Agency and Transitivity: A Review of Grammatical Relations and Voice

This post is based on a paper reviewing two articles, one on grammatical relations and agency and the other on voice & transitivity. (Carl Conrad’s discussion of Ancient Greek Voice discusses the latter article a good deal).

While the first half of this essay is not related to Greek, it ties the discussions together in such a way as to make the claims of the first article potentially relevant to future study of Greek voice, lexicography, transitivity, and Aktionsart.

I’ve included a few links to the SIL Linguistic Glossary or Wikipedia* for terms that some readers might not be able to follow. Even still much of the initial paragraphs are difficult to follow, though things become easier later on. If you can push through, I can relatively confidently promise that you will get something out of it by the end.

Key/New Terminology:




S: Subject Argument of an Intransitive Verb

A: Subject Argument of a Transitive Verb

P: Object Argument of a Transitive Verb (this is sometimes referred to as “O”)

Nominative / Accusative Systems vs. Ergative / Absolutive Systems

Split Systems

Dixon (1994) surveys split systems where a single agent argument is as an A, and the single object argument codes like an P. Within this semantic alignment, Dixon characterizes two types (71). A Split-S system places intransitive verbs into two sets, those marked as A’s and those marked as P’s. This syntactically based marking assigns each verb to a “syntactic frame” (71). The marking on these verbs is always the same, regardless of the semantics of the context. Fluid-S systems syntactically mark transitive verbs but semantically mark intransitives. Depending on the semantics of the situation, the intransitive subject can be marked as A or P.

Mandan, a Siouan language, is a Split-S language (71), distinguishing between activity predicates and stative predicates. Transitive and intransitive activities are marked with subjective suffixes and stative predicates are marked with objective ones. Pustat (2002) explores two Split Siouan languages more extensively in her article Split Intransitivity Revisited: Comparing Lakota & Osage. Like Mandan, both systems rely on lexical semantics, correlating with agency, but here these patterns occur in person prefixes rather than verbal suffixes.

Dixon views Fluid-S systems and Split-S systems as distinct, but Pustat holds the Fluid-S type is a sub-type of Split-S. She defines split intransitivity as a system allowing two patterning options for subjects of intransitives. Thus she does not differentiate between them (382).

Pustat aims to show that there are quantitative differences in the structure of Split-S systems, even within language families. Lakota and Osage split intransitives based on agency: activity predicates and state predicates pattern together, whether transitive or intransitive. Pustat gives a detailed account of the two languages and their coding for intransitive arguments. She calculates the relative percentage of lexemes coded as A’s versus those coded as P’s. Though the two languages of the same family both code intransitive predicates according to the same semantic criterion, Lakota’s results were A-dominated, while Osage data was P-dominated (397).

Of the possibilities explored, Pustat’s most promising approach to the coding discrepancy involves agency. If agency marks the same semantic concepts in both languages, lexemes should pattern the same. Based on DeLancey (1984:181) and Mithun (1991:516), Pustat suggests the idea of agency is multi-factored involving control, volition, animacy, instigation, and performance. Thus, agency not only includes the prototypical agent, but also less obvious cases. Walk, be broken, and stumble serve as examples (404). Walk is usually [+agentive], while be broken is regarded as [-agentive]. This leaves stumble, which is not so easy. It is agentive or not? To characterize agency, Pustat suggests that we can consider all of its semantic parameters. Thus, walk has positive values for all five, while be broken has negative ones. To Stumble falls on the agency scale but has mixed features (405). This semantic indeterminacy may result in discrepancies for how various languages code agents.

In light of Osage and Lakota, Pustat suggests some theoretical implications for ergativity’s origins. Much of her article is given to describing her study, but her report leads up to a concluding section where Pustat reveals the her true motivations. Split intransitivity is a possible diachronic source of ergativity. Some believe ergativity arose from reanalysis of passive and nominalization (406), but Pustat offers her case of Lakota and Osage as a different model.

In her proposal, ergativity emerges from the changes in the grammatical profile of individual items rather than entire constructions (406). Pustat sees Split intransitivity to “bridge the typological chasm” of ergativity and accusativity (406). Siouan languages are significant because, although most of them contain split intransitivity, at least one shows an accusative system with a few split-system instances (407).

Pustat desires to continue her research in the Siouan family to find out if the current theories explain the differences among languages of the same family. She seems unconvinced by the research that offers “mixed motivation” for Split-S systems (408). Comrie (1981) suggests accusative systems result from intransitive subjects converging with transitive agents due to assigning topicality to the agent. Du Bois (1987) explains how ergative systems arose using his principle of Preferred Argument Structure, which states that an argument new to a discourse is likely to be assigned as transitive patient or intransitive subject (408). However, this leaves split intransitivity unexplained. This dissatisfied Pustat because “the majority of Split-S languages documented so far lend themselves to a more homogenous, and thus, more elegant synchronic description on the basis of semantic parameters such as agency” (408).

Based on her “elegant” take on split intransitivity it is curious that she still considers it a transition point between ergativity and accusativity. Split-S systems appear more semantically driven than those systems with grammatical relations. Her consideration of the multi-factored agent may be the best motivation for split intransitivity I have encountered. Under this view, some S’s seem more like prototypical agents while other seem less so, and thus, they pattern that way. Her research is well founded since we do not know how likely it is that a split system would lead to an accusative system; however, the grander assumptions about the diachronic shift of languages from ergativity, to split, to accusativity, may be misplaced. I might ask why accusativity must be the final transformation. Could it not be the other way around? Or, perhaps split systems with their semantic alignment are the “end result” of languages dueling between ergativity and accusativity.

In line with Pustat’s claim of agency as a scale that codes many factors, Bakker (1994) uses a transitivity scale to examine middle voice in Ancient Greek. Voice, Aspect, & Aktionsart: Middle & Passive in Ancient Greek reviews two approaches to describing grammar. Discourse pragmatics and those studying language universals may take a primarily functional approach where form is “coded function” (23). On this view, Ancient Greek middle is a function, coded by certain morphology. In language-specific study, form is central; function is the meaning of a form (23). Traditional Greek Grammar as taught in most classrooms, considers the Greek middle a form, which expresses that the subject of the verb has some relationship with the action.

Bakker would like to see the approaches converge, arguing both are warranted but lacking in scope if they stand independently (24). Greek middle morphology relates to inherent events characterized in the verbs themselves. Middle morphology may add information to what is not already expressed in the verb itself, or it may only reflect what is inherently there (24).

Middle voice is usually described as marked with respect to active voice because it expresses some kind of affectedness of the subject on the verb. Bakker claims that because middle morphology is only part of the verbal word, the idea of affectedness serves as only an “abstract cover term” (24). According to Bakker’s approach, affectedness must be studied along with Aktionsart, the “inherent nature of the event-type denoted by the verb” (25).

Hopper and Thompson (1980) introduce transitivity as a synthesis of variables (25). Bakker uses these notions with one addition of his own to present transitivity as a scale. Volitionality refers to the free will of the agent, agency to an action directed to a goal, and causation to an external agent producing an effect on something or someone. The scale increases in transitivity with volitionality as the lowest and causation as the highest (25).

Bakker examines eleven event-types, ranging from low transitivity (‘to die’) to high transitivity (‘to kill’). Because the notion of affectedness on the subject is sometimes inherent in a verb and sometimes not, middle voice is mixed, to say the least. Affectedness is a feature of the morphology as well as a feature inherent in the event-types. When affectedness is inherent in the verb, middle morphology reflects (or codes) what is already there. When an event lacks affectedness and middle markings are added, then they express affectedness in the event (26).

In discussing aspect in Ancient Greek, Bakker refers to the fundamental difference between aorist and imperfect forms, that is, perfective and imperfective aspect. Aspect is either marked on the verbal stem or as a suffix. Interestingly, Aktionsart limits aspectual morphology in Greek. The event type determines whether the imperfective or perfective aspect is considered default for a verb (26-27).

Morphologically, -sa is taught to Greek students as the “tense” formative for aorists. Bakker sees -sa as a transitivizing suffix; the forms with it are more transitive than those without. Verbs with -the/-e are low in transitivity (27). Greek students learn this morpheme as signaling aorist passives, but Bakker claims this is done to a fault since it only marks passiveness in one event-type. Most often it is used when affectedness does not apply, so it is non-middle (27).

Bakker presents his eleven event-types starting with one-participant events, the lowest in transitivity. In the future, these events (such as ‘die’ or ‘fall’) consistently have middle morphology because future tense in Greek contains a volitionality marker as its tense formative. Since it has such a close connection with volition, future can be considered a mental intention. When volitionality is the only transitivity feature (agency and causation are absent), the verb involves affectedness (29).

Objective states are verbs such as ‘be king.’ Physical processes are durative events like ‘grow.’ Then motion & emotion follow, and verbs of volitionality. Next is inherent reciprocals, such as ‘fight’ or ‘kiss.’ Transitive two-participant events are next. He follows with causatives, which focus on what happens to the causee when in middle voice, such as ‘raising one’s child.’ Patient-directed actions are events where the agent is not affected and the patients are true patients in that they exert no volition of their own. ‘Kill’ has its own event-type and the highest transitivity. There is no inherent affectedness and no middle forms.

With these event-types, Bakker establishes a study of transitivity cannot ignore affectedness. By presenting his event-types in order of increasing transitivity he shows that affectedness rises with transitivity. At the midpoint the trend reverses and affectedness decreases while transitivity increases. The highest point of transitivity and affectedness is in events where the agent is the beneficiary of his own acts. The reversal occurs with events of causation, because the affected participant begins to change from having volition of its own to being a non-volitional patient. Middle voice in Ancient Greek must be understood in terms of the affectedness the verb has on the agent, but transitivity cannot be ignored.

Bakker demonstrates the variability of middle voice in relation to transitivity. The traditional notion of middle voice in Greek is described as an agent’s special interest in the action of the verb. Just as agents can be more agentive or less agentive (in terms of control, volition, animacy, instigation, and performance), so also transitivity includes volition, agency, and causation. Perhaps Pustat and Bakker are working with the same semantic claims – that our notion of subjects and verbs can be quite narrow or even perhaps too broad. In our haste, we assign arguments the role of agent as a purely grammatical notion, regardless of the semantics involved. Then when analyzing split systems, we consider them a bit deviant from the two major grammatical systems. When studying middle voice, we may be unable to characterize it because semantically it is not easily classified, such as in Ancient Greek. It may be that these variable notions can give us food for thought in our assumptions about agency and transitivity.


Comrie, Bernard. 1981. Language Universals and Linguistic Typology. Oxford: Blackwell.

Bakker, Egbert J. 1994. Voice, Aspect and Aktionsart. Voice: Form and Function. Typological Studies in Language (TSLang): 27: 23-47.

DeLancey, Scott. 1984. Notes on agentivity and causation. Studies in Language 8: 181-213.

Dixon, Robert M. W. 1994. Ergativity. Cambridge University Press. 70-109.

Du Bois, John W. 1987. The discourse basis of ergativity. Language 63: 547-619.

Hopper, Paul J. and Sandra A. Thompson. 1980. Transitivity in grammar and discourse. Language 56: 251-99.

Mithun, Marianne. 1991. Active/agentive case marking and its motivations. Language 67: 510-46.

Pustat, Regina. 2002. Split Intransitivity Revisited: Comparing Lakota and Osage. International Journal of American Linguistics 68, no. 4 (October): 381-427.

*The nice thing about Wikipedia and Linguistics is that unlike Biblical Studies, there’s very little vested interest for people who don’t know what they’re talking about. Thus many articles tend to be quite accurate simply because normal people wouldn’t think to write an article on split nominative/ergative systems.