In this post and in a few that are planned for the future, I will highlight some things that (I now think) were wrong in my earlier scientific writings. Science is supposed to be self-correcting, and these posts will demonstrate that this can happen also within a scholar’s career (not only when a new generation takes over, as is expressed by “Planck’s principle”: “Science progresses one funeral at a time”).
Of course, we all make many small mistakes, so I will focus on a few that seem bigger mistakes to me which are worth highlighting, also because I’m sometimes not the only one to have made them.
I will begin with the idea that basic-derived relations in morphology show the kinds of asymmetries that we observe because of an iconic principle. This was expressed very clearly in the intro section in my most widely cited paper: “More on the typology of inchoative-causative verb alternations” (Haspelmath 1993):
“The formally derived (or marked) words are generally also semantically derived in that they have some additional meaning element that is lacking in the formally basic (or unmarked) word. This correlation between the formal and the semantic basic-derived (or markedness) relationships has been identified as an instance of diagrammatic iconicity, an interesting case of external motivation for linguistic structures (e.g. Haiman 1980, Mayerthaler 1981).” (Haspelmath 1993: 87)
At the time, the idea of iconicity was quite popular, and I was particularly influenced by Bybee (1985) and Croft (1990), who in turn were influenced by Haiman (1980; 1983; 1985). Givón (1991: 106) expressed the idea as follows: “Categories that are cognitively marked –i.e. complex – tend to also to structurally marked” (he called it the “meta-iconic markedness principle”).
In the 1993 paper, the specific question I discussed was the directionality of asymmetric coding in causal-noncausal alternations such as those in (2) (at the time, I used “causative” for causal, and “inchoative” for noncausal): Even for the same verb meaning, some languages have an additional marker for the noncausal member of a verb pair (an anticausative marker), and others for the causal member (a causative marker).
But the paper found a general tendency for some verb meanings (such as ‘freeze’ and ‘dry’) to show causative coding of the verb pair, while others (such as ‘break’ and ‘close’) tend to show anticausative coding. Croft (1990: 50) had noted that “the more typically the change of state requires an external agent, the more likely the causative type will be unmarked”. This was supported by my findings, but how did it relate to the expected iconicity explanation? I had started out my paper with the observation (due to Mel’čuk 1967) that there was an apparent problem with anticausatives, because their coding seems to be counter-iconic: The causal member (e.g. Russian rasplavit’ ‘melt (tr.)’) is basic, while the noncausal member (which is semantically simpler) has an additional anticausative marker (e.g. Russian rasplavit’s -sja ‘melt (intr.)’).
So how did I solve this apparent problem in the 1993 paper? I appealed to a difference between “objective meaning” and “conceptual/cognitive meaning” (in this, I was influenced by Lakoff 1987), as follows:
“If the semantic properties of a word are only the objective semantic features discovered by semantic decomposition (as, e.g., in Mel’čuk 1967), then causatives are always semantically more complex than inchoatives and the existence of or even preference for anticausatives is a mystery. But iconicity in language is based on conceptual meaning, and the correlation between formal and semantic basic-derived (or markedness) relationships should be understood in cognitive terms.” (Haspelmath 1993: 106)
Unfortunately, there was no good basis for this distinction, and in the next paragraph I went on to talk about “prototypes” and other things that I did not really understand. The typological contribution of the paper was clear (and it was quickly picked up by Levin & Rappaport Hovav (1995) in their MIT Press book, paving the way for my paper’s subsequent fame), but the explanation was murky and implausible. This part of my 1993 paper was thus wrong.
It was only after I read a little-known thesis by Saundra K. Wright (supervised by Beth Levin in 2001) that I realized that the decisive factor might be frequency, not anything about the (“objective” or “cognitive”) meaning. I had met Saundra accidentally during a visit of California State University at Chico in 2003, where she told me that her thesis was also inspired by my 1993 paper. So this was a good instance of a wrong but interesting paper inspiring new research and helping to correct the earlier claims.
After having understood that the explanation in terms of “conceptual meaning” in my 1993 paper was wrong, I went on to criticize the concept of markedness (in 2006), and to argue for frequency-based explanations over iconicity-based explanations in morphology more generally (2008). But it was only in 2014 that we published a paper where we showed that the frequency skewing found for English by Wright was valid more generally (Haspelmath et al. 2014), and this was followed by an even more general paper where I argued that frequency-based predictability explains not only the tendencies for typical causative/anticausative pairs, but also various other tendencies (Haspelmath 2016).
Quite generally, we have been observing a shift away from the iconicity explanations of the 1980s and 1990s, toward efficiency-based explanations which take into account frequency of use and predictability (e.g. Gibson et al. 2019; Levshina 2022). So my mistake in 1993 was also due to the fashion of the time. Fashions exert a powerful influence on all people at all times (especially on younger people), but in science, they can be rather detrimental, and scientists must try to focus on evidence and arguments, and not to be unduly influenced by fashion. So is the new focus on efficiency that I noted yet another fashion? Maybe, but it is interesting to see that Jespersen (1941) expressed very much the same basic idea eighty years ago, at a time when it was not fashionable and when Jespersen was eighty years old:
“In linguistic changes we see the constant interplay of two opposite tendencies, one of an individual, and the other of a social character, one towards ease and the other toward distinctness. The former is the tendency to take things easy and to follow the line of least resistance–to say it bluntly, an outcome of human indolence or laziness… The opposite tendency is an effort to be clearly and precisely understood, and to make as vivid and convincing an impression on the hearer as possible; each articulation is therefore made slowly and distinctly, and great exertion is made to choose the most lucid and forcible expression… In extreme cases this may lead to pompousness and over-emphasis.” (Jespersen 1941: 15-17)
Bybee, Joan L. 1985. Morphology: A study of the relation between meaning and form. Amsterdam: Benjamins.
Croft, William. 1990. Possible verbs and the structure of events. In Tsohatzidis, Savas L. (ed.), Meanings and prototypes: Studies in linguistic categorization, 48–73. London: Routledge.
Gibson, Edward & Futrell, Richard & Piandadosi, Steven T. & Dautriche, Isabelle & Mahowald, Kyle & Bergen, Leon & Levy, Roger. 2019. How efficiency shapes human language. Trends in Cognitive Sciences 23(5). 389–407. (doi:10.1016/j.tics.2019.02.003)
Givón, T. 1991. Markedness in grammar: Distributional, communicative and cognitive correlates of syntactic structure. Studies in Language. International Journal sponsored by the Foundation “Foundations of Language” 15(2). 335–370. (doi:10.1075/sl.15.2.05giv)
Haiman, John. 1980. The iconicity of grammar: Isomorphism and motivation. Language 56(3). 515–540. (doi:10.2307/414448)
Haiman, John. 1983. Iconic and economic motivation. Language 59(4). 781–819. (doi:10.2307/413373)
Haiman, John. 1985. Natural syntax: Iconicity and erosion. Cambridge: Cambridge University Press.
Haspelmath, Martin. 1993. More on the typology of inchoative/causative verb alternations. In Comrie, Bernard & Polinsky, Maria (eds.), Causatives and transitivity, 87–120. Amsterdam: Benjamins.
Haspelmath, Martin. 2006. Against markedness (and what to replace it with). Journal of Linguistics 42(1). 25–70. (doi:10.1017/S0022226705003683)
Haspelmath, Martin. 2008. Frequency vs. iconicity in explaining grammatical asymmetries. Cognitive linguistics 19(1). 1–33.
Haspelmath, Martin. 2016. Universals of causative and anticausative verb formation and the spontaneity scale. Lingua Posnaniensis 58(2). 33–63.
Haspelmath, Martin & Calude, Andreea & Spagnol, Michael & Narrog, Heiko & Bamyacı, Eli̇f. 2014. Coding causal–noncausal verb alternations: A form–frequency correspondence explanation1. Journal of Linguistics 50(3). 587–625. (doi:10.1017/S0022226714000255)
Jespersen, Otto. 1941. Efficiency in linguistic change. Copenhagen: Munksgaard.
Lakoff, George. 1987. Women, fire, and dangerous things: What categories reveal about the mind. Chicago: The University of Chicago Press.
Levin, Beth & Rappaport Hovav, Malka. 1995. Unaccusativity: At the syntax-lexical semantics interface. Cambridge, MA: MIT Press.
Levshina, Natalia. 2022. Communicative efficiency: Language structure and use. Cambridge: Cambridge University Press.
Mayerthaler, Willi. 1981. Morphologische Natürlichkeit. Wiesbaden: Athenaion.
Mel’čuk, Igor. 1967. K ponjatiju slovoobrazovanija. Izvestija AN SSSR: Serija literatury i jazyka 26(4). 352–362.
Wright, Saundra K. 2001. Internally caused and externally caused change of state verbs. Northwestern University. (Ph.D. dissertation.)
Interestingly in the context in which you quote it, the passage above from Jespersen (1941) carries unmistakable echoes of a work from fifty years earlier, Georg von der Gabelentz’s (1891) Die Sprachwissenschaft. His discussion in book 3, section 2, §1 describes change as resulting from the interplay of two forces, Bequemlichkeit (“comfort/ease”) and Deutlichkeit (“clarity/distinctness”).
Just as in Jespersen, ease is related to a general principle of least effort: “so gilt für die Sprache der wirthschaftliche Grundsatz, dass der Zweck mit möglichst geringem Aufwande erreicht werden soll” (“the economic principle that the goal should be reached with as little effort as possible also holds for language”). Just as in Jespersen, distinctness is linked to articulatory clarity: “da sprechen wir mit besonderer Anstrengung der Sprachorgane … Es kann aber auch geschehen, dass sie übertreibt” (“then we speak with particular exertion of the speech organs … but it can also be exaggerated”).
We know that Jespersen read Gabelentz and built on his ideas, so the focus on efficiency is old indeed. It’s even tempting to speculate that, like language change, developments in linguistic theory might be cyclic… though that’s a depressing thought.
But was Gabelentz as influential during Jespersen’s lifetime as he is well-known nowadays? I’m not sure. Jespersen first expressed similar ideas in his 1894 book, which may have been influenced by Gabelentz, though this is still less likely. I kind of think it is more likely that these ideas were floating around in multiple places at the time, and Jespersen’t main innovation in 1894 was to take a more optimistic view of the results of the changes (compared with Schleicherian pessimism). But it’s hard to tell without knowing a lot more about the entire context of the 19th century.
“I think the answer is that they cannot be explained in general terms, because they do not represent a very general phenomenon”
Do we know this?
“General explanations should be invoked only if we have good reasons (and not just suspicions) to believe that the patterns we study are indeed cross-linguistically general.”
I always thought that “suspicions” that an idea might be true are the driving force behind hypothesis formation and the advancement of science — not a something negative to avoid acting on. In the present case, we have a phenomenon general across Slavic and Romance languages: a marker present on the non-causative side of a morphological causative~non-causative alternation disappears more generally under other kinds of morphological derivation, and possibly in periphrastic causatives as well (this is not uniform across Romance either, I believe). Repeating my question above: is it unattested elsewhere? If we do find other cases, does the phenomenon correlate with anything else? Do you think we know the answers to these questions already? Perhaps, but I’m not so sure (and alas, I myself did not pursue these questions that I implicitly posed in the 1995 book).
You are right that suspicions are often the driving force behind hypothesis formation, and one might of course form the hypothesis that the phenomenon is truly general if we formulate it abstractly enough. But this would be a highly speculative proposal, and speculation should not play such an important role in science, I think (this also applies to other, more applied fields such as economics and epidemiology). So yes, I do think that one should avoid focusing too much on suspicions (especially in teaching, where we should focus on methods and established findings). Of course, I do not already know (for sure) whether there are any further correlations of the sort that you suggest. But I have seen many descriptions of the world’s languages, and I do not recall having seen things of the sort that you are talking about. (In fact, polysemous “reflexive(-like) markers” are typically verbal affixes in the world’s languages, as can be seen, for example, in Bahrt’s 2021 book on voice syncretism: https://langsci-press.org/catalog/book/315. The Romance and Slavic type of “reflexive(-like) markers” are rather atypical.)
The similiarities in Slavic and Romance are intriguing, and would justify looking at more languages to see whether these patterns hold across many language families.
If, after looking at many language families, it turns out that these patterns are found in many families, (aka, if we successfully establish that these patterns really do represent a general phenomenon), then it would be reasonable to attempt to provide some sort of g-explanation for them.
The g-explanation could potentially be grammar internal, of the sort David mentions, but it could also potentially be grammar external.
Yes, I agree.
Of course it is important to establish that this is a general phenomenon by seeing if it is found in historically and geographically disconnected languages. I am still impressed that it occurs in disconnected contexts (anti-causative, event and agent nominalizations) within individual languages. And I also don’t think that the reasonableness of considering “g-explanations” bears a special burden of proof that other conjectures lack. But of course in the end what we want to discover is truth — and there is still work to be done in this area before we can declare that we have discovered it.
There is another kind of explanation that you (i.e. one) should consider, which is grammar-internal. I discussed the rasplavit’/rasplavit’-sja kind of alternation at length in chapter 4 of my 1995 book Zero Syntax, where my starting point was this kind of alternation in the domain of psychological predicates: e.g. Russian udivljat’ ‘suprise’ (causative) vs. udivljat’+sja ‘be surprised at’ (non-causative), French étonner ‘amaze’ (causative) vs. s’+ étonner (→ s’étonner) ‘be amazed at’ — but the discussion was extended to ‘melt’-type predicates as well. In previous chapters, I had argued that there is a null causative morpheme (non-null in languages like Japanese) in the causative side of this alternation, so ‘étonner’ is really amaze+CAUS. In chapter 4 of the book, I note that the Romance reflexive and Slavic -sja-type morpheme disappears quite generally even when overt morphology is added to a non-causative root. For example, the Russian verb stremit’-sja ‘strive, aspire to’ is inherently reflexive (with no causative counterpart or any counterpart lacking -sja, at least in modern Russian) — but when nominalized to yield a noun ‘striving, aspiration’ happily and oblgatorily loses the reflexive morpheme: streml-enie ‘attempt’. Likewise French s’+évanouir ‘faint’ with obligatory ‘se’, nominalization ‘evanouissement’ with no ‘se’. Furthermore, Italian and Spanish periphrastic causative constructions also appear to drop the reflexive clitic, for which I cited earlier work by Burzio and Zubizarreta:
a. Le nubi si dissipano. ‘the clouds SI dissipate’ (oblig. refl.)
b. Il vento dissiipa le nubi ‘the wind dissipates the clouds’ (no refl)
c. II vento ha fatto dissipare/*dissiparsi Ie nubi. ‘the wind AUX made dissipate the clouds’ (✓no reflexive, *with the reflexive)
If I was correct to suggest that there is a common factor behind the disappearance of the reflexive-like marker in these other two situations and in the cases you discuss, I would be interested to see the frequency-based explanation that covers them.
My suspicion (and I am not willing to attribute my suspicion to fashion, but rather to thought and experience) is that the explanation will turn out to be grammar-internal, though exactly what is going on remains to be fully explored. What I suggested is that the reflexive is required by the underived form for partly semantic reasons that I sketch in that chapter (it’s a kind of “ambient conditions” marker, like the “it” in English weather constructions), and silenced for grammatical reasons when the verb ends up in a tight syntactic unit with something else. For Slavic, by the way, this idea was anticipated by Robert Rothstein in his 1970s Harvard dissertation on Polish, which I mention in the chapter.
Of course the generalization over these cases that I suggested in 1995 could be spurious, in which case my concerns are irrelevant. But I wanted to bring them up here, in particular to suggest that the universe of possible explanations is a bit larger than your blog post might seem to suggest.
Many thanks, David, for bringing up these very interesting points! You are absolutely right that there are more possible explanations, and in several of my papers, I did consider generative-style explanations. (For causative and anticausative marking, I discussed Hale (2000) and Hale & Keyer (2002) in my 2005 LSA class at MIT: https://zenodo.org/record/6054181.) For the specific phenomenon that you bring up, the variable occurrence of reflexive(-like) markers in Romance and Russian, I think the answer is that they cannot be explained in general terms, because they do not represent a very general phenomenon. So I don’t think that there is a common factor behind them and the highly general patterns that I am concerned with. This means that we need some kind of p-explanation, not a g-explanation either in terms of frequency-induced predictability or in terms of UG (universal grammar). I’m aware that g-explanations carry more prestige than p-explanations, but since p-explanations are inherently more likely, I think that they should be the default. General explanations should be invoked only if we have good reasons (and not just suspicions) to believe that the patterns we study are indeed cross-linguistically general.
I think that the best p-explanation for many cases like the lack of -sja in Russian stremlénie derived from stremít’-sja etc. is purely historical. For instance, in Old Irish verb nominalizations functionally close to Russian or Latin infinitives typically lack all kinds of morphology present in the verb they belong too. In many cases such nominalizations – traditionally called ‘verbal nouns’ – don’t even share the root with the relevant verb. There is no synchronic rule how to derive a ‘verbal noun’ and, given the many cases of suppletion, such a rule probably can’t be formulated. ‘In order to drive a verbal noun replace everything in the verb by anything’ is not very appealing. At the same time, the historical background of the situation is, to some degree, clear. For some unknown reason, Old Irish abandoned the inherited infinitive (not unlike e.g. Bulgarian where this happened for areal reasons). Generations later Old Irish, again for unknown reasons, developed a new infinitive = ‘verbal noun’ by making use of nouns with suitable semantics (i.e. taking ‘feast’ for the verb to drink etc.). This explains why in Old Irish, in synchronical perspective, all kinds of morphs are dropped in the ‘verbal noun’. Similarly for Russian where -sja used to be (i.e. was in 11th-14th c.) a sentence clitic which modified the predicate of the clause and therefore could not be used as affix on nouns. Stremít’-sja → stremlénie just follows the normal pattern učít’ ‘to teach’ ~ učít’-sja ‘to learn’ → učénie ‘teaching; learning’ etc. To be shure, these historical facts do not explain everything (why would Old Irish abandon the inherited infinitive …?) but they do explain the morphological peculiarities of the nominalizations in question. Do we need an additional functional/cognitive/UG explanation? I think we only need one if we assume that each generation of speakers is constructing the grammar anew by recycling the morphs inherited from the previous generation.
Yes, I agree that it is enlightening to look at the historical development, and indeed, Russian stremlenie represents an ancient Indo-European pattern that normally would not be expected to combine with reflexive markers that go back to pronouns – simply because such deverbal nouns normally do not take accusative objects. So the Russian facts are not surprising at all, but I actually do not have a very deep understanding of the cross-linguistic generalizations: WHY is it the case that deverbal nouns tend to prohibit accusative objects, while they are quite permissive with prepositional objects (e.g. “rely on one’s friends” – “reliance on one’s friends”)? Note also that a closely related language, Polish, does allow reflexive “się” with deverbal nouns (e.g. zafascynowanie się, Rozwadowska 2020), so there does not seem to be any deep reason why Russian didn’t develop it.
I think my point is being missed here. One might expect that when the requirement that a particular verb combine with a reflexive-ish clitic conflicts with the impossibility of using that clitic with a nominalization, the result would be: no nominalization. But instead, the requirement of the clitic simply goes away. And if we treat the causative side of an unaccusative/causative alternations as involving a phonologically null but morphologically real causative affix, the disappearance of the same kind of clitic there can be seen as an instance of the same effect. This conclusion is reinforced by the periphrastic causative constructions that I cited in which the same clitic disappears, but here the causative morpheme is overt and free-standing.