Limits of variability in complexity of valency class systems

PI(s): Dr. Sergey S. Say, Prof. Dr. Ilja A. Seržant

In order to be able to advance causal explanations of a particular linguistic behavior in terms of universal cognitive and processing constraints as well as mental representations, cross-linguistically and diachronically stable variability limits must be established on empirically sound, areally and genealogically balanced, representative data. This typological approach allows disentangling various local “noises” and language-specific, accidental phenomena (idiosyncrasies of particular linguistic cultures) from universal limits of variability because, in the long run, linguistic phenomena survive through the evolution of languages only if they are advantageous for the human processor in production, comprehension, and learning (Hawkins 2004; MacDonald 2013; O’Grady 2015; Christiansen & Chater 2016; Seržant & Moroz 2022). Valency class systems provide a suitable research ground here because, on the one hand, they show quite a high degree of variation across languages (from very few classes to a high number thereof) but, on the other hand, there is a quite stable core of high and low transitivity-prominence verbs that do not show variation at all (Haspelmath 2015; Haspelmath & Hartmann 2015; Say 2014; Seržant et al. 2022). Arguably the main function of argument-coding devices (such as e.g. cases, adpositions or verbal indexes) is to establish links between referents in discourse and their argument roles. Given that verbs rarely have more than three arguments, a maximally simple system with just two or three valency classes, in which verbs with the same numerical valency uniformly assign argument-coding devices to their arguments, could in principle successfully fulfill this disambiguating function. Such a hypothetical system would maximize learnability and minimize the production effort to code the argument-related information. At the same time, such a system might be disadvantageous for comprehension since semantic cues available to the hearer would be highly susceptible to interpretation errors in a situation where parts of the signal are lost.

A hypothetical opposite extreme point is a maximally complex valency class system where each verb is associated with a unique argument-coding frame. Such a system would be maximally informative for the hearer but would be highly costly in memory retrieval during production as well as in learnability. It would also require transmitting a large amount of redundant information.

Clearly, both hypothetical scenarios outlined above are not observed in human languages: valency class systems are somewhere between these two hypothetical extremes. For example, most languages have a large semantically flexible class of transitive verbs (cf. kill, see, have, contain in English), but also at least several classes of non-transitive bivalent verbs (cf. look (for), wait (for), depend (on)). Thus, any actual valency class system probes a point of equilibrium between efficiency in learning and lexicon composition vs. maximal informativeness. Having said this, languages demonstrably vary in the degree of complexity of their valency classes, that is, they find their point of equilibrium at different parts of the hypothetical scale (Haspelmath 2015; Say 2014: 148ff.). Some languages mainly make use of basic monovalent, transitive and ditransitive constructions, while others possess a rich system of variegated valency classes (split S-marking; bivalent verbs with oblique arguments, etc.). This aspect of cross-linguistic variability remains underrepresented in typological research on valency and alignment, which is mainly concerned with major valency classes, cf. the standard notation in terms of S, A, P, sometimes augmented by T and R standing for the two non-agentive arguments in ditransitive constructions (Haspelmath 2011), which is normally applicable to only basic monovalent, transitive and ditransitive constructions.

The ultimate goal of this project is twofold: (i) the empirical work on establishing cross-linguistically stable limits of grammar in the domain of valency classes (contributing to Cluster C; Phase 2+3) but also (ii) identifying causal explanations for these limits in terms of universal constraints on language processing (contributing to Cluster B; Phase 3).


Dr. Sergey S. Say
Campus Am Neuen PalaisHaus 1, Raum 0.02
(+49) 331 977-124485
Prof. Dr. Ilja A. Seržant
Campus Am Neuen PalaisHaus 1, Raum 2.06
(+49) 331 977-4152


  • Peer-Reviewed: Papers, Journals, Books, Articles of the CRC
  • Talk or Presentation: Talks, Presentations, Posters of the CRC
  • SFB-Related: not produced in connection with the CRC, but are thematically appropriate
  • Other: Papers, Journals, Books, Articles of the CRC, but not peer-reviewed
Author(s)TitleYearPublished inLinksType
Say, S.Plural formation in Northeastern Neo-Aramaic (NENA) varieties spoken in Russia: towards superirregularity.2023Paper presented at the Vielfaltslinguistik 5, Universität Hamburg, Hamburg, Germany. 19-20 May.Talk or Presentation
Say, S.Understanding argument roles through the prism of BivalTyp (a typological database of bivalent verbs and their encoding frames).2023Invited talk at the Linguistische Werkstatt, Otto-Friedrich-Universität Bamberg, Bamberg, Germany. 31 May.Talk or Presentation
Say, S., & Ovsjannikova, M.Valency-encoding devices in a spoken Northeastern Neo-Aramaic corpus.2023Paper presented at the GRelSpoC 2023: Grammatical Relations in Spoken Corpora, Institut National des Langues et Civilisations Orientales (INALCO), Paris. 15-16 June. Talk or Presentation
Say, S.Cross-linguistic variability in complexity of valency class systems: implications for efficiency.2023Paper presented at the Efficiency in grammar: Patterns and explanations, University of Freiburg. 5 July.Talk or Presentation