“Irrealis is real“, but what are the predictions?

Semantic typology is based on semantic comparative concepts (or comparison meanings, as I have come to call them), and these concepts usually have labels that come from a particular language. Thus, the Concepticon, a systematic list of 3971 comparison meanings for word comparison, uses English labels for its comparative concepts.

In grammar, we often use terms derived from descriptions of European languages (e.g. genitive, perfective, 2nd person), but increasingly our general terms come from the grammars of non-European languages (e.g. evidential, antipassive). One such term is irrealis, which was apparently first used in the context of North American languages (such as Southern Paiute) and Oceanic languages. Some authors such as Bybee (1998) and Cristofaro (2012) have criticized the notion “irrealis” in the context of broader (diachronically oriented) theories of tense, aspect and mood, but in a recent paper, von Prince et al. (2022) defend it, especially on the basis of Oceanic languages and a branching-time model of temporal-modal meaning (“Irrealis is real”). They claim that “TAM systems of many languages are structured around a basic realis/irrealis distinction”, and that “minor deviations from these functions do not necessarily preclude the categorization of a marker as irrealis” (2022: 244).

In their branching-time model, there are three comparison meanings (or “modal domains”): (i) actual situations, (ii) potential situations, and (iii) counterfactual situations. These three can be defined formally in a model as summarized in Figure 1 (where the solid line stands for the actual present (ic) and earlier actual situations, dashed lines for potential situations, and dotted lines for counterfactual situations).

Given this model, one can define an irrealis marker as in (1).

(1) irrealis marker: An irrealis marker is a marker that signals some potential and some counterfactual meanings and no actual meanings.

This comes fairly close to what von Prince and colleagues are saying, though they do not distinguish very clearly between semantic domains and grammatical markers (as linguists sometimes do, e.g. when they distinguish between possessive vs. genitive, speech act vs. sentence type, property concept vs. adjective; Haspelmath 2025: §3). Much of the paper is taken up by discussions of systems that are more complex than the Oceanic systems with a binary irrealis/realis distinction (e.g. Mavea), and by discussions of phenomena that have sometimes been treated under the heading of “irrealis” but that von Prince and colleagues wish to exclude (negation, questions, subjunctives, habituals).

Overall, I really liked the paper because it is well-informed by the tradition of temporal logic and formal semantics (see also von Prince 2019), but at the same time focuses primarily on Oceanic languages (as the authors have experience with Vanuatu languages) and offers a rich discussion of the typological and descriptive literature in the traditions of Sapir and Bybee.

So one may accept their assertion that “irrealis is real” (contrary to Bybee 1998, who thought that “irrealis” was not a good concept for comparative purposes), but what exactly is their claim? They repeatedly say that the “realis/irrealis distinction is meaningful as a comparative concept”, but does this really go against what Bybee said? Are there “non-meaningful comparative concepts”? And is it the task of (cross-linguistic or language-particular) semantic analysis to make predictions? These questions were on my mind after reading the paper. My own take on comparative concepts is that they are methodological tools (somewhat like units of measurement) that allow us to formulate hypotheses (“predictions”) which we can then test, but they are not themselves predictive (Haspelmath 2018: 84). So I would not agree that

“Assuming the realis/irrealis distinction to be a valid comparative concept generates at least the following predictions: (i) a sizeable number of languages exhibit the distinction; (ii) in languages where one of these categories has an interpretation gap (one of the expected functions is unavailable), other more specified expressions will be found to fill this gap.” (von Prince et al. 2022: 222)

One can easily create comparative concepts that are useless because they do not correspond to any distinction that languages make (e.g. the distinction between ‘taller sister’ and ‘shorter sister’, as opposed to the very useful ‘elder/younger sister’ distinction), but that does not invalidate the concept. I would say that a comparative concept is “valid” if it can be applied to all languages using the same criteria – in fact, if it cannot, it is not a comparative concept. (Linguists sometimes use grammatical concepts for different languages using different criteria – such comparisons may be valid if the grammatical concepts are hypothesized to be innate natural-kind categories, but I leave this aside here.) I used the term “cross-linguistically valid” for comparative concepts myself in the past for the term converb, but I now think that that was a mistake. Converb is a good comparative concept, but not a “cross-linguistic category”.

It is true, of course, that one often has the feeling that by bringing a set of disparate phenomena under a general label that was not around previously (such as converb, and maybe irrealis), new comparisons are possible, and new insights are gained. But in the end, we need testable predictions and hypotheses. Von Prince et al. make another prediction, about the availability of “more specified expressions” when there is an interpretation gap. This is illustrated by Nakanai, which has a general irrealis marker that can be used for future/potential and counterfactual situations, but is blocked from expressing imminent actions by the existence of a special imminent irrealis marker. Their Figure 7 illustrates this (the dark shading shows the range of uses of the general irrealis marker, and the white shading shows the range for the imminent marker).

Nakanai

In this context, the branching-time model proposed by the authors looks like a semantic map (or coexpression diagram) – contiguity is typically said to be a general prediction of such semantic maps, though exceptions (via blocking) have occasionally been noted (e.g. Croft et al. 1987). Coexpression diagrams are hypotheses/claims and thus they make predictions – unlike comparative concepts. It seems to me that the authors are trying to do two things at the same time: define and contextualize a comparative concept (that has a long tradition terminologically) AND discuss predictions (or at least expectations) for cross-linguistic comparison. But what exactly is predicted (e.g. what kinds of languages they would predict don’t exist) did not become quite clear to me.

Still, the paper is very useful and instructive, and I would not want to defend Bybee’s (1998) paper to which they are reacting. Bybee is not very clear either about the relation between comparative concepts, language-particular categories, and predictions/hypotheses. She contrasts her functional-typological approach (which gives a prominent role to common diachronic pathways) with the structuralist Jakobsonian approach (using binary features), but the latter is primarily motivated by the goal of elegant language-particular description – it needs no comparative concepts. (Jakobson was not clear about this, though, and he later talked as if the features that he was working with were universal.) So I would not want to contrast “bad structuralism” with “good substance-based linguistics”, and instead prefer to say that we need structuralism for p-linguistics, while our cross-linguistic comparison (for g-linguistics) must be based on substantive concepts.

In semantics, the distinction between comparative concepts and language-particular categories has not often been discussed, but it is just as important as in morphosyntx and phonology. Perhaps there are a few dozen (innate) semantic primes (as in the NSM approach), and if this were so, we would be able to use the same concepts for semantic analysis and semantic comparison. But as long as this is not well-established, we need comparison meanings, and when we develop “models”, we should say clearly what their scope is: Are these models intended to account for the conventions of a particular language (or the corresponding internalized knowledge), or are they claimed to represent universal (innate) building blocks, or are they intended as summarizing typological claims (like the typologists’ coexpression diagrams)?

References

Cristofaro, Sonia. 2012. Descriptive notions vs. grammatical categories: Unrealized states of affairs and ‘irrealis.’ Language Sciences 34(2). 131–146. (doi:10.1016/j.langsci.2011.08.001)

Croft, William & Shyldkrot, Hava Bat-Zeev & Kemmer, Suzanne. 1987. Diachronic semantic processes in the middle voice. In Giacalone Ramat, Anna & Carruba, Onofrio & Bernini, Giuliano (eds.), Papers from the 7th International Conference on Historical Linguistics. Amsterdam: Benjamins.

Haspelmath, Martin. 2018. How comparative concepts and descriptive linguistic categories are different. In Van Olmen, Daniël & Mortelmans, Tanja & Brisard, Frank (eds.), Aspects of linguistic variation: Studies in honor of Johan van der Auwera, 83–113. Berlin: De Gruyter Mouton. (https://zenodo.org/record/3519206)

Haspelmath, Martin. 2025. Construction-functions and construction-strategies. In Däbritz, Chris Lasse & Budzisch, Josefine & Basile, Rodolfo (eds.), Existential and locative constructions. Berlin: Language Science Press (to appear). (https://ling.auf.net/lingbuzz/007897)

von Prince, Kilu. 2019. Counterfactuality and past. Linguistics and Philosophy 42(6). 577–615. (doi:10.1007/s10988-019-09259-6)

von Prince, Kilu & Krajinović, Ana & Krifka, Manfred. 2022. Irrealis is real. Language 98(2). 221–249. (doi:https://dx.doi.org/10.1353/lan.2022.0009)


OpenEdition suggests that you cite this post as follows:
Martin Haspelmath (October 13, 2024). “Irrealis is real“, but what are the predictions? Diversity Linguistics Comment. Retrieved March 16, 2025 from https://doi.org/10.58079/12gim


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.