coalescence”. The intuition that some pairs of forms are “more tightly interconnected” than other pairs of forms is strong, but maybe tightness is just another stereotype. Again and again, we find that different criteria or processes do not correlate as it appears at fist sight, and that the presumed (or presupposed) correlations are illusory. As Croft (2001: 36) put it: “Mismatches in distributional patterns are the norm in languages”. (A recent concrete demonstration of this with respect to the supposed “word-affix continuum” is van Gijn & Zúñiga 2014: 155.)
Thus, I have increasingly proposed sharp definitions of commonly used terms in recent years, including affix and word. In this spirit, one could define grammaticalization as follows:
Grammaticalization is the change from a free form or lexeme to a grammatical marker.
The terms free form, lexeme(-stem) and grammatical marker all have sharp definitions: A free form is a form that can occur in isolation; a grammatical marker is a clitic or an affix (i.e. a bound morph that is not a root; and lexeme-stem is defined in Haspelmath (2022).
This definition is narrower than many other definitions, but it does capture most of the core cases of grammaticalization, and it also leads to fewer exceptions to the unidirectionality claim (several of the recognized exceptions to unidirectionality in Haspelmath (2004) are not counterdirectional in the sense of showing a change from grammatical marker to free form or lexeme).
In my 2004 definition, I wanted to include the notion of “construction”, because its importance in understanding grammaticalization had recently been emphasized by authors such as Joan Bybee and Elizabeth Traugott. I was in full agreement with this, but now, two decades later, my emphasis is more on making clearly testable claims. I am skeptical that it will ever be possible to rigorously test the claim that constructions have a key role in grammaticalization changes, because hypotheses about causal factors in language change are generally very hard to test. So I have become more modest in my goals and hopes, which also explains the increasing focus on methodological issues in my work. When taking a broad historical perspective, it is quite sobering to realize how little actual progress in understanding has been made over the last two centuries (since Humboldt’s brilliant 1822 paper on the diachronic origins of grammatical forms). This does not detract from the fascination that grammar and grammatical changes hold, but it is also clear that a 20-year period (c. 1990-2010) of intensive (and perhaps also fashion-driven) research on grammaticalization does not necessarily lead to breakthroughs of understanding, or to the irrelevance of earlier work.
Bisang, Walter & Malchukov, Andrej L. & Rieder, Iris & Sun, Linlin. 2020. Measuring grammaticalization: A questionnaire. In Bisang, Walter & Malchukov, Andrej (eds.), Grammaticalization scenarios, volume 1, 1109–1128. Berlin: De Gruyter Mouton.
Bybee, Joan L. & Pagliuca, William & Perkins, Revere D. 1991. Back to the future. In Traugott, Elizabeth Closs & Heine, Bernd (eds.), Approaches to Grammaticalization: Volume II. Types of grammatical markers, vol. 19:2, 17–58. Amsterdam: Benjamins.
Bybee, Joan L. & Perkins, Revere & Pagliuca, William. 1994. The evolution of grammar: Tense, aspect and modality in the languages of the world. Chicago: The University of Chicago Press.
Croft, William. 2001. Radical Construction Grammar: Syntactic theory in typological perspective. Oxford: Oxford University Press.
Croft, William A. 2007. Beyond Aristotle and gradience: A reply to Aarts. Studies in Language. International Journal sponsored by the Foundation “Foundations of Language.” John Benjamins 31(2). 409–430. (doi:10.1075/sl.31.2.05cro)
Haspelmath, Martin. 1999. Why is grammaticalization irreversible? Linguistics 37(6). 1043–1068.
Haspelmath, Martin. 2004. On directionality in language change with particular reference to grammaticalization. In Fischer, Olga & Norde, Muriel & Perridon, Harry (eds.), Up and down the cline: The nature of grammaticalization, 17–44. Amsterdam: Benjamins.
Haspelmath, Martin. 2011a. The gradual coalescence into “words” in grammaticalization. In Narrog, Heiko & Heine, Bernd (eds.), The Oxford handbook of grammaticalization, 342–355. Oxford: Oxford University Press.
Haspelmath, Martin. 2011b. The indeterminacy of word segmentation and the nature of morphology and syntax. Folia Linguistica 45(1). 31–80. (doi:10.1515/flin-2017-1005)
Haspelmath, Martin. 2021. Bound forms, welded forms, and affixes: Basic concepts for morphological comparison. Voprosy Jazykoznanija 2021(1). 7–28. (doi:10.31857/0373-658X.2021.1.7-28)
Jespersen, Otto. 1922. Language: Its nature, origin and development. London: Allen & Unwin.
Lehmann, Christian. 1985. Grammaticalization: Synchronic variation and diachronic change. Lingua e Stile 20. 303–318.
Lehmann, Christian. 2015. Thoughts on grammaticalization. Berlin: Language Science Press. (http://langsci-press.org/catalog/book/88)
Meillet, Antoine. 1912. L’évolution des formes grammaticales. Scientia 12(26). 6.
Narrog, Heiko & Heine, Bernd (eds.). 2011. The Oxford handbook of grammaticalization. Oxford: Oxford University Press.
Traugott, Elizabeth Closs & Heine, Bernd (eds.). 1991a. Approaches to grammaticalization (Volume I: Focus on theoretical and methodological issues). Amsterdam: Benjamins.
Traugott, Elizabeth Closs & Heine, Bernd (eds.). 1991b. Approaches to grammaticalization (Volume II: Focus on types of grammatical markers). Amsterdam: Benjamins.
Traugott, Elizabeth Closs & Trousdale, Graeme. 2013. Constructionalization and constructional changes. Oxford: Oxford University Press.
Thanks Martin, this has been an interesting read (and a trip down memory lane). I agree with most of what you say, and I would like to quote your definition in a handbook entry on gzn I am currently writing. Just to be sure: does “grammatical marker” only include affixes and clitics, i.e. not function words like prepositions? In other words, does it exclude what has traditionally been labelled primary gzn?
And an addendum to quantitative approaches to to gzn there is Correia Saavedra’s thesis “Measurements of grammaticalization” (De Gruyter 2021), which includes computational methods (collocation and colligation analysis) to semantics, as a way to quantify semantic substance.
Thanks, this is helpful! I just wanted to add that in addition to the 2 teams you mention, a Ph.D thesis by me, defended in December 2022 also measures degrees of grammaticalisation (by means of 3 variables) to explain the distinctions in the productivity of Estonian spatial cases when they function as argument markers. This is chapter 5 in here: http://dspace.ut.ee/handle/10062/88033
Many thanks for pointing this out – I’ll check out your dissertation.
Lovely to read your text – I was lucky to share some of that journey with you.
Your new definition “Grammaticalization is the change from a free form or lexeme to a grammatical marker” strikes me as very modest. Wasn’t that Lehmann’s starting point. and didn’t he capture what he perceived as the gradual nature of grammaticalization by distinguishing the various parameters? Don’t they mean that grammaticalization may *look* gradual, especially in a comparative perspective, since the changes implied by the various parameters need not overlap?
Yes, for a long time I was convinced that Lehmann’s approach was exactly right, and I thought that Bybee et al. had shown this by their quantitative measures. However, a substantial part of their quantifications was based on intuitive notions which I no longer think of as solid. For example, loss of “semantic substance” (just one part of one of Lehmann’s six parameters) requires a way of measuring the amount of semantic substance. And “bondedness” (another parameter) is itself composed of a multiplicity of more primitive elements which do not necessarily correlate with each other. Clearly, there is a lot of insight in Lehmann’s and Bybee’s work, but I don’t think that it’s as rigorous as it should eventually be.
You could potentially define something as “semantically substantial” if it refers to an thing, action, or property (the same categories that you use to define “roots”), and anything else as semantically unsubstantial. Then a “loss of semantic substance” would be any change in which something which is semantically substantial becomes semantically unsubstantial.
For example, body parts are things, thus semantically substantial. Directions are not things, actions, or properties, and thus semantically unsubstantial. The change of body part terms into directional terms/prepositions, which is a classical example of common grammaticalization pathways, would then count as a loss of semantic substance by this definition.
This definition doesn’t allow one to measure different “levels” of semantic substance (something is either semantically substantial, or semantically unsubstantial, with no middle ground), but it’s at least a start. You could check whether loss of substance is more common than gain of substance, for example.