Causal clauses as source of sentential complementation: cross-linguistic evidence and methodological issues

1 Background

Complement clause markers can show synchronic identity with causal adverbializers.^[1] One example is Modern Hebrew ki, a subordinator whose versatility can even result in “ambiguity” (Zuckermann 2006: 81):^[2]

While causal-complement isomorphism often has an explanation in common developments from sources such as interrogatives or relativizers (see, e.g., Givón 2015: 250; Kehayov and Boye 2016: 870), Hebrew complementizer ki has been argued to derive “from kī ‘because’” (Zuckermann 2006: 81). A similar CAUSE > COMPLEMENT development has been suggested for other languages. However, important questions about this type of language change remain unanswered:

1. How does the CAUSE > COMPLEMENT process operate? What evidence can we find?

2. How recurrent is this process assumed to be in cross-linguistic perspective?

Dixon and Aikhenvald (2006) cross-linguistic analysis of complementation includes 12 case studies, two of which explicitly assume a CAUSE > COMPLEMENT diachronic process (Deutscher 2006; Zuckermann 2006). This may lead us to expect it to be a recurrent phenomenon. By contrast, CAUSE > COMPLEMENT does not feature in cross-linguistic works such as Kuteva et al. (2019) or Boye and Kehayov (2016) and is only sporadically assumed in Schmidtke-Bode (2014). This tacitly points to a different expectation.^[3]

Two scenarios of CAUSE > COMPLEMENT reanalysis or grammaticalization are often cited in the literature:

1. Givón (1991) suggests some possible routes explaining the emergence of complementation in Hebrew. One of them, the “because-factive” analogical pathway, hypothesizes a morphological transfer from original relative clauses to complements through “because-adverb-clauses”. This is explained in the context of factive/presuppositional predicates (e.g., be happy, regret). In this scenario, the state or event coded in the complement (e.g., He’s happy that she’s here ) is also the cause of the state or event coded by the matrix clause (e.g., He’s happy because she’s here ) (Givón 1991: 296). A similar explanation has been proposed for the rise of the Modern Greek complementizer που (pu) (Nicholas 1998) and Latin quod (Deutscher 2000: 62). From there, complementation has been argued to spread to other types of complement-taking predicates.

2. Deutscher (2000) on the other hand, argues that the Akkadian historical record documents the emergence of complementation through the direct reinterpretation of causal clauses in the “bridging” context of speech-related verbs:

He [spoke to/wrote to/said to/informed] the governor kīma (=because/that) the barley was not collected. (Deutscher 2000: 42)

• This scenario also relies on factive predicates, but it assumes that the complement-taking predicates involved in the reanalysis are speech-related verbs, rather than emotional predicates. Furthermore, it does not presume the process to follow from a morphological transfer from original relative clauses.

The cross-linguistic study of the diachronic relationship between causal and complement clauses is severely limited by the pervasive lack of diachronic data (cf. Schmidtke-Bode 2014: 243–244), often limiting historical syntax studies to interlanguage historical comparative analysis (Ferraresi and Goldbach 2008: 24). The available diachronic record is usually insufficient to provide information about aspects like prosodic integration, differences between oral and written registers (cf. Diessel and Hetterle 2011) or pragmatic nuances that could be relevant for the categorization of clauses as embedded arguments or adjuncts (Lehmann 1988).

Nevertheless, historical records can provide informative data about the characteristics and evolution of causal and complement structures (Cristofaro 2003, 2014), particularly if we can control for aspects like the relative frequency and distribution of attested complement-taking predicates (Schmidtke-Bode 2014: Ch. 7) or the characteristics of the textual evidence and their sociolinguistic environment.

In Section 2, I will present a collection of claims and suggestions for CAUSE > COMPLEMENT processes found in the literature, focusing on ancient Indo-European languages (2.1), Semitic languages (2.2), other examples (2.3), and emotional predicates (2.4). Some relevant observations will be then discussed in Section 3. The conclusions in Section 4 summarize the diachronic and cross-linguistic evidence for CAUSE > COMPLEMENT claims and its challenges.

2 Cross-linguistic observations

References to CAUSE > COMPLEMENT processes were sought in a varied array of publications, many of which appear as bibliographical references in wide cross-linguistic works such as Dixon and Aikhenvald (2006), Kuteva et al. (2019), Noonan (2007), and especially Schmidtke-Bode (2014), a study based on a large dataset of complementation patterns. These claims or suggestions, which are summarized in Table 1, cover many geographical areas and linguistic genealogies. The clustering of Semitic and Indo-European cases correlates with the wealth of studies and diachronic data on these languages.

3 Discussion

Despite the different potential interpretations of the diachronic data, purported CAUSE > COMPLEMENT processes for Latin quod clauses and Hittite kuit clauses have served as reinforcing parallels for Akkadian (Deutscher 2000: 62). Akkadian, in turn, has suggested similar developments in related Ugaritic, Ancient Aramaic, and Hebrew k- markers (Deutscher 2000: 63–64), and provided a model for modern languages like Epena Pedee. At the same time, the analogical because-to-factive evolution of Hebrew ʾasher/she clauses hypothesized in Givón (1991) has been enhanced by an assumed equivalent development for pre-biblical kī clauses (Givón 1991: 298). Although Givón (2015) changes his view on this and suggests the opposite COMPLEMENT > CAUSE trajectory, the evolution of Modern Greek που (pu) in Nicholas (1998) is modelled after the earlier hypotheses in Givón (1991). Since the evidence for CAUSE > COMPLEMENT processes in all these cases is insufficient, we find that many of the claims or suggestions in Table 1 might have been influenced by a circular flow of inadequate mutual corroborations.

Obviously, the observations questioning the validity of specific claims or suggestions in Table 1 do not imply the general impossibility of CAUSE > COMPLEMENT, a process that might very well explain the polyfunctionality of the Moksha Mordvin marker mes or the Jamiltepec Mixtec vátyí/tyí (perhaps to be extended, on comparative grounds, to Coatzospan Mixtec causal/complementizer tsí; cf. Small 1990: 294).

Nevertheless, two considerations are in order. First, the diachronic extension of causal constructions into the domain of complementation is not necessarily only the output of independent reanalyses based on semantic cause-factive analogies. Latin quia clauses illustrate how a causal-to-complement evolution could be potentially driven by factors like convergence with other polyfunctional markers in the speakers’ repertoires (which, in turn, might themselves not be the product of CAUSE > COMPLEMENT processes). The Late Latin use of quoniam as a complementizer is also due to an increasing convergence with quia (Sznajder 2019: 174), and according to Givón (2015), BH ʾasher/she mirrored the polyfunctionality of ki.^[7]

The second consideration refers to the recurrent allusion to CAUSE > COMPLEMENT analogical bridges provided by the semantic ambiguity of emotional predicate environments. This explanation is absent for OB Akkadian, where verbs denoting abstract mental state are scarcely attested (Deutscher 2000: 96) or simply do not occur with finite complementation (like ‘fear’). Instead, a similar bridging context was assumed for OB speech-related verbs (Deutscher 2000). The arguments against this in Section 2.2 and Hernáiz (2024) highlight the potential problem of relying on apparent cause/complement ambiguity, especially for historical doculects with insufficient documentation and no possible access to informants. As De Smet (2009) points out, surface ambiguity can transpire in retrospect as a consequence of a specific language development, rather than being its cause.

The frequent allusion to emotional predicates observed in the literature can also be explained without invoking CAUSE > COMPLEMENT or COMPLEMENT > CAUSE reanalyses: if complement and causal functions had independent developmental paths and the causal one resulted from sources such as temporal or relative markers (cf. Kortmann 1997), it could be the case that grammars tend to showcase causal uses in emotional contexts because they provide more canonical examples for this causal reading. For example, the causal meaning of Yankunytjatjara’s polyfunctional “circumstantial ending” in Goddard (1985) might appear mentioned almost exclusively in contexts of emotional predicates because these entail less ambiguous causal/temporal/simultaneous interpretations. Similar explanations whereby emotional predicates provide semantic meeting points rather than bridging contexts between cause and complement meanings could be considered for other references in Section 2: Greek που (pu) is traditionally considered causal only in contexts of emotional predicates (Nicholas 1998: 69) but is considered temporal in clauses with the matrix predicate ‘remember’ (Nicholas 1998: 460). In Hausa, causal dà could also derive from temporal meanings (Abdoulaye 2009: 20), and Ugaritic “causal clauses, particularly, are often difficult to distinguish from temporal/circumstantial clauses” (Bordreuil and Pardee 2009: 69).

A subsequent “language-internal” extension of complementation from potentially original emotional predicate environments to other non-factive complement-taking predicates also remains uncertain for cases mentioned in Section 2. The description of complementation for Moksha Mordvin, Modern Greek (που), and Amharic remains mostly limited to emotional predicates. For Latin, even if quod/quia complements had appeared first in emotional predicate environments, their selection by other complement-taking predicates seems to rely substantially on the effects of language contact and convergence with other polyfunctional markers (quia) or be an independent development from (cor)relative structures (quod; cf. Cuzzolin 2013). Regarding independent developments, Lezgian luhun offers an insight that illustrates further the possibility of parallel or intertwined sources for complementation: if one conjecturally wanted to assume an original CAUSE > COMPLEMENT process in emotional predicate environments (based on the shared semantic properties mentioned in Section 2.4), a subsequent extension of complementation from emotional predicates to utterance complement-taking predicates would be rejected, as the latter surely grammaticalized directly from quotative uses (Haspelmath 1993: 367).^[8]

The causal-complement ambiguity in contexts of emotional predicates can be problematic for cross-linguistic comparisons relying on grammar descriptions. Schmidtke-Bode’s study (2014: 275) has shown the peculiarity of emotional predicates in cross-linguistic clustering of complement-taking predicate classes, and potential difficulties in distinguishing causal and complement functions in emotional contexts are often expressed in language descriptions (e.g., Harrison 2019: 287; Kholodilova 2018: 664; Pinkster 2021: 61). Different interpretations can also be found for the same phenomenon: Geez ʾasma-clauses in emotional predicate constructions are considered causal adverbial in Dillmann and Bezold (2005: 542) but complements in Tropper and Hasselbach-Andee (2021: 344). Similar ambiguity applies to the description of Greek που clauses in emotional contexts (Nicholas 1998: 69 and 460). Even for English, there can be different views for examples like (4): while the clause introduced by that is usually recognized as a complement (Dixon 2006: 11; Noonan 2007: 127), Payne (1997: 317) considers it a reason clause (not constituting a logical argument).

The potential cross-linguistic inconsistency in describing emotional predicate constructions is not completely resolved in Dixon (2006: 20), where one of the four main criteria defining complements assumes that they “must be in O (or E) function for one or more of a set of prototypical complement-taking verbs”. Crucially, Dixon (2006) does not consider emotional predicates to be prototypical complement-taking predicates, which questions whether some cases in Table 1 in which complements are only (or mainly) realized in emotional contexts would be analysed as complements at all. Conversely, similar clauses in complement-taking environments can be considered complements in some language descriptions but disregarded as such in others.^[9]

Interestingly, Schmidtke-Bode (2014: 22) shows that the syntactic notion of “argument”, used for categorizing clauses as complements, is less straightforward for emotional predicates. Even though specific mechanisms to code semantic or syntactic nuances can indeed exist in languages with isomorphic cause/complement markers (for example, different complementizers in Ancient Greek [Cristofaro 2008: 588–589]; valence-increasing morphemes in Mokilese [Harrison 2019: 288]), the semantic nuances that differentiate be happy that and be happy because constructions in languages like English might not necessarily be expressed or overtly coded in languages with identical markers for cause and complement, potentially obscuring further the distinction between complements and causal adverbial clauses in emotional contexts.

When it comes to diachronic studies, our data lacks reliable prosodic information, and we cannot ask informants about semantic and syntactic nuances in emotional contexts. Furthermore, the analysis of morphosyntactic integration might not always be useful, as one of the ways in which complementation might originate is through clauses with low integration with the matrix clause (Lühr 2008: 22). Whilst it could be argued that a later occurrence of complement patterns with other complement-taking predicates could prove an earlier complement status with emotional predicates, the diachronic data is often inadequate to discard alternative multicausal or interdependent developments. The causal/complement status of clauses in emotional predicate constructions might then remain ambiguous for the modern researcher, increasing the possibility of translation or anachronistic biases.

4 Conclusions

The goal of this paper was to collect and assess diachronic and cross-linguistic evidence for CAUSE > COMPLEMENT claims in the literature. Fourteen cause/complement constructions were found for which this process had been suggested with varying degrees of confidence.

Causal clauses may be the source of complements in Jamiltepec Mixtec and Moksha Mordvin, but more information is required to substantiate the process and reject alternative explanations. The diachronic evidence analysed for other languages, however, is mostly uncompelling.

Except for Latin quia/quoniam and derivates, the limited size and representativeness of the earliest language records is insufficient to demonstrate the anteriority and the explicit source role of CAUSE in polyfunctional causal/complement constructions. Alternative explanations based on the cross-linguistically well-described RELATIVE > COMPLEMENT path (see, e.g., Hendery 2012; Schmidtke-Bode 2014: 248–251) have been convincingly invoked for Hittite kuit complements and Latin quod complements (Lühr 2008, among others) and for Hebrew ʾasher/she clauses (Cristofaro 1998; Givón 2015; Rubin 2005).

A significant observation is that the assessment of the Akkadian textual record, which had been confidently postulated as one of the best documented examples for the process concerned, suggests in fact a non-causal MANNER > COMPLEMENT development for kīma clauses (Hernáiz 2024), which opens up the possibility that this path may also be shared by cognate Semitic k- complementizers.

Most of the CAUSE > COMPLEMENT hypotheses analysed imply some sort of independent reanalysis based on semantic cause/factive analogies in ambiguous contexts. However, the development of complements formally identical to pre-existing causal constructions might also be driven, to some extent, by convergence with other polyfunctional cause/complement markers in the speakers’ repertoires (including those introduced via contact with other languages), as appears to be central in the evolution of Latin quia and quoniam clauses.

The study also finds that one particular CAUSE > COMPLEMENT scenario recurs in the literature: the reanalysis of causal adverbial clauses into complement arguments of emotional predicates. An opposite COMPLEMENT > CAUSE process has been alternatively suggested as responsible for this type of semantic overlap (Givón 2015 for BH; Cristofaro 1998 for Ancient Greek).^[10] However, the more prominently causal nuance in emotional contexts of otherwise more ambiguous or loosely causal constructions might sometimes influence the perceived overlap with complements. Nevertheless, the analysis of these scenarios is often challenging, particularly for extinct languages, and a potential inconsistent categorization of these structures as either adverbial or complements in individual language descriptions should be addressed in cross-linguistic generalizations.